{ "reviewed_on": "2026-10-25", "policy": "Exclude 2023/2024 and older AI evidence, including old experiments republished later. Hold undated original cases and mixed-period results that cannot be separated. Recent publications with unstated measurement dates remain illustrative references, explicitly labeled; their dates do not establish current-model performance.", "range_policy": "Seven ranges re-estimated with explicit before/after scenario budgets. All ten numeric ranges remain conditional analyst scenarios; no full-workflow empirical interval is established.", "sources": { "accounting-deloitte_zora": { "id": "accounting-deloitte_zora", "url": "https://www.deloitte.com/us/en/what-we-do/case-studies/hands-off-the-task-eyes-on-the-outcome.html", "title": "Deloitte: Hands off the task, eyes on the outcome", "publication_date": "2026-10-19", "evidence_period": "Not stated. Describes an implemented Zora AI deployment after earlier ERP/process/RPA work; measurement dates and model version are absent.", "status": "retain", "reason": "First-party page metadata establishes a recent publication. Useful AP deployment case, with wider automation than the modeled human-reviewed draft-bill project.", "verified_date_evidence": { "url": "https://www.deloitte.com/us/en/what-we-do/case-studies/hands-off-the-task-eyes-on-the-outcome.html", "method": "Fetched first-party HTML metadata", "detail": "createDate 07 Jul 2026; dateModified 08 Oct 2026; datePublished and publishDate 19 Oct 2026. Page has no visible date." }, "project_ids": [ "AI-01" ], "kind": "Consultancy-authored anonymous customer deployment", "result": "Reported invoice-processing time reduction of 50–75%, 50% fewer invoice-processing FTEs and 92% touchless processing.", "denominator_and_limits": "Invoice counts, baseline FTE count, sampling period, human-time measurement and review boundary are absent. Processing-time wording is imprecise. Autonomous rejection/approval and prior process redesign exceed a draft assistant with every invoice checked.", "quantitative_use": "Illustrative broader AP outcome; cannot establish the human-reviewed project's lower/upper bounds.", "plain_summary": "Deloitte reports 50–75% less invoice-processing time and 50% fewer processing FTEs at an unnamed company. Its system handles most invoices without people; that is broader than preparing drafts for AP review.", "doc_bullet": "[Deloitte AP case (2026)](https://www.deloitte.com/us/en/what-we-do/case-studies/hands-off-the-task-eyes-on-the-outcome.html): 50–75% less invoice-processing time and 50% fewer processing FTEs reported. Most invoices are handled automatically, so the scope is broader than this assistant; measurement dates are unstated." }, "accounting-auditoria_university": { "id": "accounting-auditoria_university", "url": "https://www.auditoria.ai/case-studies/large-public-research-university/", "title": "Auditoria: From 100K Emails to Zero Chaos — A University Innovates AP", "publication_date": null, "evidence_period": "Not dated. Text refers to improvements over the last two years without anchoring them to a publication date.", "status": "hold", "reason": "No verifiable publication or deployment dates. A current 2026 site banner does not date the case. Direct HTML metadata request was blocked.", "verified_date_evidence": { "url": "https://www.auditoria.ai/case-studies/large-public-research-university/", "method": "Live primary-page review and metadata attempt", "detail": "Primary page accessible to web reader but shows no date; direct fetch returned HTTP 403. Date remains unverified." }, "project_ids": [ "AI-01" ], "kind": "Vendor-authored anonymous university case", "result": "Time per invoice reportedly fell from close to 10 minutes to under 5; turnaround from 14 to 7 business days; 75% of intake/queue distribution automated.", "denominator_and_limits": "Eight-person invoice team, more than 5,800 invoices/month; no matched sample or time-study method. Turnaround days must remain separate from labor minutes.", "quantitative_use": "Remove from active recent evidence until date is verified.", "plain_summary": "The reported per-invoice improvement is relevant, but the case cannot yet pass the date cutoff." }, "accounting-auditoria_ringcentral": { "id": "accounting-auditoria_ringcentral", "url": "https://www.auditoria.ai/case-studies/ringcentral/", "title": "Auditoria: RingCentral AP case", "publication_date": null, "evidence_period": "Unverified", "status": "hold", "reason": "No verifiable first-party publication date obtained; primary page failed in web reader and direct fetch returned HTTP 403. Do not infer freshness from a 2026 directory listing.", "verified_date_evidence": { "url": "https://www.auditoria.ai/case-studies/ringcentral/", "method": "Live fetch attempts", "detail": "No accessible dated primary record found." }, "project_ids": [ "AI-01" ], "kind": "Vendor-authored named customer case", "result": "Existing model records 75%+ lower process workload, 96% extraction accuracy and most invoices processed in under two minutes; not newly reverified in this pass.", "denominator_and_limits": "No verified human-hours denominator. Process workload, accuracy and processing runtime are different metrics.", "quantitative_use": "Remove from active recent evidence pending date/content verification.", "plain_summary": "RingCentral's date remains unverified, so this case should leave the active evidence set." }, "accounting-pamarco": { "id": "accounting-pamarco", "url": "https://www.microsoft.com/en/customers/story/27289-pamarco-technologies-dynamics-365-finance", "title": "Microsoft customer story: Pamarco Technologies", "publication_date": "2026-09-18", "evidence_period": "Deployment and measurement dates not stated; 2026 case describes current ERP-connected Copilot Studio/MCP use.", "status": "retain", "reason": "Visible primary publication date is recent. Keep reported coding success separate from projected labor saving.", "verified_date_evidence": { "url": "https://www.microsoft.com/en/customers/story/27289-pamarco-technologies-dynamics-365-finance", "method": "Visible date on live primary page", "detail": "9/18/2026" }, "project_ids": [ "AI-01" ], "kind": "Vendor-authored named customer deployment", "result": "About 95% freight-invoice coding success reported. The fully adopted process is estimated to reduce a daily task from one hour to five minutes.", "denominator_and_limits": "The one-hour-to-five-minute claim is an estimate for full adoption, not a demonstrated before/after labor result. Automatic posting of non-PO invoices exceeds current draft-only scope. Broader 80% coding headline and 95% freight success concern different described applications.", "quantitative_use": "Supports feasibility; projected saving cannot calibrate measured workflow bounds.", "plain_summary": "Pamarco reports about 95% success coding freight invoices. It estimates much larger time savings after full adoption, including automatic posting.", "doc_bullet": "[Pamarco (2026)](https://www.microsoft.com/en/customers/story/27289-pamarco-technologies-dynamics-365-finance): about 95% freight-invoice coding success. The estimate of one hour reduced to five minutes is for full adoption, including automatic posting; it is not a measured saving for this draft-only scope." }, "accounting-asurion": { "id": "accounting-asurion", "url": "https://www.microsoft.com/en/customers/story/27456-asurion-insurance-services-microsoft-copilot-studio", "title": "Microsoft customer story: Asurion invoice processing", "publication_date": "2026-10-08", "evidence_period": "Exact deployment/measurement dates not stated; current 2026 customer report.", "status": "retain", "reason": "Recent primary case with a clear process-coverage result; no labor-time percentage.", "verified_date_evidence": { "url": "https://www.microsoft.com/en/customers/story/27456-asurion-insurance-services-microsoft-copilot-studio", "method": "Visible date on live primary page", "detail": "10/8/2026" }, "project_ids": [ "AI-01" ], "kind": "Vendor-authored named customer deployment", "result": "Touchless processing grew from below 20% to above 70% for North/South American vendors after embedded agents corrected defined mapping and header exceptions.", "denominator_and_limits": "Share of invoices handled without intervention, not share of human time saved. No timed baseline, correction burden or representative study sample disclosed. An 85% target is future ambition and excluded.", "quantitative_use": "Supports exception-handling feasibility; no conversion to human-time saving.", "plain_summary": "Asurion reduced selected invoice exceptions enough to raise touchless processing above 70%. That does not tell us the percentage of staff time saved.", "doc_bullet": "[Asurion (2026)](https://www.microsoft.com/en/customers/story/27456-asurion-insurance-services-microsoft-copilot-studio): touchless invoice processing rose from below 20% to above 70% in North and South America. This measures invoices requiring no intervention, rather than staff hours saved." }, "accounting-choi_xie_final": { "id": "accounting-choi_xie_final", "url": "https://onlinelibrary.wiley.com/doi/10.1111/1475-679x.70052", "title": "Choi and Xie: Human + AI in Accounting — 2025 framed experiment only", "publication_date": "2026-04-16", "evidence_period": "June–July 2025 framed experiment. Exclude the separate January 2023–March 2025 platform records and November 2024/March 2025 surveys from recent quantitative support.", "status": "retain", "reason": "The published paper explicitly dates the experimental component to 2025; separate components must not inherit that date. Model version is not disclosed in the accessible text.", "verified_date_evidence": { "url": "https://onlinelibrary.wiley.com/doi/10.1111/1475-679x.70052", "method": "Live full-text article", "detail": "First published 16 April 2026; Figure 1 caption dates the framed experiment to June and July 2025; Table 10 reports its results." }, "project_ids": [ "AI-01", "AI-02", "AI-03" ], "kind": "Peer-reviewed randomized framed experiment", "result": "99 accountants classified the same 43 transactions. AI improved accuracy by 17.5 percentage points; average completion time did not significantly change. Seconds per correct classification improved.", "denominator_and_limits": "52 treated and 47 control; transaction categorization into 153 account categories, not invoice intake or reconciliation. Completion-time coefficient +92 seconds, statistically insignificant. Accuracy-adjusted efficiency is not ordinary task-time reduction. No speed-linked incentive.", "quantitative_use": "Important direct coding counterexample; cannot justify selecting an arbitrary zero-low/60%-high stage range or transfer to collections/reconciliation.", "plain_summary": "In a 2025 coding experiment, AI made accountants more accurate without a statistically significant average time saving. It tested coding, not the full invoice workflow.", "doc_bullet": "[Choi and Xie (2026; experiment June–July 2025)](https://onlinelibrary.wiley.com/doi/10.1111/1475-679x.70052): 99 accountants classified 43 transactions. AI improved accuracy by 17.5 percentage points, but did not significantly reduce average completion time." }, "accounting-choi_xie_old_field_component": { "id": "accounting-choi_xie_old_field_component", "url": "https://onlinelibrary.wiley.com/doi/10.1111/1475-679x.70052", "title": "Choi and Xie: mixed 2023–2025 field records and survey components", "publication_date": "2026-04-16", "evidence_period": "Platform records January 2023–March 2025; survey waves November 2024 and March 2025.", "status": "exclude", "reason": "These reported associations incorporate excluded 2023/2024 use. A 2026 final publication does not refresh the technology or measurement period.", "verified_date_evidence": { "url": "https://onlinelibrary.wiley.com/doi/10.1111/1475-679x.70052", "method": "Live full-text timeline", "detail": "Figure 1 caption and survey-method section explicitly identify the component periods." }, "project_ids": [ "AI-01", "AI-02", "AI-03" ], "kind": "Mixed-period observational/survey components", "plain_summary": "Exclude older platform and survey associations; retain only the separately dated 2025 coding experiment." }, "accounting-qje_support": { "id": "accounting-qje_support", "url": "https://academic.oup.com/qje/article/140/2/889/7990658", "title": "Brynjolfsson, Li and Raymond: Generative AI at Work", "publication_date": "2025-02-04", "evidence_period": "GPT-3 support-assistance rollout primarily fall 2020 and winter 2021.", "status": "exclude", "reason": "2025 journal publication reports much older technology/use. It is unsuitable for the user's recent-technology reference set.", "verified_date_evidence": { "url": "https://academic.oup.com/qje/article/140/2/889/7990658", "method": "Live article sections III and IV.A", "detail": "Text states GPT-3; rollout primarily fall 2020/winter 2021, with adoption rising October 2020–January 2021." }, "project_ids": [ "AI-02" ], "kind": "Peer-reviewed customer-support field study", "result": "Prior model used 8.5% lower average handling time and about 15% more resolutions/hour.", "denominator_and_limits": "Customer-service chats rather than collection-email drafting. Historical result, not a current technology estimate.", "quantitative_use": "Remove old handling-time anchor and reasoning based on it.", "plain_summary": "Exclude: the 2025 paper studies a 2020–2021 GPT-3 deployment." }, "accounting-noy_zhang": { "id": "accounting-noy_zhang", "url": "https://doi.org/10.1126/science.adh2586", "title": "Noy and Zhang: Experimental evidence on the productivity effects of generative artificial intelligence", "publication_date": "2023-07-14", "evidence_period": "2023-era ChatGPT writing tasks", "status": "exclude", "reason": "Explicitly excluded publication year; also an indirect writing-task analogue.", "verified_date_evidence": { "url": "https://api.crossref.org/works/10.1126/science.adh2586", "method": "Publisher-deposited DOI metadata fetched live", "detail": "published and published-print date-parts: 2023-07-14. Science HTML returned HTTP 403." }, "project_ids": [ "AI-02", "AI-08" ], "kind": "Peer-reviewed writing experiment", "result": "Prior model used a 40% task-time reduction.", "quantitative_use": "Remove this drafting-stage anchor; no new workpaper/email percentage is supplied in its place.", "plain_summary": "Exclude the 2023 writing experiment from collections and audit-workpaper calculations." }, "accounting-microsoft_collections": { "id": "accounting-microsoft_collections", "title": "Microsoft Frontier Finance: Collections Email Agent", "url": "https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/mcaps/documents/fy26/Collections-Email-Agent.pdf", "publication_date": "Document created 2026-05-08; public by 2026-05-11 according to completed source review", "evidence_period": "Measurement dates and model version unstated.", "status": "retain", "kind": "First-party internal deployment case", "reported_result": "About 60,000 incoming emails/month, about 55% resolved automatically and 36,000 annual hours saved.", "transfer": "Demonstrates intent handling and routine reply preparation at scale. Autonomous sending exceeds this project's approval boundary.", "limits": "No baseline human-hours or time-study denominator. Neither 55% automated volume nor the annual hours supplies a percentage saving for a reviewed assistant.", "verified_this_pass": "Live first-party PDF text; document date inherited from source-review-2026-10-25/review.json, whose PDF metadata check is recorded.", "project_ids": [ "AI-02" ], "evidence_type": "First-party internal deployment case", "date_verification": "Live first-party PDF text; document date inherited from source-review-2026-10-25/review.json, whose PDF metadata check is recorded.", "research_status": "retain_as_illustrative", "plain_summary": "About 60,000 incoming emails/month, about 55% resolved automatically and 36,000 annual hours saved.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live first-party PDF text; document date inherited from source-review-2026-10-25/review.json, whose PDF metadata check is recorded." }, "accounting-microsoft_finance_2025_ga": { "id": "accounting-microsoft_finance_2025_ga", "title": "Microsoft: Empowering Finance with an AI Assistant in Microsoft 365 Copilot", "url": "https://www.microsoft.com/en-us/dynamics-365/blog/it-professional/2025/10/20/empowering-finance-with-an-ai-assistant-in-microsoft-365-copilot/", "publication_date": "2025-10-20", "evidence_period": "Product availability in October 2025; unnamed pilots have unstated measurement periods.", "status": "retain", "kind": "Product announcement", "reported_result": "ERP-grounded draft customer replies for staff review and Excel reconciliation matching/exception assistance.", "transfer": "Confirms both specified human-reviewed workflow shapes are technically available.", "limits": "Days-to-hours pilot wording has no auditable human-time denominator. Supplies no task percentage or endpoint.", "verified_this_pass": "Live primary page with visible date.", "project_ids": [ "AI-02", "AI-03" ], "evidence_type": "Product announcement", "date_verification": "Live primary page with visible date.", "research_status": "retain_as_workflow_reference", "plain_summary": "ERP-grounded draft customer replies for staff review and Excel reconciliation matching/exception assistance.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live primary page with visible date." }, "accounting-us_venture": { "id": "accounting-us_venture", "url": "https://www.microsoft.com/en/customers/story/20600-us-venture-microsoft-365-copilot-for-finance", "title": "Microsoft: U.S. Venture gains efficiencies and insights with Dynamics 365 and Copilot for Finance", "publication_date": "2025-01-28", "evidence_period": "The manager reports three months of use by the January 2025 publication, implying use spanning late 2024. Exact start and measurement dates unstated.", "status": "hold", "reason": "Recent publication contains a result from a likely mixed late-2024/early-2025 period. Hold the quantitative result under the technology-date cutoff rather than presenting it as a clean 2025 measurement.", "verified_date_evidence": { "url": "https://www.microsoft.com/en/customers/story/20600-us-venture-microsoft-365-copilot-for-finance", "method": "Visible primary date and customer quotation", "detail": "1/28/2025; manager describes three months of reconciliation use. Late-2024 overlap is an inference from that duration, not a disclosed exact launch date." }, "project_ids": [ "AI-03" ], "kind": "Vendor-authored named customer case", "result": "Named manager reports 80% less US Bank reconciliation time; accounting team saves more than 30 hours/month across credit-card and bank-statement reconciliation.", "denominator_and_limits": "Individual statement for one reconciliation; team total spans two recurring processes. No timed sample, independent review or task-stage distribution.", "quantitative_use": "Remove 80% from active recent range anchors; retain only in exclusion/hold audit log.", "plain_summary": "The 80% result was published in January 2025 after three months of use, so it includes likely 2024 use. Hold it rather than relabeling it as current technology evidence." }, "accounting-microsoft_reconciliation_pilot": { "id": "accounting-microsoft_reconciliation_pilot", "url": "https://www.microsoft.com/en-us/copilot/blog/2024/02/29/introducing-microsoft-copilot-for-finance-transform-finance-with-next-generation-ai-in-microsoft-365/", "title": "Microsoft: Introducing Copilot for Finance", "publication_date": "2024-02-29", "evidence_period": "Pre-launch pilot by February 2024", "status": "exclude", "reason": "Explicitly excluded publication year and old pilot technology.", "verified_date_evidence": { "url": "https://www.microsoft.com/en-us/copilot/blog/2024/02/29/introducing-microsoft-copilot-for-finance-transform-finance-with-next-generation-ai-in-microsoft-365/", "method": "Visible date on live primary page", "detail": "Feb 29, 2024" }, "project_ids": [ "AI-03" ], "kind": "First-party product launch / internal pilot", "result": "Existing model used a one-to-two-hour weekly task reduced to ten minutes.", "quantitative_use": "Remove the old 83–92% narrow-task anchor and reasoning based on it.", "plain_summary": "Exclude the February 2024 reconciliation pilot." }, "accounting-capitec": { "id": "accounting-capitec", "url": "https://www.microsoft.com/en/customers/story/19093-capitec-bank-azure-open-ai-service", "title": "Microsoft: Capitec Bank employees save more than one hour per week", "publication_date": "2024-10-15", "evidence_period": "Deployment before October 2024", "status": "exclude", "reason": "Original primary case is October 2024. Later 2025/2026 directories must not refresh its date.", "verified_date_evidence": { "url": "https://www.microsoft.com/en/customers/story/19093-capitec-bank-azure-open-ai-service", "method": "Visible date on live primary page", "detail": "October 15, 2024" }, "project_ids": [ "AI-03" ], "kind": "Vendor-authored named customer anecdote", "result": "Six weekly hours of an employee's reconciliation replaced by one-minute runs; run frequency not specified.", "denominator_and_limits": "Weekly baseline and per-run new runtime have mismatched time bases; no valid percentage without frequency and human-review time.", "quantitative_use": "Already unsuitable for normalization; now exclude for date as well.", "plain_summary": "Exclude the original 2024 Capitec story, including later retellings." }, "accounting-chatham_2026": { "id": "accounting-chatham_2026", "title": "OpenAI: Chatham scales its capital markets expertise with OpenAI", "url": "https://openai.com/index/chatham-financial/", "publication_date": "2026-10-02", "evidence_period": "Early measurement reported October 2026; exact dates unstated; GPT-5.6 generation explicitly identified.", "status": "retain", "kind": "Vendor-authored named customer early measurement", "reported_result": "Trade-validation review reportedly falls from about 30 minutes to under four; evidence is assembled, terms compared and discrepancies flagged.", "transfer": "Supports substantial reduction in narrow comparison/evidence preparation; can inform feasibility of the upper reconciliation scenario.", "limits": "Trade validation differs from ledger reconciliation. No sample, error rate or timing protocol; comparison against experienced reviewers is ongoing. It does not justify removing independent review.", "verified_this_pass": "Live primary article and visible date.", "project_ids": [ "AI-03" ], "evidence_type": "Vendor-authored named customer early measurement", "date_verification": "Live primary article and visible date.", "research_status": "retain_as_adjacent_reference", "plain_summary": "Trade-validation review reportedly falls from about 30 minutes to under four; evidence is assembled, terms compared and discrepancies flagged.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live primary article and visible date." }, "accounting-provectus_johnson_lambert": { "id": "accounting-provectus_johnson_lambert", "title": "Provectus: Johnson Lambert reinvents insurance auditing with generative AI", "url": "https://www.provectus.com/case-studies/johnson-lambert-audit/", "publication_date": "2026-04-20, first-party metadata verified in completed source review", "evidence_period": "Exact measurement period absent; listed Claude 3.7 Sonnet and Gemini 2.5 Pro establish post-2024 technology.", "status": "retain", "kind": "Vendor-authored named customer deployment", "reported_result": "50% lower document-processing time; auditors inspect extracted tables against originals and correct values.", "transfer": "Relevant human-checked extraction analogue for the lower audit preparation scenario.", "limits": "PDF financial-report extraction differs from ticket testing. No timed sample or task breakdown. Other broader audit headlines conflict; do not use them or convert an efficiency increase into an equal time reduction. Its 50% document-processing result is neither a universal cap nor a guaranteed stage effect.", "verified_this_pass": "Live primary article confirms workflow, named model generations and narrower result; publication date inherited from recorded first-party metadata check.", "project_ids": [ "AI-08" ], "evidence_type": "Vendor-authored named customer deployment", "date_verification": "Live primary article confirms workflow, named model generations and narrower result; publication date inherited from recorded first-party metadata check.", "research_status": "retain_as_adjacent_reference", "plain_summary": "50% lower document-processing time; auditors inspect extracted tables against originals and correct values.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live primary article confirms workflow, named model generations and narrower result; publication date inherited from recorded first-party metadata check." }, "accounting-microsoft_sox": { "id": "accounting-microsoft_sox", "title": "Microsoft Frontier Finance: SOX Change Management Testing Agent", "url": "https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/mcaps/documents/fy26/SOX-Change-Management-Testing-Agent.pdf", "publication_date": "File modified 2026-05-07; public by 2026-05-11 according to completed source review", "evidence_period": "Deployment/measurement dates and model version absent; listed technology includes Bing Chat. A 2026 file date does not establish recent deployment technology.", "status": "retain", "kind": "First-party internal deployment case", "reported_result": "850 annual hours saved. Two agents extract ticket attributes, prepopulate rule checks and mark the matching source evidence.", "transfer": "Direct match to ticket evidence extraction and annotation; supports preparing standardized source-linked workpapers.", "limits": "No baseline hours, ticket volume, correction rate or timed sample. Cannot infer a percentage or establish current-model speed. Auditor verification remains in the scenario.", "verified_this_pass": "Downloaded first-party PDF with curl and extracted its text with pdftotext after web reader failed; date provenance remains the existing metadata audit.", "project_ids": [ "AI-08" ], "evidence_type": "First-party internal deployment case", "date_verification": "Downloaded first-party PDF with curl and extracted its text with pdftotext after web reader failed; date provenance remains the existing metadata audit.", "research_status": "retain_as_illustrative", "plain_summary": "850 annual hours saved. Two agents extract ticket attributes, prepopulate rule checks and mark the matching source evidence.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Downloaded first-party PDF with curl and extracted its text with pdftotext after web reader failed; date provenance remains the existing metadata audit." }, "accounting-audit_rct": { "id": "accounting-audit_rct", "url": "https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6095166", "title": "Jezierski et al.: The Effect of Large Language Models on Audit Quality", "publication_date": "Posted 2026-02-03; paper dated 2025-12-19; last revised 2026-01-19", "evidence_period": "Experiment dates and model version unverified. Authors' university 2025 research poster describes the RAG treatment but not a completed experimental period; it is not enough to establish a study date.", "status": "retain", "reason": "Recent primary preprint record verified. Retain only qualitative abstract findings until full paper and experiment timing can be inspected; no numerical effect is available here.", "verified_date_evidence": { "url": "https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6095166", "method": "Live search-indexed primary SSRN record; direct fetch returned HTTP 403", "detail": "SSRN record lists 41 pages; Posted 3 Feb 2026; revised 19 Jan 2026; Date Written December 19, 2025. University poster: https://www.heicad.hhu.de/fileadmin/redaktion/HeiCad/heicad_pdf/Poster_Retreat_2025/Jezierski_AI_Upskilling__Poster.pdf" }, "project_ids": [ "AI-08" ], "kind": "Preregistered randomized field experiment; preprint", "result": "Abstract reports faster completion and better documentation for 100 professional auditors given a firm-specific LLM, with no improvement in adherence to internal audit guidelines.", "denominator_and_limits": "Complex inventory-valuation task rather than ticket evidence extraction. Full paper inaccessible, so no numerical time effect, exact model or experiment date newly verified. Do not label peer reviewed.", "quantitative_use": "Qualitative support only; cannot supply the missing workpaper-drafting percentage.", "plain_summary": "A recent working paper reports faster inventory-audit work and better documentation, but no improvement in following internal guidelines. Its experiment dates and numerical timing result remain unverified.", "doc_bullet": "[Jezierski et al. (2026 preprint)](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6095166): 100 auditors with an AI assistant completed an inventory-valuation task faster and produced better documentation. Experiment dates and numerical time effects remain unverified; the task differs from SOX ticket testing." }, "accounting-afm_controls": { "id": "accounting-afm_controls", "title": "Netherlands AFM: 12 Building blocks for controlled use of auditing tools", "url": "https://www.afm.nl/~/profmedia/files/rapporten/2025/12-building-blocks-for-controlled-use-of-auditing-tools.pdf", "publication_date": "2025-12", "evidence_period": "2025 supervisory guidance; older firm-selection data are unused.", "status": "retain", "kind": "Regulator guidance", "reported_result": "No productivity result. Requires reliable input, verifiable outputs and continuing auditor responsibility.", "transfer": "Supports retaining substantive review and evidence tracing. It neither fixes review hours nor prevents review from getting faster through organized evidence.", "limits": "No numerical effect; none used.", "verified_this_pass": "Live primary PDF.", "project_ids": [ "AI-03", "AI-08" ], "evidence_type": "Regulator guidance", "date_verification": "Live primary PDF.", "research_status": "retain_as_requirements_reference", "plain_summary": "No productivity result. Requires reliable input, verifiable outputs and continuing auditor responsibility.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live primary PDF." }, "planning-natura_sap": { "title": "SAP: Natura gross-margin application", "url": "https://news.sap.com/2026/07/natura-co-transforming-finance-generative-ai/", "publication_date": "2026-07-08", "date_verification": "Visible first-party article date and explicit August 2025 production go-live.", "evidence_period": "Production since August 2025; use in Ecuador described July 2026. Originating early-2024 prototype is excluded from quantitative support.", "evidence_type": "Named vendor/customer production case", "reported_result": "ERP-connected gross-margin revenue/cost explanations with human interpretation and decision control; no numeric human-time saving.", "use": "Supports the AI-04 workflow and approved-calculation boundary; supplies no task saving rate.", "limits": "Six months describes implementation. No labor baseline or saving is reported.", "checked": "2026-10-25", "id": "planning-natura_sap", "research_status": "retain", "status": "retain", "kind": "Named vendor/customer production case", "plain_summary": "ERP-connected gross-margin revenue/cost explanations with human interpretation and decision control; no numeric human-time saving.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Visible first-party article date and explicit August 2025 production go-live." }, "planning-pwc_cfo_advisor": { "title": "PwC CFO Advisor implementation case", "url": "https://cee.pwc.com/case-studies/how-cfo-advisor-enhances-and-automates-financial-analysis-and-storytelling-with-ai.html", "publication_date": "2026-04-01", "date_verification": "Live first-party HTML: pwcReleaseDate=2026-04-01T17:45:00.000+03:00; pwcLastModifiedDate=2026-04-01T14:00:19.185Z.", "evidence_period": "Deployment and observation calendar unstated; operating implementation described in the 2026 case.", "evidence_type": "Consultancy-authored anonymous TMT customer case", "reported_result": "About 90% less variance-analysis effort across approximately 20 configured analysis types, with controllers and FP&A editing and approving drafts.", "use": "90% applies only to eligible recurring variance preparation in the AI-04 scenario; it is not reduced by an invented efficiency multiplier.", "limits": "No baseline hours, sample, timing protocol or quality comparison. Tool consolidation and fewer report pages accompany the result. The scenario assumes that stage result can transfer at unchanged output; it is not demonstrated for the scoped workflow. Ambiguous FTE wording is unused.", "checked": "2026-10-25", "id": "planning-pwc_cfo_advisor", "research_status": "retain", "status": "retain", "kind": "Consultancy-authored anonymous TMT customer case", "plain_summary": "About 90% less variance-analysis effort across approximately 20 configured analysis types, with controllers and FP&A editing and approving drafts.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live first-party HTML: pwcReleaseDate=2026-04-01T17:45:00.000+03:00; pwcLastModifiedDate=2026-04-01T14:00:19.185Z." }, "planning-clickup_pigment": { "title": "Pigment: ClickUp Analyst Agent case", "url": "https://www.pigment.com/customer-stories/clickup", "publication_date": "2026-09-03", "date_verification": "Live first-party JSON-LD: datePublished=2026-09-03T18:56:58.573Z; dateModified=2026-09-03T18:53:21.520Z.", "evidence_period": "Monthly operating use described in 2026; deployment/measurement months unstated. An agent launch article visibly published September 2025 also describes the result, so this is not assumed to originate in 2026.", "evidence_type": "Named vendor/customer deployment case", "reported_result": "Reported 1-1.5 days saved per person monthly across six finance staff on first-draft budget-variance reports and driver identification.", "use": "Independent named workflow support for AI-05; saved days are not converted to a percentage.", "limits": "No baseline reporting hours, hours/day or timing protocol. Separate Modeler Agent claims are excluded.", "checked": "2026-10-25", "id": "planning-clickup_pigment", "research_status": "retain", "status": "retain", "kind": "Named vendor/customer deployment case", "plain_summary": "Reported 1-1.5 days saved per person monthly across six finance staff on first-draft budget-variance reports and driver identification.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live first-party JSON-LD: datePublished=2026-09-03T18:56:58.573Z; dateModified=2026-09-03T18:53:21.520Z." }, "planning-microsoft_search": { "id": "planning-microsoft_search", "url": "https://www.microsoft.com/en-us/worklab/work-trend-index/copilots-earliest-users-teach-us-about-generative-ai-at-work", "title": "What Can Copilot’s Earliest Users Teach Us About Generative AI at Work?", "publication_date": "2023-11-15", "evidence_period": "2023 early Copilot studies", "status": "exclude", "reason": "Explicitly excluded publication and technology period under the user's date rule. Remove as an active retrieval-time anchor.", "verified_date_evidence": "Visible report date November 15, 2023; methodology describes 2023 surveys and early-access product experiments.", "project_ids": [ "AI-04", "AI-06" ], "kind": "Vendor controlled simulated-work experiments", "plain_summary": "This supplied the old approximately 25–27% search-time reduction. It should no longer calibrate the current finance knowledge or margin ranges." }, "planning-noy_zhang": { "id": "planning-noy_zhang", "url": "https://doi.org/10.1126/science.adh2586", "title": "Experimental evidence on the productivity effects of generative artificial intelligence", "publication_date": "2023-07-13", "evidence_period": "2023 ChatGPT writing experiment", "status": "exclude", "reason": "2023 publication and experiment. Remove the 40% writing-time result from active calibration rather than retain it indirectly in drafting assumptions.", "verified_date_evidence": "Science DOI publication record; existing registry records July 13, 2023 online publication.", "project_ids": [ "AI-04", "AI-05", "AI-06", "AI-07" ], "kind": "Randomized writing experiment", "plain_summary": "The old 40% writing-task result was an adjacent-task reference, not measured finance work. Its removal affects all four existing modeled ranges." }, "planning-jagged_frontier": { "id": "planning-jagged_frontier", "url": "https://www.hbs.edu/ris/Publication%20Files/dell-acqua-et-al-2026-navigating-the-jagged-technological-frontier_5c589c8c-fbb5-458f-b285-c944746cd717.pdf", "title": "Navigating the Jagged Technological Frontier", "publication_date": "2026-03-11", "evidence_period": "June 2023 GPT-4 experiment; original working paper 2023", "status": "exclude", "reason": "The 2026 journal date does not make the underlying 2023 experiment recent. Exclude both the speed and correctness results from active project evidence under this date rule.", "verified_date_evidence": "Final paper methods, PDF page 4, describes GPT-4 as the most capable available model in June 2023. The final article reuses the 758-consultant experiment.", "project_ids": [ "AI-04", "AI-05" ], "kind": "Randomized field experiment republished in 2026", "plain_summary": "Its experimental speed and error findings describe 2023 GPT-4. Human verification can still be a project requirement, but should not be justified here as recent empirical evidence." }, "planning-bcg_knowledge": { "id": "planning-bcg_knowledge", "url": "https://media-publications.bcg.com/BCG-Executive-Perspectives-Future-of-Procurement-with-AI-2025-27Feb2025.pdf", "title": "Procurement: Maximizing Value Potential from AI in 2025", "publication_date": "2025-02", "evidence_period": "Unnamed OEM deployment and measurement dates unstated; February 2025 report discusses work with clients over the preceding year.", "status": "retain", "reason": "Verified February 2025 publication qualifies as a recent case. The measurement calendar and model remain unknown; no evidence establishes that the reported result was actually measured in 2023/2024. Do not call it a current-2026-model experiment.", "verified_date_evidence": "PDF cover states February 2025. Slide 13 provides the OEM case but no measurement dates or model version. Intro discusses the preceding year of client work.", "project_ids": [ "AI-04", "AI-06" ], "kind": "Consultancy deployment case; unnamed global OEM", "plain_summary": "BCG reports 50–75% less internal research time, up to about 20 minutes per search, for an unnamed global OEM. Task coverage, validation boundary, observation period and model are unstated. This is adjacent internal-knowledge evidence, not full finance-workflow timing." }, "planning-deloitte_dartbot": { "id": "planning-deloitte_dartbot", "url": "https://dart.deloitte.com/USDART/home/help/dart-help-1/introducing-dartbot-generative-ai-powered-accounting", "title": "Introducing DARTbot: Generative AI-Powered Accounting Research", "publication_date": null, "evidence_period": "Current external-subscriber product description undated; original internal DARTbot launched October 2023.", "status": "hold", "reason": "No verified date for this particular external-product page. It has no measured outcome anyway. The original 2023 launch is separately excluded; do not label this undated page as a 2025/2026 case.", "verified_date_evidence": "Visible page and primary HTML provide no verified publication date. Deloitte's dated October 16, 2023 press release establishes the original internal deployment.", "project_ids": [ "AI-06" ], "kind": "Undated product capability description", "plain_summary": "The product answers accounting questions, summarizes guidance and links sources. This explains a possible use, but supplies neither recent case-study dates nor measured time savings." }, "planning-dartbot_launch_2023": { "id": "planning-dartbot_launch_2023", "url": "https://www.deloitte.com/us/en/about/press-room/deloitte-launches-innovative-dartbot-internal-chatbot.html", "title": "Deloitte Launches Innovative DARTbot Internal Chatbot", "publication_date": "2023-10-16", "evidence_period": "2023 internal launch to nearly 18,000 US audit and assurance professionals", "status": "exclude", "reason": "Explicitly a 2023 deployment account, excluded by user instruction.", "verified_date_evidence": "Visible press-release date October 16, 2023 and dateline identify launch timing.", "project_ids": [ "AI-06" ], "kind": "First-party launch announcement", "plain_summary": "The original DARTbot launch provides no measured time-saving result and is outside the permitted date period." }, "planning-amazon_fintech": { "id": "planning-amazon_fintech", "url": "https://aws.amazon.com/solutions/case-studies/amazon-finance-case-study/", "title": "Delivering regulatory insights 92% faster using AWS with Amazon Finance Technology", "publication_date": null, "evidence_period": "World Wide Watch VAT deployment and measurement dates unstated; first publication not verified.", "status": "hold", "reason": "Primary HTML only verifies an October 2026 update time, not publication or measurement. Hold until a dated primary account establishes the VAT result. The dated 2026 regulatory-inquiry article below describes a different workflow and cannot date these claims.", "verified_date_evidence": "HTML og:updated_time=2026-10-15T17:27:32.316-0700. No verified first-publication field or measurement period found.", "project_ids": [ "AI-06", "AI-07" ], "kind": "Undated first-party deployment case", "plain_summary": "Reports VAT insight delivery falling from 26 to 2 minutes and 80% expert approval of suggested content. Complete human review/correction labor is not measured. A separate chatbot accuracy figure is also not a labor-time result." }, "planning-amazon_regulatory_rag_2026": { "id": "planning-amazon_regulatory_rag_2026", "url": "https://aws.amazon.com/blogs/machine-learning/how-amazon-finance-streamlines-regulatory-inquiries-by-using-generative-ai-on-aws/", "title": "How Amazon Finance streamlines regulatory inquiries by using generative AI on AWS", "publication_date": "2026-05-12", "date_verification": "Visible 12 MAY 2026 dateline.", "evidence_period": "May 2026 account names Claude Sonnet 4.5; exact go-live/observation dates unstated.", "status": "retain", "kind": "First-party technical deployment account", "result": "Finds historical documents and precedents and supports regulatory-response preparation. Parallel retrieval falls from 10 seconds to under 2 seconds.", "denominator_and_limits": "Retrieval number compares system configurations; no human-hours result. Regulatory inquiries differ from ongoing tax monitoring.", "quantitative_use": "Mechanism only; no numeric human-time factor is derived from retrieval latency.", "project_ids": [ "AI-06", "AI-07" ], "evidence_type": "First-party technical deployment account", "reported_result": "Finds historical documents and precedents and supports regulatory-response preparation. Parallel retrieval falls from 10 seconds to under 2 seconds.", "limits": "Retrieval number compares system configurations; no human-hours result. Regulatory inquiries differ from ongoing tax monitoring.", "research_status": "retain", "plain_summary": "Finds historical documents and precedents and supports regulatory-response preparation. Parallel retrieval falls from 10 seconds to under 2 seconds.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Visible 12 MAY 2026 dateline." }, "planning-fm_knowledge_2026": { "id": "planning-fm_knowledge_2026", "url": "https://www.microsoft.com/en/customers/story/26503-fm-azure", "title": "FM empowers 1,500+ engineers with trusted, AI-assisted access to engineering knowledge using Azure OpenAI", "publication_date": "2026-05-28", "date_verification": "Visible 5/28/2026 dateline.", "evidence_period": "First two production months, without calendar dates. Model version and timing protocol unstated.", "status": "retain", "kind": "Vendor/customer estimate; adjacent engineering knowledge search", "result": "Estimated 6–10 minutes saved per search; 17,000 queries in the first two production months across more than 1,500 engineers.", "denominator_and_limits": "No baseline duration or full checking-time boundary. Estimated minutes cannot be converted to a percentage.", "quantitative_use": "Supports potentially substantial lookup savings and source-grounded implementation; neither calibrates a general percentage nor caps lookup savings.", "project_ids": [ "AI-06" ], "evidence_type": "Vendor/customer estimate; adjacent engineering knowledge search", "reported_result": "Estimated 6–10 minutes saved per search; 17,000 queries in the first two production months across more than 1,500 engineers.", "limits": "No baseline duration or full checking-time boundary. Estimated minutes cannot be converted to a percentage.", "research_status": "retain", "plain_summary": "Estimated 6–10 minutes saved per search; 17,000 queries in the first two production months across more than 1,500 engineers.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Visible 5/28/2026 dateline." }, "planning-heineken_hoppy_2025": { "id": "planning-heineken_hoppy_2025", "url": "https://news.sap.com/africa/2025/05/leading-brands-showcase-sap-fuelled-ai-cloud-innovation-at-sapphire-2025/", "title": "Leading Brands Showcase SAP-fuelled AI, Cloud Innovation at Sapphire 2025", "publication_date": "2025-05-29", "evidence_period": "Hoppy existed during 2024, according to Heineken's 2024 annual report. The 15-to-1-minute measurement period is unstated.", "status": "hold", "reason": "Verified May 2025 primary publication qualifies as a recent case. Prior existence in 2024 does not prove the reported timing was measured in 2024. Explicitly disclose the unstated measurement calendar and model; do not relabel it a 2025-only experiment. Root review holds the timing claim: Hoppy existed in 2024 and no newer measurement period or model is identified.", "verified_date_evidence": "SAP visible May 29, 2025 dateline. Heineken annual report 2024, published February 2025, describes Hoppy supporting finance teams: https://www.heinekenholding.com/sites/heinekenholding-v2/files/2025-03/heineken_n_v_annual_report_2024_final_20feb2025.pdf .", "project_ids": [ "AI-06" ], "kind": "Named vendor/customer deployment case; observation date uncertain", "plain_summary": "SAP reports information retrieval falling from 15 to 1 minute with Hoppy over enterprise data. Heineken had already described Hoppy in its 2024 annual report. The 2025 timing claim has no stated measurement dates, sample or review boundary, so it supports a large task-level example rather than a general full-workflow percentage." }, "planning-persistent_knowledge_undated": { "id": "planning-persistent_knowledge_undated", "url": "https://www.persistent.com/client-success/when-customer-conversations-depend-on-finding-the-right-answer-fast/", "title": "When Customer Conversations Depend on Finding the Right Answer Fast", "publication_date": null, "evidence_period": "Multi-year financial-services engagement; initial deployment and measurement dates unstated.", "status": "hold", "reason": "No publication metadata, dateline or dated primary corroboration found. Date-like asset URLs do not establish the article or evidence date.", "verified_date_evidence": "Primary HTML retrieved using system curl. No datePublished, article:published_time or equivalent publication field found; search index describes a multi-year engagement without dates.", "project_ids": [ "AI-06" ], "kind": "Undated consultant case; unnamed financial-services client", "plain_summary": "Reports about two minutes saved per customer interaction with a RAG assistant over internal policies and service documents, supporting over 18,000 users. Human validation is retained. No verifiable measurement period or timing protocol is provided." }, "planning-tr_cocounsel_corporate": { "id": "planning-tr_cocounsel_corporate", "url": "https://tax.thomsonreuters.com/en/products/cocounsel-tax/corp", "title": "CoCounsel Tax: The AI tax assistant for in-house teams", "publication_date": "2026-06-29", "date_verification": "Primary HTML w-published-date and analytics publishDate=2026-06-29; visible survey footnote states November–December 2025.", "evidence_period": "Self-reported user survey November–December 2025.", "status": "retain", "kind": "Vendor user survey", "result": "Users report an average 32% less tax-research time.", "denominator_and_limits": "Sample size, task mix and review boundary not disclosed; not independently timed or verified. Research differs from scanning five sources and obtaining approval for every relevant update.", "quantitative_use": "Adjacent research reference used to illustrate the lower read/draft scenario assumption; no causal/full-workflow claim.", "project_ids": [ "AI-06", "AI-07" ], "evidence_type": "Vendor user survey", "reported_result": "Users report an average 32% less tax-research time.", "limits": "Sample size, task mix and review boundary not disclosed; not independently timed or verified. Research differs from scanning five sources and obtaining approval for every relevant update.", "research_status": "retain", "plain_summary": "Users report an average 32% less tax-research time.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Primary HTML w-published-date and analytics publishDate=2026-06-29; visible survey footnote states November–December 2025." }, "planning-tax_plat": { "id": "planning-tax_plat", "url": "https://arxiv.org/abs/2503.03444", "title": "Taxation Perspectives from Large Language Models: A Case Study on Additional Tax Penalties", "publication_date": "2025-03-05; revised 2026-02-08", "evidence_period": "Recent 2025 model snapshots in revised benchmark; exact runs unstated.", "status": "retain", "kind": "Tax reasoning benchmark", "result": "Continuing difficulties applying tax law to facts and conclusions.", "denominator_and_limits": "300 Korean tax-penalty questions; no human-time study.", "quantitative_use": "Supports preserving applicability review; no percentage or correction-hours factor derived.", "project_ids": [ "AI-07" ], "evidence_type": "Tax reasoning benchmark", "reported_result": "Continuing difficulties applying tax law to facts and conclusions.", "limits": "300 Korean tax-penalty questions; no human-time study.", "research_status": "retain_as_accuracy_context", "plain_summary": "Continuing difficulties applying tax law to facts and conclusions.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "" }, "planning-datarails_review": { "id": "planning-datarails_review", "url": "https://www.datarails.com/blog/cfo-survey-ai-finance/", "title": "2026 CFO Survey: Finance Spends 26% of Week Checking AI", "publication_date": "2026-10-07", "evidence_period": "July 2026 survey", "status": "retain", "reason": "Explicit 2026 survey dates. Retain as general finance context only; it does not estimate incremental verification time for an individual project.", "verified_date_evidence": "Visible October 7, 2026 article date and methodology stating July 2026 online survey of 270 US CFOs and finance leaders.", "project_ids": [ "AI-04", "AI-05", "AI-06", "AI-07" ], "kind": "Vendor-commissioned panel survey", "plain_summary": "Large-company finance leaders report substantial checking of AI output and little trust in unattended board reports. There is no pre-AI comparison or task-level denominator; the 26% figure cannot be subtracted from project savings." }, "ir-procurement-q4-forrester-2025": { "id": "ir-procurement-q4-forrester-2025", "url": "https://tei.forrester.com/go/Q4/platform/index.html", "title": "The Total Economic Impact of Q4 Platform", "publication_date": "2025-06", "evidence_period": "Interview, deployment and measurement dates undisclosed; three-year financial projection, not three years of observed results. Model version undisclosed.", "status": "retain", "reason": "Clearly dated 2025 primary commissioned research. Suitable as a reported task example; current model generation and complete-workflow effect remain unverified.", "verified_date_evidence": "Report cover says commissioned by Q4, June 2025. Method describes one decision-maker interview. Tables A7-A14 give the task hours.", "project_ids": [ "AI-09" ], "kind": "Sponsored single-customer interview and financial model", "plain_summary": "One energy company's inputs to a June 2025 Forrester/Q4 model put script drafting at 36 to 4 hours and each peer-call summary at 10 to 2 hours. Full editing, checking and approval time was not separately measured.", "numerical_support": "Task-specific reported inputs; no directly observed 25-55% complete-workflow range.", "independence_note": "Anonymous energy-company interview could overlap the Murphy USA case. Do not present the two as proven independent replications." }, "ir-procurement-murphy-q4": { "id": "ir-procurement-murphy-q4", "url": "https://www.q4inc.com/resource-center/blog/how-murphy-usa-unlocked-their-full-potential-with-q4", "title": "From Manual to Strategic: Murphy USA's IR Transformation", "publication_date": "2025-04-30", "updated_date": "2026-08-11", "evidence_period": "Deployment, measurement dates and model version undisclosed.", "status": "retain", "reason": "First-party page explicitly distinguishes original 2025 publication from 2026 update. Result combines AI and other Q4 platform features.", "verified_date_evidence": "Visible byline: Published April 30, 2025; Updated August 11, 2026.", "project_ids": [ "AI-09" ], "kind": "Named vendor customer case", "plain_summary": "Murphy USA reports three days saved per earnings cycle and half as many people needed on competitor calls. It used Q4's AI, transcripts and other platform tools together; the article does not isolate AI's contribution or give total staff hours.", "numerical_support": "Reported platform benefit, with no compatible workflow denominator." }, "ir-procurement-q4-workflow-2026": { "id": "ir-procurement-q4-workflow-2026", "url": "https://www.q4inc.com/resource-center/blog/can-ai-actually-handle-earnings-season-a-workflow-test-for-ir-teams?hs_amp=true", "title": "Can AI Actually Handle Earnings Season? A Workflow Test for IR Teams", "publication_date": "2026-08-14", "evidence_period": "2026 product/workflow guidance; no experiment or deployment measurement.", "status": "retain", "reason": "Dated current guidance can explain retained work, but provides no measured savings.", "verified_date_evidence": "Visible author/date line: The Q4 Insights Team, August 14, 2026.", "project_ids": [ "AI-09" ], "kind": "Vendor workflow guidance", "plain_summary": "Q4's 2026 workflow guide describes AI drafting and summarization, followed by source checks and finance, legal and executive approval. It reports no time-saving result.", "numerical_support": "None." }, "ir-procurement-niri-uf-research": { "id": "ir-procurement-niri-uf-research", "url": "https://www.niri.org/publications/research/", "title": "2025 NIRI and University of Florida Research Survey on AI within IR", "publication_date": "2025", "evidence_period": "Survey period and methods unavailable on public listing.", "status": "hold", "reason": "Public page verifies a 2025 survey exists, but results are members-only. It cannot verify workflow savings.", "verified_date_evidence": "NIRI's public research index labels the survey 2025 and its results Members Only.", "project_ids": [ "AI-09" ], "kind": "Survey index; underlying report inaccessible", "plain_summary": "NIRI lists a 2025 AI survey with the University of Florida. The public page gives no time-saving findings." }, "ir-procurement-bcg-procurement-2025": { "id": "ir-procurement-bcg-procurement-2025", "url": "https://media-publications.bcg.com/BCG-Executive-Perspectives-Future-of-Procurement-with-AI-2025-27Feb2025.pdf", "title": "Maximizing Value Potential from AI in 2025: Procurement", "publication_date": "2025-02", "evidence_period": "Case measurement and model dates undisclosed. Manufacturer took 12 months from first workshop to deployment before publication; project therefore began in 2024 or earlier, but result measurement year is unconfirmed.", "status": "retain", "reason": "Verified February 2025 publication meets the date policy. Measurement/model periods remain unknown; a 12-month project alone cannot date the benefit measurement. Do not present this as verified 2025 model capability.", "verified_date_evidence": "Cover says February 2025. PDF pages 12-13 contain cases and the 12-month rollout statement.", "project_ids": [ "AI-10" ], "kind": "Anonymous consulting delivery cases", "plain_summary": "BCG reports 40% less tender-creation time at an energy provider and roughly 50% improvements in tender drafting and offer comparison at a manufacturer. The models, measurement dates and review protocol are undisclosed.", "numerical_support": "Task-specific case results inform the existing analyst transfer, but do not directly measure the 15-35% complete-event range." }, "ir-procurement-gatx-levelpath": { "id": "ir-procurement-gatx-levelpath", "url": "https://www.levelpath.com/customer-story/how-gatx-scaled-rfp-capacity-and-contract-savings-with-ai-procurement", "title": "How GATX Scaled RFP Capacity and Contract Savings With AI Procurement", "publication_date": null, "evidence_period": "Undisclosed.", "status": "hold", "reason": "Main customer page has no verified original date. Replace with the dated July 2025 first-party webinar recap covering the same claims.", "verified_date_evidence": "No visible publication date found. Footer copyright is not publication evidence.", "project_ids": [ "AI-10" ], "kind": "Undated vendor customer case", "plain_summary": "Main GATX case reports greater RFP capacity and fast bid analysis, but the undated page adds no reliable age evidence beyond the dated companion article." }, "ir-procurement-gatx-levelpath-2025-july": { "id": "ir-procurement-gatx-levelpath-2025-july", "url": "https://www.levelpath.com/blog/how-gatx-is-scaling-procurement-with-an-ai-first-strategy", "title": "How GATX is Scaling Procurement with an AI-First Strategy", "publication_date": "2025-07-21", "evidence_period": "Reports a recent webinar with GATX's procurement director. Deployment and measurement dates and model version undisclosed.", "status": "retain", "reason": "Dated first-party account with named customer participants; useful replacement for the undated customer page.", "verified_date_evidence": "Visible date July 21, 2025; names Bob Campbell, GATX Director of Sourcing and Procurement, and Mari Carmichael, Levelpath.", "project_ids": [ "AI-10" ], "kind": "Vendor recap of named customer webinar", "plain_summary": "GATX reports ten times the RFP capacity with the same headcount, and bid-comparison output in 10-15 seconds rather than hours. These are customer reports; the account does not time the full sourcing event or subsequent human checking.", "numerical_support": "Capacity and subtask speed evidence; does not establish a percentage of total event labor saved." }, "ir-procurement-gatx-levelpath-2025-august": { "id": "ir-procurement-gatx-levelpath-2025-august", "url": "https://www.levelpath.com/blog/how-gatx-uses-levelpaths-ai-to-tackle-procurement-challenges-head-on", "title": "How GATX Uses Levelpath's AI to Tackle Procurement Challenges Head-On", "publication_date": "2025-08-20", "evidence_period": "Deployment/model dates undisclosed.", "status": "retain", "reason": "Dated named case supports additional work made feasible, with no complete workflow time estimate.", "verified_date_evidence": "Visible first-party date August 20, 2025; names Thomas Falstreau, VP of Strategic Sourcing.", "project_ids": [ "AI-10" ], "kind": "Vendor customer account", "plain_summary": "GATX describes using AI to draft RFPs in unfamiliar categories and let stakeholders start sourcing requests themselves. This supports expanded coverage; no whole-event labor reduction is measured.", "numerical_support": "None; mechanism and expanded-work example." }, "ir-procurement-procurement-offers-experiment": { "id": "ir-procurement-procurement-offers-experiment", "url": "https://www.sciencedirect.com/science/article/pii/S1877050926005880", "title": "Enhancing Procurement Processes in Supply Chain Management with Large Language Models", "publication_date": "2026", "evidence_period": "Experiment date and GPT-4o snapshot unverified; only publisher abstract accessible.", "status": "retain", "reason": "Verified 2026 publisher record meets the publication-year policy. Experiment age, full methods and human+AI review labor remain unverified; no precise numerical calibration from abstract alone.", "verified_date_evidence": "Publisher abstract identifies Procedia Computer Science volume 278 (2026), pages 308-315, DOI 10.1016/j.procs.2026.02.467.", "project_ids": [ "AI-10" ], "kind": "Controlled task experiment; abstract only", "plain_summary": "A 2026 paper compares GPT-4o with procurement professionals and reports faster offer analysis. The accessible abstract does not establish experiment date, a transferable savings percentage, or total human review time." }, "ir-procurement-bid-scoring-volatility-2026": { "id": "ir-procurement-bid-scoring-volatility-2026", "url": "https://onlinelibrary.wiley.com/doi/10.1111/jbl.70072", "title": "Do Humans and GAI See Eye to Eye? Implications of LLM Scoring Volatility in Supplier Evaluations", "publication_date": "2026-05-14", "evidence_period": "AI evaluation uses 2025 reasoning models o3, Grok-3-Mini and DeepSeek R1-0528. Exact run dates unreported. Input bids came from 2023-2024 solicitations.", "status": "retain", "reason": "A newer AI benchmark on historical documents. The old bid dates do not establish an old AI experiment.", "verified_date_evidence": "Publisher says first published May 14, 2026; methods section identifies the three 2025 reasoning models and 15 repeated evaluations per model/bid.", "project_ids": [ "AI-10" ], "kind": "Peer-reviewed controlled supplier-scoring benchmark", "plain_summary": "A 2026 study applied three 2025 reasoning models to 123 bids. Compliance checks aligned more consistently with human judgments than competitive scoring. It supports retaining human award judgment; it does not measure labor savings.", "numerical_support": "No time-saving measure." }, "ir-procurement-ieee-tender-2025": { "id": "ir-procurement-ieee-tender-2025", "url": "https://ieeexplore.ieee.org/document/11411639/", "title": "AI in Public Procurement: Case Study of tender Evaluation AI-assisted tender evaluation framework", "publication_date": "2025-11", "online_date": "2026-03-04", "evidence_period": "2025 conference case; experiment date and ChatGPT model undisclosed in accessible abstract. Archived tenders used as inputs.", "status": "retain", "reason": "2025 conference and 2026 publisher release verified. Qualify as abstract-level study with uncertain model/measurement date, pending full methods.", "verified_date_evidence": "IEEE lists conference November 19-21, 2025 and Xplore addition March 4, 2026; university conference schedule lists paper 950.", "project_ids": [ "AI-10" ], "kind": "Conference case study; accessible abstract", "plain_summary": "A 2025 public-procurement study reports over 60% less bid-evaluation time using OCR and ChatGPT, with human validation. The public abstract does not state the model, sample size, experiment date or how validation time was counted.", "numerical_support": "Promising scoped case; insufficient detail for complete-event range calibration." }, "ir-procurement-bcg-procurement-2026": { "id": "ir-procurement-bcg-procurement-2026", "url": "https://www.bcg.com/publications/2026/scaling-agentic-ai-in-tech-procurement", "title": "Scaling Agentic AI in Procurement Is an Organizational Challenge", "publication_date": "2026-07-21", "evidence_period": "BCG calls this its 2026 tech-procurement study; more than 200 respondents and interviews. Exact field dates/model versions unreported.", "status": "retain", "reason": "Explicitly identified 2026 primary survey, useful for dependencies and adoption conditions. Not a task-time experiment.", "verified_date_evidence": "Visible July 21, 2026 date; body describes 2026 survey of CIOs, procurement leaders and specialist buyers.", "project_ids": [ "AI-10", "AI-11" ], "kind": "Consulting survey and interviews", "plain_summary": "BCG's 2026 survey finds production deployments report stronger benefits than pilots. Data quality, legacy integration, staff capability and process redesign constrain progress. It gives no matching percentage of labor saved for one RFP.", "numerical_support": "Qualitative support for foundations and workflow redesign; no calibrated hours." }, "ir-procurement-aws-finance-foundation": { "id": "ir-procurement-aws-finance-foundation", "url": "https://aws.amazon.com/blogs/machine-learning/how-amazon-finance-streamlines-regulatory-inquiries-by-using-generative-ai-on-aws/", "title": "How Amazon Finance streamlines regulatory inquiries by using generative AI on AWS", "publication_date": "2026-05-12", "evidence_period": "2026 architecture/deployment account; implementation and measurement periods unreported.", "status": "retain", "reason": "Dated first-party finance architecture account supports work involved, with no benefit-time measurement. Custom AWS system is larger than the bounded vendor-product setup.", "verified_date_evidence": "AWS byline explicitly says 12 MAY 2026.", "project_ids": [ "AI-11" ], "kind": "First-party finance deployment architecture", "plain_summary": "Amazon Finance's 2026 account shows teams building document retrieval, access controls, conversation history and monitoring for regulatory inquiries. It illustrates foundation work and team upkeep; it gives no setup-hours estimate or reuse saving.", "numerical_support": "No directly measured foundation labor or reuse percentage." }, "ir-procurement-microsoft-shared-governance": { "id": "ir-procurement-microsoft-shared-governance", "url": "https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/sec-gov-intro", "title": "Manage your Copilot Studio projects, an overview", "publication_date": null, "updated_date": "2026-09-14", "evidence_period": "Current documentation; original publication date unverified; no experiment.", "status": "hold", "reason": "2026 update alone does not verify first publication after 2024. Replace for the dated source list with Microsoft's May 2025 controls announcement; current documentation remains background guidance.", "verified_date_evidence": "Page footer explicitly says last updated September 14, 2026, with no original date.", "project_ids": [ "AI-11" ], "kind": "Product guidance", "plain_summary": "Microsoft lists shared access, data policies, testing, release management and monitoring duties. This is guidance, with no measured labor saving." }, "ir-procurement-microsoft-controls-2025": { "id": "ir-procurement-microsoft-controls-2025", "url": "https://techcommunity.microsoft.com/blog/microsoft-security-blog/enterprise-grade-controls-for-ai-apps-and-agents-built-with-azure-ai-foundry-and/4414757", "title": "Enterprise-grade controls for AI apps and agents built with Azure AI Foundry and Copilot Studio", "publication_date": "2025-05-19", "evidence_period": "2025 product capability announcement; no deployment or labor experiment.", "status": "retain", "reason": "Dated Microsoft primary announcement supports foundation capabilities; does not establish savings.", "verified_date_evidence": "First-party indexed article text shows May 19, 2025, Microsoft Security Community Blog and named Microsoft author. Direct body rendering returned empty, so findings limited to indexed first-party text.", "project_ids": [ "AI-11" ], "kind": "Product/security capability announcement", "plain_summary": "Microsoft's May 2025 controls announcement describes shared identity, security and evaluation capabilities for enterprise agents. It supports the foundation's task list; it provides no implementation-hour or percentage-saving result.", "numerical_support": "None." }, "accounting-microsoft_collections_2026_insidetrack": { "id": "accounting-microsoft_collections_2026_insidetrack", "title": "Microsoft Inside Track: Streamlining finance cash collection at Microsoft with AI", "url": "https://www.microsoft.com/insidetrack/blog/streamlining-finance-cash-collection-at-microsoft-with-ai/", "publication_date": "2026-06-04", "evidence_period": "Editor note says the internal data were gathered during writing; exact observation window, model versions and launch dates are unstated.", "status": "retain", "kind": "First-party internal deployment report", "reported_result": "Up to 60% lower inquiry handling time using inline suggestions, call summaries and automatically drafted replies.", "transfer": "Closer to human-assisted collections than the autonomous email PDF. Supports retrieval/context/drafting as saving mechanisms. The maximum is not an average or endpoint estimate.", "limits": "SAP/Dynamics consolidation, standardization and automated communications accompany AI assistance. No sample, task timer or fixed-output quality study is disclosed. Do not transfer its matching-accuracy or resolution-speed metrics into labor percentages.", "verified_this_pass": "Live primary page: visible June 4, 2026 date; handling-time result in operating-results paragraph; period caveat in editor note.", "project_ids": [ "AI-02" ], "evidence_type": "First-party internal deployment report", "date_verification": "Live primary page: visible June 4, 2026 date; handling-time result in operating-results paragraph; period caveat in editor note.", "research_status": "new_recent_primary_reference", "plain_summary": "Up to 60% lower inquiry handling time using inline suggestions, call summaries and automatically drafted replies.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live primary page: visible June 4, 2026 date; handling-time result in operating-results paragraph; period caveat in editor note." }, "accounting-ted_taylor_2026": { "id": "accounting-ted_taylor_2026", "title": "OpenAI: Franchisee reconciles cash in 94% less time with ChatGPT Work", "url": "https://openai.com/index/ted-taylor-restaurant/", "publication_date": "2026-10-20", "evidence_period": "Current customer use described in October 2026; exact deployment/measurement dates and model version unstated. No known older experiment is identified.", "status": "retain", "kind": "Vendor-authored named customer case", "reported_result": "Ted Taylor compares cash transactions with bank deposits monthly across three restaurants. Reported time falls from three hours to five to ten minutes; the article says he reviews results before implementing.", "transfer": "Direct cash-reconciliation feasibility reference. From stated times alone the reported reduction is 94.4–97.2%; use the raw minutes rather than extending its headline to other workflows.", "limits": "No transaction count, exception mix, timed sample, error rate, independent reviewer or breakdown of preparation/review is disclosed. It is unclear whether five to ten minutes includes every retained check. The catalog's two-dataset controlled reconciliation has a more explicit review standard.", "verified_this_pass": "Live primary article, visible date and back-office reconciliation paragraph.", "project_ids": [ "AI-03" ], "evidence_type": "Vendor-authored named customer case", "date_verification": "Live primary article, visible date and back-office reconciliation paragraph.", "research_status": "new_recent_primary_reference", "plain_summary": "Ted Taylor compares cash transactions with bank deposits monthly across three restaurants. Reported time falls from three hours to five to ten minutes; the article says he reviews results before implementing.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live primary article, visible date and back-office reconciliation paragraph." }, "planning-docker_pigment": { "title": "Pigment: Docker monthly close commentary and review", "url": "https://www.pigment.com/customer-stories/docker", "publication_date": "2026-09-11", "date_verification": "Live first-party JSON-LD: datePublished=2026-09-11T19:10:43.460Z; dateModified=2026-09-11T17:15:48.513Z.", "evidence_period": "Current monthly operating use in a 2026 case; exact measurement months unstated.", "evidence_type": "Named vendor/customer deployment; customer self-estimate", "reported_result": "Estimated 30-50% less time on monthly close commentary and review. The process covers actual-versus-budget analysis, narrative drafting and packaging; staff refine judgment and audience context. Analyst Agent and Claude/MCP are used.", "use": "Transfer 30% and 50% to the combined commentary/packaging/review block in AI-05, where the task boundary is closest. Do not assign this result separately to each constituent task.", "limits": "No timed sample, baseline hours, unchanged-output test or detailed correction boundary. Final pack release approval is separately retained. Modeler Agent build/formula results are excluded.", "checked": "2026-10-25", "id": "planning-docker_pigment", "research_status": "retain", "status": "retain", "kind": "Named vendor/customer deployment; customer self-estimate", "plain_summary": "Estimated 30-50% less time on monthly close commentary and review. The process covers actual-versus-budget analysis, narrative drafting and packaging; staff refine judgment and audience context. Analyst Agent and Claude/MCP are used.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live first-party JSON-LD: datePublished=2026-09-11T19:10:43.460Z; dateModified=2026-09-11T17:15:48.513Z." }, "planning-bcg_finance_roi_2025": { "title": "BCG: How to Get ROI from AI in the Finance Function", "url": "https://www.bcg.com/publications/2025/how-finance-leaders-can-get-roi-from-ai", "publication_date": "2025-06-04", "date_verification": "Visible first-party article date; separate finance survey conducted March 2025.", "evidence_period": "Consumer-goods deployment timing unstated; described in the 2025 article.", "evidence_type": "Consultancy-authored anonymous customer transformation", "reported_result": "50% shorter report-generation time after constructing a driver-tree model, integrating data and adding a GenAI query interface.", "use": "Triangulation only: demonstrates the model-backed answer mechanism.", "limits": "Bundled model/data transformation and unclear elapsed-versus-human-time boundary. No numerical rate is transferred because these projects already have usable data and a validated model.", "checked": "2026-10-25", "id": "planning-bcg_finance_roi_2025", "research_status": "retain", "status": "retain", "kind": "Consultancy-authored anonymous customer transformation", "plain_summary": "50% shorter report-generation time after constructing a driver-tree model, integrating data and adding a GenAI query interface.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Visible first-party article date; separate finance survey conducted March 2025." }, "planning-amazon_finance_discovery_2025": { "id": "planning-amazon_finance_discovery_2025", "url": "https://aws.amazon.com/blogs/machine-learning/how-amazon-finance-built-an-ai-assistant-using-amazon-bedrock-and-amazon-kendra-to-support-analysts-for-data-discovery-and-business-insights/", "title": "How Amazon Finance built an AI assistant using Amazon Bedrock and Amazon Kendra to support analysts for data discovery and business insights", "publication_date": "2025-08-28", "date_verification": "Visible dateline and article:published_time / datePublished HTML time: 2025-08-28T12:14:22-08:00.", "evidence_period": "Measurement and deployment calendar dates unstated; explicitly uses Claude 3 Sonnet, a 2024-era model. The 2025 publication does not establish a 2025 experiment or current-model performance.", "status": "hold", "kind": "First-party finance implementation account", "result": "Reports 30% less search time in analyst benchmarking, and separately 45–60 minutes to 5–10 minutes to find relevant information (called 85% improvement). Also reports roughly 8/10 first-attempt retrievals.", "denominator_and_limits": "The 30% benchmark and 85% before/after descriptions have no reconciled task/sample boundary. Over 50 business queries; no human sample size, timing protocol, calendar or complete checking labor. Data discovery/knowledge search, rather than a full finance workflow; 8/10 retrieval is not accepted-answer coverage.", "quantitative_use": "Task-level feasibility for expanded metrics/datasets/precedents scope. Report both time claims or only explicit before/after minutes with a boundary caveat. Do not choose 85% as a general measured effect or set workflow endpoints from it. Model and measurement calendar remain qualified.", "project_ids": [ "AI-06" ], "evidence_type": "First-party finance implementation account", "reported_result": "Reports 30% less search time in analyst benchmarking, and separately 45–60 minutes to 5–10 minutes to find relevant information (called 85% improvement). Also reports roughly 8/10 first-attempt retrievals.", "limits": "The 30% benchmark and 85% before/after descriptions have no reconciled task/sample boundary. Over 50 business queries; no human sample size, timing protocol, calendar or complete checking labor. Data discovery/knowledge search, rather than a full finance workflow; 8/10 retrieval is not accepted-answer coverage.", "reason": "Claude 3 Sonnet is 2024 technology; measurement dates are unknown and the article reports unreconciled 30% and 85% search/discovery results.", "research_status": "hold", "plain_summary": "Reports 30% less search time in analyst benchmarking, and separately 45–60 minutes to 5–10 minutes to find relevant information (called 85% improvement). Also reports roughly 8/10 first-attempt retrievals.", "verified_date_evidence": "Visible dateline and article:published_time / datePublished HTML time: 2025-08-28T12:14:22-08:00." }, "planning-tr_copeland_buhl_2026": { "id": "planning-tr_copeland_buhl_2026", "url": "https://tax.thomsonreuters.com/en/insights/case-studies/from-hours-to-minutes-how-cocounsel-reimagined-tax-workflow-efficiency-at-copeland-buhl", "title": "From hours to minutes: How CoCounsel reimagined tax workflow efficiency at Copeland Buhl", "publication_date": "2026-03-09", "date_verification": "Fetched primary HTML: w-published-date=2026-03-09T00:00:00.000Z and analytics publishDate=2026-03-09.", "evidence_period": "14,000+ messages over six months, exact months unstated. CoCounsel Tax & Audit launched March 2025 per separate dated first-party announcement.", "status": "retain", "kind": "Named vendor/customer implementation account", "result": "Research projects reported 2–3 times faster; reviewers save 15–30 minutes per research memo. 67 active licenses.", "denominator_and_limits": "No timed sample, matched task baseline, exact dates or average review duration. 2–3 times faster implies roughly 50–67% shorter research only if the same task and output are compared. Memo-review minutes do not justify eliminating expert tax applicability approval.", "quantitative_use": "Supports a conditional high research-stage reduction around 65%; exact modeled factor remains an assumption.", "project_ids": [ "AI-06", "AI-07" ], "evidence_type": "Named vendor/customer implementation account", "reported_result": "Research projects reported 2–3 times faster; reviewers save 15–30 minutes per research memo. 67 active licenses.", "limits": "No timed sample, matched task baseline, exact dates or average review duration. 2–3 times faster implies roughly 50–67% shorter research only if the same task and output are compared. Memo-review minutes do not justify eliminating expert tax applicability approval.", "research_status": "retain_with_measurement_limits", "plain_summary": "Research projects reported 2–3 times faster; reviewers save 15–30 minutes per research memo. 67 active licenses.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Fetched primary HTML: w-published-date=2026-03-09T00:00:00.000Z and analytics publishDate=2026-03-09." }, "planning-tr_tax_launch_confirmation_2026": { "id": "planning-tr_tax_launch_confirmation_2026", "url": "https://www.thomsonreuters.com/en-us/posts/innovation/the-real-ai-story-in-tax-and-audit-isnt-adoption-its-impact/", "title": "The Real AI Story in Tax and Audit Isn’t Adoption — It’s Impact", "publication_date": "2026-04-16", "date_verification": "Primary HTML datePublished=2026-04-16T08:00:32+00:00; modified 2026-04-23.", "evidence_period": "States CoCounsel Tax & Audit launched March 2025.", "status": "retain", "kind": "First-party executive article", "result": "Confirms March 2025 product launch; this establishes product-era context for the Copeland Buhl account.", "denominator_and_limits": "Does not calendar-date the Copeland Buhl six-month period or provide a controlled time study.", "quantitative_use": "Date corroboration only.", "project_ids": [ "AI-06", "AI-07" ], "evidence_type": "First-party executive article", "reported_result": "Confirms March 2025 product launch; this establishes product-era context for the Copeland Buhl account.", "limits": "Does not calendar-date the Copeland Buhl six-month period or provide a controlled time study.", "research_status": "retain_as_product_date_confirmation", "plain_summary": "Confirms March 2025 product launch; this establishes product-era context for the Copeland Buhl account.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Primary HTML datePublished=2026-04-16T08:00:32+00:00; modified 2026-04-23." } }, "projects": { "AI-01": { "source_ids": [ "accounting-deloitte_zora", "accounting-pamarco", "accounting-choi_xie_final" ], "bullets": [ { "source_id": "accounting-deloitte_zora", "label": "Deloitte (2026)", "text": "", "url": "https://www.deloitte.com/us/en/what-we-do/case-studies/hands-off-the-task-eyes-on-the-outcome.html" }, { "source_id": "accounting-pamarco", "label": "Pamarco / Microsoft (2026)", "text": "", "url": "https://www.microsoft.com/en/customers/story/27289-pamarco-technologies-dynamics-365-finance" }, { "source_id": "accounting-choi_xie_final", "label": "Choi & Xie (2025 experiment)", "text": "", "url": "https://onlinelibrary.wiley.com/doi/10.1111/1475-679x.70052" } ], "range_status": "analyst_scenario", "range_support_impact": "Recent deployment evidence remains, including large broader-automation outcomes. The 15–55% whole-workflow range is still an analyst scenario, not a bound estimated by these studies. Hold both undated Auditoria cases. Use only the 2025 randomized component of Choi/Xie and preserve its null average completion-time result. Pamarco's projected saving and Asurion's touchless share cannot become human-time percentages." }, "AI-02": { "source_ids": [ "accounting-microsoft_collections_2026_insidetrack", "accounting-microsoft_collections", "accounting-microsoft_finance_2025_ga" ], "bullets": [ { "source_id": "accounting-microsoft_collections_2026_insidetrack", "label": "Microsoft collections (2026)", "text": "", "url": "https://www.microsoft.com/insidetrack/blog/streamlining-finance-cash-collection-at-microsoft-with-ai/" }, { "source_id": "accounting-microsoft_collections", "label": "Microsoft collections case (2026 document)", "text": "", "url": "https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/mcaps/documents/fy26/Collections-Email-Agent.pdf" }, { "source_id": "accounting-microsoft_finance_2025_ga", "label": "Microsoft finance assistant (2025)", "text": "", "url": "https://www.microsoft.com/en-us/dynamics-365/blog/it-professional/2025/10/20/empowering-finance-with-an-ai-assistant-in-microsoft-365-copilot/" } ], "range_status": "analyst_scenario", "range_support_impact": "Recent first-party assisted-collections report gives an up-to-60% handling-time result; exact catalog range remains an analyst scenario. An autonomous collections case confirms broader capabilities but supplies no valid percentage denominator for reviewed replies.", "reestimated_in": "reestimation-2026-10-25/model.json", "prior_source_review_impact": "The 15–40% range loses its historical measured anchors: 2020/2021 GPT-3 support assistance and the 2023 writing experiment. The recent Microsoft case reports annual savings but automates sending and lacks baseline handling hours. The product announcement confirms human-reviewed ERP-grounded replies, with no measured percentage. No recent direct study found here validates either end of the range. Choi's coding experiment should not be presented as a collection-email case." }, "AI-03": { "source_ids": [ "accounting-ted_taylor_2026", "accounting-chatham_2026", "accounting-microsoft_finance_2025_ga", "accounting-afm_controls" ], "bullets": [ { "source_id": "accounting-ted_taylor_2026", "label": "Ted Taylor / OpenAI (2026)", "text": "", "url": "https://openai.com/index/ted-taylor-restaurant/" }, { "source_id": "accounting-chatham_2026", "label": "Chatham / OpenAI (2026)", "text": "", "url": "https://openai.com/index/chatham-financial/" }, { "source_id": "accounting-microsoft_finance_2025_ga", "label": "Microsoft finance assistant (2025)", "text": "", "url": "https://www.microsoft.com/en-us/dynamics-365/blog/it-professional/2025/10/20/empowering-finance-with-an-ai-assistant-in-microsoft-365-copilot/" }, { "source_id": "accounting-afm_controls", "label": "AFM audit guidance (2025)", "text": "", "url": "https://www.afm.nl/~/profmedia/files/rapporten/2025/12-building-blocks-for-controlled-use-of-auditing-tools.pdf" } ], "range_status": "analyst_scenario", "range_support_impact": "New named 2026 cash-reconciliation example reports 180 to five-to-ten minutes; adjacent current-model trade validation reports about thirty to under four. Both are vendor cases with incomplete review and timing methods. They support a potentially large narrow-workflow reduction, not empirical endpoints for the catalog scope.", "reestimated_in": "reestimation-2026-10-25/model.json", "prior_source_review_impact": "The 25–55% modeled range loses its 80–92% narrow-task anchors after excluding the 2024 pilot and holding U.S. Venture's likely late-2024/mixed-period result. Chatham is a recent named narrow transaction-validation example, not bank/ledger reconciliation. Microsoft's October 2025 announcement confirms current reconciliation capability but supplies no auditable labor-time denominator. No clean recent full-reconciliation percentage found; do not manufacture a replacement bound." }, "AI-04": { "source_ids": [ "planning-pwc_cfo_advisor", "planning-natura_sap", "planning-bcg_finance_roi_2025" ], "bullets": [ { "source_id": "planning-pwc_cfo_advisor", "label": "PwC CFO Advisor (2026)", "text": "", "url": "https://cee.pwc.com/case-studies/how-cfo-advisor-enhances-and-automates-financial-analysis-and-storytelling-with-ai.html" }, { "source_id": "planning-natura_sap", "label": "Natura / SAP (2026)", "text": "", "url": "https://news.sap.com/2026/07/natura-co-transforming-finance-generative-ai/" }, { "source_id": "planning-bcg_finance_roi_2025", "label": "BCG finance case (2025)", "text": "", "url": "https://www.bcg.com/publications/2025/how-finance-leaders-can-get-roi-from-ai" } ], "range_status": "analyst_scenario", "range_support_impact": "Conditional scenario informed by a recent variance-analysis case; no measured end-to-end margin-answer result.", "reestimated_in": "reestimation-2026-10-25/model.json", "prior_source_review_impact": "Current 10–35% band needs re-estimation: its 2023 search and drafting anchors are excluded. Recent sources show a live margin project, repeatable variance work and adjacent internal retrieval, but no measured end-to-end margin-answer saving. Do not replace the range with PwC's 90%." }, "AI-05": { "source_ids": [ "planning-docker_pigment", "planning-pwc_cfo_advisor", "planning-clickup_pigment" ], "bullets": [ { "source_id": "planning-docker_pigment", "label": "Docker / Pigment (2026)", "text": "", "url": "https://www.pigment.com/customer-stories/docker" }, { "source_id": "planning-pwc_cfo_advisor", "label": "PwC CFO Advisor (2026)", "text": "", "url": "https://cee.pwc.com/case-studies/how-cfo-advisor-enhances-and-automates-financial-analysis-and-storytelling-with-ai.html" }, { "source_id": "planning-clickup_pigment", "label": "ClickUp / Pigment (2026)", "text": "", "url": "https://www.pigment.com/customer-stories/clickup" } ], "range_status": "analyst_scenario", "range_support_impact": "Recent named customer self-estimate closely matches commentary and review; the complete pack percentage still depends on an assumed task mix.", "reestimated_in": "reestimation-2026-10-25/model.json", "prior_source_review_impact": "Current 20–45% band needs re-estimation because the 2023 writing experiment and Jagged Frontier are excluded. PwC and ClickUp remain useful task-level references, with unknown measurement calendars. Neither measures the same complete 10-page pack with sign-off." }, "AI-06": { "source_ids": [ "planning-fm_knowledge_2026", "planning-amazon_regulatory_rag_2026", "planning-tr_copeland_buhl_2026" ], "bullets": [ { "source_id": "planning-fm_knowledge_2026", "label": "FM / Microsoft (2026)", "text": "", "url": "https://www.microsoft.com/en/customers/story/26503-fm-azure" }, { "source_id": "planning-amazon_regulatory_rag_2026", "label": "Amazon Finance / AWS (2026)", "text": "", "url": "https://aws.amazon.com/blogs/machine-learning/how-amazon-finance-streamlines-regulatory-inquiries-by-using-generative-ai-on-aws/" }, { "source_id": "planning-tr_copeland_buhl_2026", "label": "Copeland Buhl / Thomson Reuters (2026)", "text": "", "url": "https://tax.thomsonreuters.com/en/insights/case-studies/from-hours-to-minutes-how-cocounsel-reimagined-tax-workflow-efficiency-at-copeland-buhl" } ], "range_status": "analyst_scenario", "range_support_impact": "A 2026 knowledge-search case reports saved minutes without a baseline duration. Reference lookup duration, task mix, coverage and answer-success rates are explicit assumptions; no complete finance-knowledge workflow is measured.", "reestimated_in": "reestimation-2026-10-25/model.json", "prior_source_review_impact": "Current 15–35% range needs re-estimation: its main 2023 Microsoft search and Noy writing anchors are excluded. BCG and Hoppy offer much larger task-level retrieval examples, with measurement calendars unstated; FM reports minutes saved without a baseline duration. The user's broader knowledge category includes metrics, datasets, policies and precedents, while the numeric model remains policy-only. Reconcile scope and estimate a representative question mix. No evidence supports retaining 35% as a general cap, and these cases do not justify an invented replacement band." }, "AI-07": { "source_ids": [ "planning-tr_cocounsel_corporate", "planning-tr_copeland_buhl_2026", "planning-amazon_regulatory_rag_2026", "planning-tax_plat" ], "bullets": [ { "source_id": "planning-tr_cocounsel_corporate", "label": "Thomson Reuters (2025 survey)", "text": "", "url": "https://tax.thomsonreuters.com/en/products/cocounsel-tax/corp" }, { "source_id": "planning-tr_copeland_buhl_2026", "label": "Copeland Buhl / Thomson Reuters (2026)", "text": "", "url": "https://tax.thomsonreuters.com/en/insights/case-studies/from-hours-to-minutes-how-cocounsel-reimagined-tax-workflow-efficiency-at-copeland-buhl" }, { "source_id": "planning-amazon_regulatory_rag_2026", "label": "Amazon Finance / AWS (2026)", "text": "", "url": "https://aws.amazon.com/blogs/machine-learning/how-amazon-finance-streamlines-regulatory-inquiries-by-using-generative-ai-on-aws/" }, { "source_id": "planning-tax_plat", "label": "PLAT tax benchmark (2026)", "text": "", "url": "https://arxiv.org/abs/2503.03444" } ], "range_status": "analyst_scenario", "range_support_impact": "Late-2025 self-reported tax-research survey plus named 2026 tax-research implementation account; no measured end-to-end tax monitoring labor result. Conditional estimate.", "reestimated_in": "reestimation-2026-10-25/model.json", "prior_source_review_impact": "Current 10–35% band needs re-estimation: Noy drafting evidence is excluded and the undated World Wide Watch case is held. Late-2025 self-reported tax-research savings remain, but do not calibrate scanning, no-change sources, expert applicability review and corrections for the full monitoring workflow." }, "AI-08": { "source_ids": [ "accounting-microsoft_sox", "accounting-provectus_johnson_lambert", "accounting-afm_controls" ], "bullets": [ { "source_id": "accounting-microsoft_sox", "label": "Microsoft SOX case (file updated 2026)", "text": "", "url": "https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/mcaps/documents/fy26/SOX-Change-Management-Testing-Agent.pdf" }, { "source_id": "accounting-provectus_johnson_lambert", "label": "Johnson Lambert / Provectus (2026)", "text": "", "url": "https://www.provectus.com/case-studies/johnson-lambert-audit/" }, { "source_id": "accounting-afm_controls", "label": "AFM audit guidance (2025)", "text": "", "url": "https://www.afm.nl/~/profmedia/files/rapporten/2025/12-building-blocks-for-controlled-use-of-auditing-tools.pdf" } ], "range_status": "analyst_scenario", "range_support_impact": "Direct SOX extraction/annotation case supplies annual hours without a denominator; recent human-reviewed audit-document case reports fifty percent less processing time. Exact ticket-workflow endpoints remain conditional analyst assumptions.", "reestimated_in": "reestimation-2026-10-25/model.json", "prior_source_review_impact": "The direct Microsoft SOX case is now freshly verified and dated through 2026 file metadata, but 850 annual hours has no baseline denominator. Provectus is verifiably a 2026 publication with post-2024 model names, while its document-processing outcome is adjacent and self-reported. Remove the 2023 writing anchor. The 20–35% whole-ticket-workflow range remains uncalibrated; neither recent case measures that scope at a fixed sample and review standard." }, "AI-09": { "source_ids": [ "ir-procurement-q4-forrester-2025", "ir-procurement-murphy-q4", "ir-procurement-q4-workflow-2026" ], "bullets": [ { "source_id": "ir-procurement-q4-forrester-2025", "label": "Forrester / Q4 (2025)", "text": "", "url": "https://tei.forrester.com/go/Q4/platform/index.html" }, { "source_id": "ir-procurement-murphy-q4", "label": "Murphy USA / Q4 (2025)", "text": "", "url": "https://www.q4inc.com/resource-center/blog/how-murphy-usa-unlocked-their-full-potential-with-q4" }, { "source_id": "ir-procurement-q4-workflow-2026", "label": "Q4 workflow guide (2026)", "text": "", "url": "https://www.q4inc.com/resource-center/blog/can-ai-actually-handle-earnings-season-a-workflow-test-for-ir-teams?hs_amp=true" } ], "range_status": "analyst_scenario", "range_support_impact": "The 25-55% workflow band remains an analyst work-mix transfer. Its strongest retained numerical evidence is a sponsored single-customer model of drafting and summarization. Source publication years are eligible, but model and measurement dates remain undisclosed. No defensible new range follows from this audit alone." }, "AI-10": { "source_ids": [ "ir-procurement-bcg-procurement-2025", "ir-procurement-gatx-levelpath-2025-july", "ir-procurement-bid-scoring-volatility-2026" ], "bullets": [ { "source_id": "ir-procurement-bcg-procurement-2025", "label": "BCG procurement cases (2025)", "text": "", "url": "https://media-publications.bcg.com/BCG-Executive-Perspectives-Future-of-Procurement-with-AI-2025-27Feb2025.pdf" }, { "source_id": "ir-procurement-gatx-levelpath-2025-july", "label": "GATX / Levelpath (2025)", "text": "", "url": "https://www.levelpath.com/blog/how-gatx-is-scaling-procurement-with-an-ai-first-strategy" }, { "source_id": "ir-procurement-bid-scoring-volatility-2026", "label": "Supplier-scoring study (2026)", "text": "", "url": "https://onlinelibrary.wiley.com/doi/10.1111/jbl.70072" } ], "range_status": "analyst_scenario", "range_support_impact": "The 15-35% band remains an analyst work-mix transfer from BCG task-level case results, whose 2025 publication is eligible but measurement/model dates are unknown. GATX and newer studies corroborate task/capacity potential without a compatible complete-event denominator. This audit does not independently revalidate either endpoint or justify a new range." }, "AI-11": { "source_ids": [ "ir-procurement-aws-finance-foundation", "ir-procurement-microsoft-controls-2025" ], "bullets": [ { "source_id": "ir-procurement-aws-finance-foundation", "label": "Amazon Finance / AWS (2026)", "text": "", "url": "https://aws.amazon.com/blogs/machine-learning/how-amazon-finance-streamlines-regulatory-inquiries-by-using-generative-ai-on-aws/" }, { "source_id": "ir-procurement-microsoft-controls-2025", "label": "Microsoft enterprise controls (2025)", "text": "", "url": "https://techcommunity.microsoft.com/blog/microsoft-security-blog/enterprise-grade-controls-for-ai-apps-and-agents-built-with-azure-ai-foundry-and/4414757" } ], "range_status": "indirect", "range_support_impact": "Keep Indirect. Retained sources describe architecture and work; none measures saved implementation labor across comparable deployments. Application workflow savings should remain on application rows." } } }