{ "reviewed_on": "2026-10-25", "metric": "Percentage of baseline human operating time saved within the defined workflow", "range_status": "Conditional planning scenarios, not measured population bounds or confidence intervals.", "reference_baseline": "The same work is completed manually using existing systems and approved inputs. Sources and calculation models are usable. These are workflow reference cases, not company types.", "conditions": [ "The workflow is in steady-state use and passes checks for equivalent output quality.", "Volume, coverage and output remain unchanged. Saved capacity is not assumed to reduce staffing.", "Human review, corrections, exceptions and existing approvals are included.", "Baseline task shares and transferred task-saving rates are analyst assumptions, not measured industry allocations." ], "technology_boundary": "Savings include conventional extraction, matching, retrieval and integration delivered with each scoped project. The separate marginal contribution of GenAI is not established.", "exclusions": [ "Implementation and training", "Separate project/platform upkeep", "Upstream remediation outside the project", "Additional output and newly worthwhile work" ], "upkeep_formula": "Monthly operating hours saved = baseline monthly workflow hours × saving percentage. Net capacity released = operating hours saved − annual project upkeep / 12 − allocated shared upkeep / 12. Shared upkeep is allocated once across the portfolio.", "outside_range": "Already-automated work, low adoption, difficult exceptions or unreliable output can produce smaller savings, zero or additional work. The positive lower endpoint is not a guaranteed minimum.", "calculation": "Normalize manual work to 100 human-hours. The seven revised projects save 100 minus the sum of remaining human-hour budgets, including incremental AI correction rows with zero baseline hours. Three earlier scenarios retain their weighted task-rate calculation. Round endpoints outward to five percentage points.", "sensitivity": "Revised projects show explicit alternatives for work coverage or near-automatic preparation. Only the three earlier scenarios retain the illustrative extra-15-hour review-share sensitivity. No sensitivity is an empirical performance bound.", "review_revisions": [ "Seven flagged ranges restored from reviewed before/after scenarios with sourced results separate from assumptions.", "Audit reference range 25–35%; the unvalidated near-automatic 65% scenario is shown separately.", "Tax research averages are applied once, without a second coverage/success discount.", "Amazon discovery result held out because its model is from 2024 and timing boundaries conflict.", "Existing review and incremental AI corrections are counted separately; shared upkeep remains outside the workflow percentage." ], "sources": { "accounting-deloitte_zora": { "title": "Deloitte: Hands off the task, eyes on the outcome", "url": "https://www.deloitte.com/us/en/what-we-do/case-studies/hands-off-the-task-eyes-on-the-outcome.html", "type": "Consultancy-authored anonymous customer deployment", "published": "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.", "finding": "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.", "measurement": "", "limits": [], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "accounting-pamarco": { "title": "Microsoft customer story: Pamarco Technologies", "url": "https://www.microsoft.com/en/customers/story/27289-pamarco-technologies-dynamics-365-finance", "type": "Vendor-authored named customer deployment", "published": "2026-09-18", "evidence_period": "Deployment and measurement dates not stated; 2026 case describes current ERP-connected Copilot Studio/MCP use.", "finding": "Pamarco reports about 95% success coding freight invoices. It estimates much larger time savings after full adoption, including automatic posting.", "measurement": "", "limits": [], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "accounting-choi_xie_final": { "title": "Choi and Xie: Human + AI in Accounting — 2025 framed experiment only", "url": "https://onlinelibrary.wiley.com/doi/10.1111/1475-679x.70052", "type": "Peer-reviewed randomized framed experiment", "published": "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.", "finding": "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.", "measurement": "", "limits": [], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "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", "type": "First-party internal deployment case", "published": "Document created 2026-05-08; public by 2026-05-11 according to completed source review", "evidence_period": "Measurement dates and model version unstated.", "finding": "About 60,000 incoming emails/month, about 55% resolved automatically and 36,000 annual hours saved.", "measurement": "", "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." ], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "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/", "type": "Product announcement", "published": "2025-10-20", "evidence_period": "Product availability in October 2025; unnamed pilots have unstated measurement periods.", "finding": "ERP-grounded draft customer replies for staff review and Excel reconciliation matching/exception assistance.", "measurement": "", "limits": [ "Days-to-hours pilot wording has no auditable human-time denominator. Supplies no task percentage or endpoint." ], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "accounting-chatham_2026": { "title": "OpenAI: Chatham scales its capital markets expertise with OpenAI", "url": "https://openai.com/index/chatham-financial/", "type": "Vendor-authored named customer early measurement", "published": "2026-10-02", "evidence_period": "Early measurement reported October 2026; exact dates unstated; GPT-5.6 generation explicitly identified.", "finding": "Trade-validation review reportedly falls from about 30 minutes to under four; evidence is assembled, terms compared and discrepancies flagged.", "measurement": "", "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." ], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "accounting-provectus_johnson_lambert": { "title": "Provectus: Johnson Lambert reinvents insurance auditing with generative AI", "url": "https://www.provectus.com/case-studies/johnson-lambert-audit/", "type": "Vendor-authored named customer deployment", "published": "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.", "finding": "50% lower document-processing time; auditors inspect extracted tables against originals and correct values.", "measurement": "", "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." ], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "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", "type": "First-party internal deployment case", "published": "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.", "finding": "850 annual hours saved. Two agents extract ticket attributes, prepopulate rule checks and mark the matching source evidence.", "measurement": "", "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." ], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "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", "type": "Regulator guidance", "published": "2025-12", "evidence_period": "2025 supervisory guidance; older firm-selection data are unused.", "finding": "No productivity result. Requires reliable input, verifiable outputs and continuing auditor responsibility.", "measurement": "", "limits": [ "No numerical effect; none used." ], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "planning-natura_sap": { "title": "SAP: Natura gross-margin application", "url": "https://news.sap.com/2026/07/natura-co-transforming-finance-generative-ai/", "type": "Named vendor/customer production case", "published": "2026-07-08", "evidence_period": "Production since August 2025; use in Ecuador described July 2026. Originating early-2024 prototype is excluded from quantitative support.", "finding": "ERP-connected gross-margin revenue/cost explanations with human interpretation and decision control; no numeric human-time saving.", "measurement": "", "limits": [ "Six months describes implementation. No labor baseline or saving is reported." ], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "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", "type": "Consultancy-authored anonymous TMT customer case", "published": "2026-04-01", "evidence_period": "Deployment and observation calendar unstated; operating implementation described in the 2026 case.", "finding": "About 90% less variance-analysis effort across approximately 20 configured analysis types, with controllers and FP&A editing and approving drafts.", "measurement": "", "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." ], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "planning-clickup_pigment": { "title": "Pigment: ClickUp Analyst Agent case", "url": "https://www.pigment.com/customer-stories/clickup", "type": "Named vendor/customer deployment case", "published": "2026-09-03", "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.", "finding": "Reported 1-1.5 days saved per person monthly across six finance staff on first-draft budget-variance reports and driver identification.", "measurement": "", "limits": [ "No baseline reporting hours, hours/day or timing protocol. Separate Modeler Agent claims are excluded." ], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "planning-amazon_regulatory_rag_2026": { "title": "How Amazon Finance streamlines regulatory inquiries by using generative AI on AWS", "url": "https://aws.amazon.com/blogs/machine-learning/how-amazon-finance-streamlines-regulatory-inquiries-by-using-generative-ai-on-aws/", "type": "First-party technical deployment account", "published": "2026-05-12", "evidence_period": "May 2026 account names Claude Sonnet 4.5; exact go-live/observation dates unstated.", "finding": "Finds historical documents and precedents and supports regulatory-response preparation. Parallel retrieval falls from 10 seconds to under 2 seconds.", "measurement": "", "limits": [ "Retrieval number compares system configurations; no human-hours result. Regulatory inquiries differ from ongoing tax monitoring." ], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "planning-fm_knowledge_2026": { "title": "FM empowers 1,500+ engineers with trusted, AI-assisted access to engineering knowledge using Azure OpenAI", "url": "https://www.microsoft.com/en/customers/story/26503-fm-azure", "type": "Vendor/customer estimate; adjacent engineering knowledge search", "published": "2026-05-28", "evidence_period": "First two production months, without calendar dates. Model version and timing protocol unstated.", "finding": "Estimated 6–10 minutes saved per search; 17,000 queries in the first two production months across more than 1,500 engineers.", "measurement": "", "limits": [ "No baseline duration or full checking-time boundary. Estimated minutes cannot be converted to a percentage." ], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "planning-tr_cocounsel_corporate": { "title": "CoCounsel Tax: The AI tax assistant for in-house teams", "url": "https://tax.thomsonreuters.com/en/products/cocounsel-tax/corp", "type": "Vendor user survey", "published": "2026-06-29", "evidence_period": "Self-reported user survey November–December 2025.", "finding": "Users report an average 32% less tax-research time.", "measurement": "", "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." ], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "planning-tax_plat": { "title": "Taxation Perspectives from Large Language Models: A Case Study on Additional Tax Penalties", "url": "https://arxiv.org/abs/2503.03444", "type": "Tax reasoning benchmark", "published": "2025-03-05; revised 2026-02-08", "evidence_period": "Recent 2025 model snapshots in revised benchmark; exact runs unstated.", "finding": "Continuing difficulties applying tax law to facts and conclusions.", "measurement": "", "limits": [ "300 Korean tax-penalty questions; no human-time study." ], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "ir-procurement-q4-forrester-2025": { "title": "The Total Economic Impact of Q4 Platform", "url": "https://tei.forrester.com/go/Q4/platform/index.html", "type": "Sponsored single-customer interview and financial model", "published": "2025-06", "evidence_period": "Interview, deployment and measurement dates undisclosed; three-year financial projection, not three years of observed results. Model version undisclosed.", "finding": "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.", "measurement": "", "limits": [], "use": "Task-specific reported inputs; no directly observed 25-55% complete-workflow range.", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "ir-procurement-murphy-q4": { "title": "From Manual to Strategic: Murphy USA's IR Transformation", "url": "https://www.q4inc.com/resource-center/blog/how-murphy-usa-unlocked-their-full-potential-with-q4", "type": "Named vendor customer case", "published": "2025-04-30", "evidence_period": "Deployment, measurement dates and model version undisclosed.", "finding": "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.", "measurement": "", "limits": [], "use": "Reported platform benefit, with no compatible workflow denominator.", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "ir-procurement-q4-workflow-2026": { "title": "Can AI Actually Handle Earnings Season? A Workflow Test for IR Teams", "url": "https://www.q4inc.com/resource-center/blog/can-ai-actually-handle-earnings-season-a-workflow-test-for-ir-teams?hs_amp=true", "type": "Vendor workflow guidance", "published": "2026-08-14", "evidence_period": "2026 product/workflow guidance; no experiment or deployment measurement.", "finding": "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.", "measurement": "", "limits": [], "use": "None.", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "ir-procurement-bcg-procurement-2025": { "title": "Maximizing Value Potential from AI in 2025: Procurement", "url": "https://media-publications.bcg.com/BCG-Executive-Perspectives-Future-of-Procurement-with-AI-2025-27Feb2025.pdf", "type": "Anonymous consulting delivery cases", "published": "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.", "finding": "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.", "measurement": "", "limits": [], "use": "Task-specific case results inform the existing analyst transfer, but do not directly measure the 15-35% complete-event range.", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "ir-procurement-gatx-levelpath-2025-july": { "title": "How GATX is Scaling Procurement with an AI-First Strategy", "url": "https://www.levelpath.com/blog/how-gatx-is-scaling-procurement-with-an-ai-first-strategy", "type": "Vendor recap of named customer webinar", "published": "2025-07-21", "evidence_period": "Reports a recent webinar with GATX's procurement director. Deployment and measurement dates and model version undisclosed.", "finding": "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.", "measurement": "", "limits": [], "use": "Capacity and subtask speed evidence; does not establish a percentage of total event labor saved.", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "ir-procurement-bid-scoring-volatility-2026": { "title": "Do Humans and GAI See Eye to Eye? Implications of LLM Scoring Volatility in Supplier Evaluations", "url": "https://onlinelibrary.wiley.com/doi/10.1111/jbl.70072", "type": "Peer-reviewed controlled supplier-scoring benchmark", "published": "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.", "finding": "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.", "measurement": "", "limits": [], "use": "No time-saving measure.", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "ir-procurement-aws-finance-foundation": { "title": "How Amazon Finance streamlines regulatory inquiries by using generative AI on AWS", "url": "https://aws.amazon.com/blogs/machine-learning/how-amazon-finance-streamlines-regulatory-inquiries-by-using-generative-ai-on-aws/", "type": "First-party finance deployment architecture", "published": "2026-05-12", "evidence_period": "2026 architecture/deployment account; implementation and measurement periods unreported.", "finding": "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.", "measurement": "", "limits": [], "use": "No directly measured foundation labor or reuse percentage.", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "ir-procurement-microsoft-controls-2025": { "title": "Enterprise-grade controls for AI apps and agents built with Azure AI Foundry and Copilot Studio", "url": "https://techcommunity.microsoft.com/blog/microsoft-security-blog/enterprise-grade-controls-for-ai-apps-and-agents-built-with-azure-ai-foundry-and/4414757", "type": "Product/security capability announcement", "published": "2025-05-19", "evidence_period": "2025 product capability announcement; no deployment or labor experiment.", "finding": "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.", "measurement": "", "limits": [], "use": "None.", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "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/", "type": "First-party internal deployment report", "published": "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.", "finding": "Up to 60% lower inquiry handling time using inline suggestions, call summaries and automatically drafted replies.", "measurement": "", "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." ], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "accounting-ted_taylor_2026": { "title": "OpenAI: Franchisee reconciles cash in 94% less time with ChatGPT Work", "url": "https://openai.com/index/ted-taylor-restaurant/", "type": "Vendor-authored named customer case", "published": "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.", "finding": "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.", "measurement": "", "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." ], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "planning-docker_pigment": { "title": "Pigment: Docker monthly close commentary and review", "url": "https://www.pigment.com/customer-stories/docker", "type": "Named vendor/customer deployment; customer self-estimate", "published": "2026-09-11", "evidence_period": "Current monthly operating use in a 2026 case; exact measurement months unstated.", "finding": "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.", "measurement": "", "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." ], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "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", "type": "Consultancy-authored anonymous customer transformation", "published": "2025-06-04", "evidence_period": "Consumer-goods deployment timing unstated; described in the 2025 article.", "finding": "50% shorter report-generation time after constructing a driver-tree model, integrating data and adding a GenAI query interface.", "measurement": "", "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." ], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" }, "planning-tr_copeland_buhl_2026": { "title": "From hours to minutes: How CoCounsel reimagined tax workflow efficiency at Copeland Buhl", "url": "https://tax.thomsonreuters.com/en/insights/case-studies/from-hours-to-minutes-how-cocounsel-reimagined-tax-workflow-efficiency-at-copeland-buhl", "type": "Named vendor/customer implementation account", "published": "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.", "finding": "Research projects reported 2–3 times faster; reviewers save 15–30 minutes per research memo. 67 active licenses.", "measurement": "", "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." ], "use": "", "anchors": [], "eligibility": "retain", "checked": "2026-10-25" } }, "projects": { "AI-01": { "title": "Invoice processing", "range_percent": [ 15, 55 ], "baseline": "Intake an invoice, extract/rekey its details, find the ERP supplier, choose coding, prepare a draft bill, resolve exceptions and review it before approval. Material manual entry and draft preparation remain in the current process.", "evidence_strength": "Analyst work-share scenario. Recent references remain, but none establishes these whole-workflow endpoints.", "blocks": [ { "work": "Invoice intake and extraction", "baseline_hours": 35, "reduction_percent": [ 50, 80 ], "basis": "Analyst-selected baseline share and task rates; combined document extraction can remove much rekeying, but leave format exceptions and source checking.", "source_ids": [ "accounting-deloitte_zora" ] }, { "work": "Supplier lookup and accounting coding", "baseline_hours": 25, "reduction_percent": [ 0, 60 ], "basis": "Zero low-case coding saving reflects the randomized null time result. High case assumes stable coding rules and repetitive suppliers; 95% success and projected autoposting gains are not imported.", "source_ids": [ "accounting-choi_xie_final", "accounting-pamarco" ] }, { "work": "Prepare the ERP draft bill", "baseline_hours": 15, "reduction_percent": [ 50, 80 ], "basis": "Draft assembly removes duplicate typing; rates are transfer assumptions, not published stage measurements.", "source_ids": [ "accounting-deloitte_zora" ] }, { "work": "Check drafts, correct exceptions and approve", "baseline_hours": 25, "reduction_percent": [ -30, 0 ], "basis": "Approval remains. Low scenario adds 7.5 checking hours per 100 baseline hours, allowing review/correction to grow; high scenario only holds existing checking time flat.", "source_ids": [ "accounting-choi_xie_final" ] } ], "source_ids": [ "accounting-deloitte_zora", "accounting-pamarco", "accounting-choi_xie_final" ], "assumptions": [], "lower_case": "", "upper_case": "", "reasoning": "The manual extraction/draft-bill scope can benefit even when coding suggestions save no time. Full-autonomy AP reports around 50% staffing or 50–75% processing-time gains, but all-invoice human review, null coding-time evidence and unresolved exceptions warrant a substantially lower planning floor.", "outside_band": [ "Coding-only assistance, an already touchless intake/draft workflow or a failed deployment can save zero or add time. No positive lower bound is established for those situations.", "A fully autonomous posting/payment deployment has a different scope and risk profile." ], "uncertainties": [], "validation_needed": [ "Time a representative invoice sample before/after with intake, coding, draft preparation, initial review and rework separately recorded; retain the same invoice mix and quality standard." ], "research_file": "source-review-2026-10-25/review.json", "review_block_index": 3, "estimate_status": "analyst_scenario", "reference_review": "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.", "calculated_percent": [ 17.5, 55.0 ], "remaining_hours_per_100": [ 45.0, 82.5 ], "review_sensitivity": { "review_before": 25, "review_after": 40, "calculated_percent": [ 8.0, 44.0 ], "range_percent": [ 5, 45 ] } }, "AI-02": { "title": "Collections", "range_percent": [ 25, 60 ], "baseline": "Five supported inquiry types from intake and account lookup through reply drafting, factual verification, correction and approved dispatch. Dispute negotiation and wider receivables management are excluded. The same inbound mix is compared at both endpoints.", "evidence_strength": "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.", "blocks": [ { "work": "Read, classify and route incoming inquiries", "baseline_hours": 10, "reduction_percent": [ 20.0, 80.0 ], "after_hours": [ 8, 2 ], "basis": "Lower: staff still read most messages and correct routing. Upper: intent and routing are prepared automatically; staff use the review step below to read the request and confirm it belongs to the supported customer/type.", "source_ids": [ "accounting-microsoft_collections_2026_insidetrack", "accounting-microsoft_finance_2025_ga" ] }, { "work": "Retrieve account records, invoice detail and supporting evidence", "baseline_hours": 30, "reduction_percent": [ 33.33333333, 83.33333333 ], "after_hours": [ 20, 5 ], "basis": "Lower: one-third of manual lookup effort disappears but ambiguous identifiers and missing records still require searches. Upper: correctly linked account/invoice evidence is presented together for most cases; five hours remain for unresolved identifiers, stale sources and manual fallback.", "source_ids": [ "accounting-microsoft_collections_2026_insidetrack", "accounting-microsoft_finance_2025_ga" ] }, { "work": "Prepare reply text and perform preparer edits", "baseline_hours": 30, "reduction_percent": [ 66.66666667, 90.0 ], "after_hours": [ 10, 3 ], "basis": "Lower: approved templates supply first drafts but material edits remain. Upper: repetitive supported replies are populated from verified records and approved language; three hours cover preparer edits and rejected drafts. Reading and factual approval are counted in the next block.", "source_ids": [ "accounting-microsoft_collections_2026_insidetrack", "accounting-microsoft_finance_2025_ga" ] }, { "work": "Verify customer/account facts, correct errors, approve and dispatch every reply", "baseline_hours": 30, "reduction_percent": [ -23.33333333, 0.0 ], "after_hours": [ 37, 30 ], "basis": "Every outgoing reply retains human approval. Lower adds seven hours for checking unfamiliar drafts and repairing errors; upper retains the full existing thirty-hour verification/dispatch budget. No gain from autonomous sending is counted.", "source_ids": [ "accounting-microsoft_collections_2026_insidetrack", "accounting-microsoft_finance_2025_ga" ] } ], "source_ids": [ "accounting-microsoft_collections_2026_insidetrack", "accounting-microsoft_collections", "accounting-microsoft_finance_2025_ga" ], "assumptions": [ "Baseline task hours, eligible coverage and remaining human-work budgets are scenario assumptions.", "Published task results inform the scenario; no source establishes the complete workflow range.", "Same work volume and required output; manual fallback and corrections are included. Separate upkeep is excluded." ], "lower_case": "Substantial manual lookup and typing in the starting process Read-only current account/invoice evidence joined by reliable customer identifiers Narrow approved reply policies and repetitive supported inquiry types Per-reply human identity, facts, disclosure and dispatch check All repair and fallback time included within the four budgets Connected data and draft templates remove 32 preparation hours, while correction and approval add seven: 100 - (8 + 20 + 10 + 37) = 25 hours saved.", "upper_case": "Substantial manual lookup and typing in the starting process Read-only current account/invoice evidence joined by reliable customer identifiers Narrow approved reply policies and repetitive supported inquiry types Per-reply human identity, facts, disclosure and dispatch check All repair and fallback time included within the four budgets Reliable customer links and standardized replies reduce preparation from 70 to ten hours; every response still gets the original thirty-hour approval/dispatch allowance: 100 - (2 + 5 + 3 + 30) = 60 hours saved.", "reasoning": "Re-estimated from retained references with explicit before/after human-hour budgets.", "outside_band": [], "uncertainties": [ "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." ], "validation_needed": [ "Time the same five inquiry types before and after, tracking collector and approver labor separately", "Stratify by identity ambiguity, record completeness and draft acceptance; record every fallback and correction", "Require unchanged factual accuracy, disclosure controls and resolution quality before treating the high timing budget as achieved" ], "research_file": "reestimation-2026-10-25/model.json", "estimate_status": "analyst_scenario", "reference_review": "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.", "calculated_percent": [ 25, 60 ], "scenario_method": "before_after_hours", "scenarios": { "constrained": { "conditions": [ "Substantial manual lookup and typing in the starting process", "Read-only current account/invoice evidence joined by reliable customer identifiers", "Narrow approved reply policies and repetitive supported inquiry types", "Per-reply human identity, facts, disclosure and dispatch check", "All repair and fallback time included within the four budgets", "Connected data and draft templates remove 32 preparation hours, while correction and approval add seven: 100 - (8 + 20 + 10 + 37) = 25 hours saved." ], "remaining_hours": 75, "saved_hours_per_100": 25 }, "fast": { "conditions": [ "Substantial manual lookup and typing in the starting process", "Read-only current account/invoice evidence joined by reliable customer identifiers", "Narrow approved reply policies and repetitive supported inquiry types", "Per-reply human identity, facts, disclosure and dispatch check", "All repair and fallback time included within the four budgets", "Reliable customer links and standardized replies reduce preparation from 70 to ten hours; every response still gets the original thirty-hour approval/dispatch allowance: 100 - (2 + 5 + 3 + 30) = 60 hours saved." ], "remaining_hours": 40, "saved_hours_per_100": 60 } }, "previous_range_percent": [ 15, 40 ], "remaining_hours_per_100": [ 40, 75 ] }, "AI-03": { "title": "Account reconciliation", "range_percent": [ 35, 75 ], "baseline": "Prepare and align two datasets, apply three agreed matching rules, investigate up to ten exception types, produce source-linked notes and verify balances/conclusions. No autonomous journal posting. Savings include configured deterministic matching delivered in this project.", "evidence_strength": "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.", "blocks": [ { "work": "Prepare and align datasets; check completeness and totals", "baseline_hours": 20, "reduction_percent": [ 25.0, 85.0 ], "after_hours": [ 15, 3 ], "basis": "Lower: exports and mapping still need manual repair. Upper: stable scheduled exports and reusable mappings remove assembly; three hours remain for completeness, control totals and feed exceptions.", "source_ids": [ "accounting-ted_taylor_2026", "accounting-chatham_2026", "accounting-microsoft_finance_2025_ga" ] }, { "work": "Match records and identify genuine differences", "baseline_hours": 35, "reduction_percent": [ 71.42857143, 97.14285714 ], "after_hours": [ 10, 1 ], "basis": "Rules execute over the entire population. Lower retains ten hours to resolve ambiguous candidates and inspect unmatched records. Upper retains one hour to initiate/check the matching result; no one manually compares already matched rows. This is mainly deterministic automation, not an attributed LLM reasoning effect.", "source_ids": [ "accounting-ted_taylor_2026", "accounting-chatham_2026", "accounting-microsoft_finance_2025_ga" ] }, { "work": "Investigate exceptions and draft source-linked notes", "baseline_hours": 25, "reduction_percent": [ 20.0, 76.0 ], "after_hours": [ 20, 6 ], "basis": "Lower: most exceptions need external follow-up and AI mainly organizes records. Upper: a small recurring set of timing, reference and split-payment differences is resolved from the two connected datasets; staff check proposed causes and handle the remaining genuinely unresolved cases. Twenty-five to six is an assumed routine-exception budget, not a measured published stage result.", "source_ids": [ "accounting-ted_taylor_2026", "accounting-chatham_2026", "accounting-microsoft_finance_2025_ga" ] }, { "work": "Independently verify balances, evidence and resolution decisions", "baseline_hours": 20, "reduction_percent": [ 0.0, 25.0 ], "after_hours": [ 20, 15 ], "basis": "The same reviewer controls and final decisions remain. Upper saves five hours from having matched items, control totals and source-linked evidence organized together; professional judgment and unresolved differences still require people.", "source_ids": [ "accounting-afm_controls" ] } ], "source_ids": [ "accounting-ted_taylor_2026", "accounting-chatham_2026", "accounting-microsoft_finance_2025_ga", "accounting-afm_controls" ], "assumptions": [ "Baseline task hours, eligible coverage and remaining human-work budgets are scenario assumptions.", "Published task results inform the scenario; no source establishes the complete workflow range.", "Same work volume and required output; manual fallback and corrections are included. Separate upkeep is excluded." ], "lower_case": "Starting reconciliation still includes significant manual matching Two reliable datasets and three reusable, deterministic matching rules Upper case has routine exception types with causes traceable within connected records Unchanged balances, materiality, evidence requirements and reviewer sign-off Every manual repair, unresolved exception and reviewer correction counted Preparation fifteen + matching ten + investigation twenty + independent verification twenty = sixty-five human-hours, a thirty-five-hour saving.", "upper_case": "Starting reconciliation still includes significant manual matching Two reliable datasets and three reusable, deterministic matching rules Upper case has routine exception types with causes traceable within connected records Unchanged balances, materiality, evidence requirements and reviewer sign-off Every manual repair, unresolved exception and reviewer correction counted Stable feeds, automatic rule matching and mostly recurring supported exceptions leave three + one + six + fifteen = twenty-five human-hours, a seventy-five-hour saving.", "reasoning": "Re-estimated from retained references with explicit before/after human-hour budgets.", "outside_band": [], "uncertainties": [ "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." ], "validation_needed": [ "Use two comparable historical periods, then a shadow production run with the same rules and exception severity", "Measure preparation, matching oversight, investigation and reviewer labor separately", "Count all falsely matched, missed, duplicated and unsupported-explanation cases; time reviewer re-performance and manual fallback", "Replace assumed task shares with observed time and rerun the exact arithmetic" ], "research_file": "reestimation-2026-10-25/model.json", "estimate_status": "analyst_scenario", "reference_review": "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.", "calculated_percent": [ 35, 75 ], "scenario_method": "before_after_hours", "scenarios": { "constrained": { "conditions": [ "Starting reconciliation still includes significant manual matching", "Two reliable datasets and three reusable, deterministic matching rules", "Upper case has routine exception types with causes traceable within connected records", "Unchanged balances, materiality, evidence requirements and reviewer sign-off", "Every manual repair, unresolved exception and reviewer correction counted", "Preparation fifteen + matching ten + investigation twenty + independent verification twenty = sixty-five human-hours, a thirty-five-hour saving." ], "remaining_hours": 65, "saved_hours_per_100": 35 }, "fast": { "conditions": [ "Starting reconciliation still includes significant manual matching", "Two reliable datasets and three reusable, deterministic matching rules", "Upper case has routine exception types with causes traceable within connected records", "Unchanged balances, materiality, evidence requirements and reviewer sign-off", "Every manual repair, unresolved exception and reviewer correction counted", "Stable feeds, automatic rule matching and mostly recurring supported exceptions leave three + one + six + fifteen = twenty-five human-hours, a seventy-five-hour saving." ], "remaining_hours": 25, "saved_hours_per_100": 75 } }, "previous_range_percent": [ 25, 55 ], "remaining_hours_per_100": [ 25, 65 ] }, "AI-04": { "title": "Variance analysis", "range_percent": [ 15, 45 ], "baseline": "Recurring revenue, cost and margin variance analysis across three approved datasets and five dimensions, comparing actuals with budget, forecast or prior periods using existing validated calculations; retrieve results, explain drivers, investigate context, verify and approve.", "evidence_strength": "Conditional scenario informed by a recent variance-analysis case; no measured end-to-end result for this complete workflow.", "blocks": [ { "work": "Retrieve approved calculation outputs and sources", "baseline_hours": 10, "after_hours": [ 10, 10 ], "saving_rate_pct": [ 0, 0 ], "basis": "Analyst task-share assumption; no numeric retrieval saving is claimed from the qualitative Natura case.", "source_ids": [ "planning-natura_sap" ], "reduction_percent": [ 0.0, 0.0 ] }, { "work": "Model-backed driver decomposition and first source-linked explanation", "baseline_hours": 45, "eligible_baseline_hours": [ 20, 45 ], "eligible_task_saving_rate_pct": [ 90, 90 ], "after_hours": [ 27, 4.5 ], "saving_rate_pct": [ 40, 90 ], "basis": "The constrained case has 20 of these 45 baseline hours eligible for a configured recurring analysis; the remaining 25 hours run manually. The fast case covers all 45 hours. PwC's 90% stage result is applied unchanged only to eligible hours: after = (45 - eligible_hours) + eligible_hours * 0.10. Eligibility is an analyst assumption about work hours, not the proportion of questions or a measured adoption rate.", "source_ids": [ "planning-pwc_cfo_advisor" ], "reduction_percent": [ 40.0, 90.0 ] }, { "work": "Investigate missing business context and assess causal explanation", "baseline_hours": 20, "after_hours": [ 20, 20 ], "saving_rate_pct": [ 0, 0 ], "basis": "Analyst task-share assumption; retain all contextual judgment. This is operating analysis, not data repair or finance-model development.", "source_ids": [ "planning-natura_sap" ], "reduction_percent": [ 0.0, 0.0 ] }, { "work": "Verify figures and citations, correct answers and approve", "baseline_hours": 25, "incremental_correction_hours": [ 3, 0 ], "after_hours": [ 28, 25 ], "saving_rate_pct": [ -12, 0 ], "basis": "Retain all 25 baseline review hours. The constrained scenario explicitly adds three correction hours per 100 baseline hours; the fast scenario fits corrections inside existing review. Neither correction assumption is an empirical result or a discount to PwC's stage rate.", "source_ids": [], "reduction_percent": [ -12.0, 0.0 ] } ], "source_ids": [ "planning-pwc_cfo_advisor", "planning-natura_sap", "planning-bcg_finance_roi_2025" ], "assumptions": [ "Baseline task hours, eligible coverage and remaining human-work budgets are scenario assumptions.", "Published task results inform the scenario; no source establishes the complete workflow range.", "Same work volume and required output; manual fallback and corrections are included. Separate upkeep is excluded." ], "lower_case": "Twenty baseline hours of recurring bridges/decompositions are within configured coverage; the rest of the analytical stage remains manual. All baseline context and validation remain; three hours of extra corrections are charged.", "upper_case": "All 45 preparation hours are recurring, model-backed questions with stable definitions, supported drivers and approved sources. The configured stage reaches the reported PwC effect, while the full 45 hours of context plus review remain.", "reasoning": "Re-estimated from retained references with explicit before/after human-hour budgets.", "outside_band": [], "uncertainties": [ "The 45/20/25/10 work mix and 20-versus-45 eligible hours are unmeasured scenario assumptions. PwC's rate also accompanies process/tool/report changes; equal-output transfer is conditional. Natura validates the use case without quantifying savings. The display range is not a technology ceiling." ], "validation_needed": [ "Time the same 20 reference questions before/after, separately logging the four blocks and corrections.", "Classify eligible recurring work by baseline human hours before applying the saving formula.", "Require equivalent figures, citations, explanation support and output volume; include failed answers and manual fallback." ], "research_file": "reestimation-2026-10-25/model.json", "estimate_status": "analyst_scenario", "reference_review": "Conditional scenario informed by a recent variance-analysis case; no measured end-to-end result for this complete workflow.", "calculated_percent": [ 15, 40.5 ], "scenario_method": "before_after_hours", "scenarios": { "constrained": { "conditions": "Twenty baseline hours of recurring bridges/decompositions are within configured coverage; the rest of the analytical stage remains manual. All baseline context and validation remain; three hours of extra corrections are charged.", "remaining_hours": 85, "saved_hours_per_100": 15 }, "fast": { "conditions": "All 45 preparation hours are recurring, model-backed questions with stable definitions, supported drivers and approved sources. The configured stage reaches the reported PwC effect, while the full 45 hours of context plus review remain.", "remaining_hours": 59.5, "saved_hours_per_100": 40.5 } }, "higher_coverage_sensitivity": { "purpose": "Expose a faster case when a larger share of the same scoped question workload is configured; the displayed range is not a technology ceiling.", "baseline_task_mix_hours": { "retrieve_approved_outputs": 10, "eligible_model_backed_variance_preparation_and_first_explanation": 55, "business_context_and_causal_investigation": 10, "verification_corrections_and_approval": 25 }, "eligible_task_saving_pct": 90, "after_hours": 50.5, "saving_pct": 49.5, "outward_rounded_upper_pct": 50, "arithmetic": "10 + 55*0.10 + 10 + 25 = 50.5; (100-50.5)/100 = 49.5%", "condition": "An organization whose same approved-question workload spends 55 baseline hours on recurring, configured model-backed variance preparation and only 10 on additional context could reach this scenario. All 25 baseline checking/correction/approval hours remain. This is a different explicit baseline mix, not a newly measured saving or added implementation scope.", "source_ids": [ "planning-pwc_cfo_advisor" ] }, "previous_range_percent": [ 10, 35 ], "remaining_hours_per_100": [ 59.5, 85 ] }, "AI-05": { "title": "Management reporting", "range_percent": [ 20, 40 ], "baseline": "Prepare the same monthly 10-page management pack and ten variance narratives from two approved sources for one audience, including numerical checking, contextual edits, corrections and release approval.", "evidence_strength": "Recent named customer self-estimate closely matches commentary and review; the complete pack percentage still depends on an assumed task mix.", "blocks": [ { "work": "Refresh approved inputs and perform source/tie-out checks", "baseline_hours": 10, "after_hours": [ 10, 10 ], "saving_rate_pct": [ 0, 0 ], "basis": "Analyst task-share assumption. Sources are usable already; retain routine checks and claim no data-transformation benefit.", "source_ids": [], "reduction_percent": [ 0.0, 0.0 ] }, { "work": "Variance commentary, pack narrative/visual placement and finance review/correction", "baseline_hours": 80, "baseline_composition_hours": { "analyze_and_draft_ten_variances": 35, "assemble_same_ten_page_pack": 20, "finance_context_numeric_source_layout_checks_and_corrections": 25 }, "after_hours": [ 56, 40 ], "saving_rate_pct": [ 30, 50 ], "basis": "Apply Docker's reported 30-50% commentary/review estimate to this combined block. The 35/20/25 split makes the 80-hour analyst work-share assumption auditable; it is not a set of independently measured task effects. Source and scenario both retain human refinement, and review/corrections are included once inside the bundle.", "source_ids": [ "planning-docker_pigment" ], "reduction_percent": [ 30.0, 50.0 ] }, { "work": "Final accountable approval and release of the checked pack", "baseline_hours": 10, "after_hours": [ 10, 10 ], "saving_rate_pct": [ 0, 0 ], "basis": "Analyst task-share assumption; retain the full approval/release step outside the customer commentary/review estimate.", "source_ids": [], "reduction_percent": [ 0.0, 0.0 ] } ], "source_ids": [ "planning-docker_pigment", "planning-pwc_cfo_advisor", "planning-clickup_pigment" ], "assumptions": [ "Baseline task hours, eligible coverage and remaining human-work budgets are scenario assumptions.", "Published task results inform the scenario; no source establishes the complete workflow range.", "Same work volume and required output; manual fallback and corrections are included. Separate upkeep is excluded." ], "lower_case": "The same pack is produced; the combined drafting, packaging and review stage achieves the lower Docker self-estimate. Normal business-context edits, numerical/source checks and correction effort fit inside its 56 remaining hours; source checks and final approval retain 20 hours.", "upper_case": "Stable pack format and supported recurring variances allow the combined block to achieve the upper Docker self-estimate. Review and corrections fit inside its 40 remaining hours; all 20 hours of source checks and final approval remain.", "reasoning": "Re-estimated from retained references with explicit before/after human-hour budgets.", "outside_band": [], "uncertainties": [ "The 80% commentary/review share is an analyst scenario assumption, not a finding that all reporting teams spend 80% of time there. Docker's 30-50% is one customer's self-estimate and does not establish the full pack rate. ClickUp lacks a baseline denominator; PwC's 90% covers a configured variance stage. Neither is averaged into the model." ], "validation_needed": [ "Log three comparable monthly reporting cycles at the same ten pages and ten narratives.", "Record source checks, combined commentary/assembly/review, final approval and every correction/fallback separately.", "Replace the assumed 80-hour share with observed baseline hours; test whether net block saving meets the scenario after required quality checks." ], "research_file": "reestimation-2026-10-25/model.json", "estimate_status": "analyst_scenario", "reference_review": "Recent named customer self-estimate closely matches commentary and review; the complete pack percentage still depends on an assumed task mix.", "calculated_percent": [ 24, 40 ], "scenario_method": "before_after_hours", "scenarios": { "constrained": { "conditions": "The same pack is produced; the combined drafting, packaging and review stage achieves the lower Docker self-estimate. Normal business-context edits, numerical/source checks and correction effort fit inside its 56 remaining hours; source checks and final approval retain 20 hours.", "remaining_hours": 76, "saved_hours_per_100": 24 }, "fast": { "conditions": "Stable pack format and supported recurring variances allow the combined block to achieve the upper Docker self-estimate. Review and corrections fit inside its 40 remaining hours; all 20 hours of source checks and final approval remain.", "remaining_hours": 60, "saved_hours_per_100": 40 } }, "higher_coverage_sensitivity": { "purpose": "Expose a faster case where existing templates make pack assembly short and configured variance preparation dominates baseline hours; neither Docker 50% nor the displayed 40% workflow endpoint is a capability limit.", "baseline_task_mix_hours": { "source_checks": 10, "eligible_configured_variance_analysis_and_first_drafts": 65, "existing_template_pack_assembly": 5, "finance_context_checks_and_corrections": 15, "final_release_approval": 5 }, "eligible_task_saving_pct": 90, "after_hours": 41.5, "saving_pct": 58.5, "outward_rounded_upper_pct": 60, "arithmetic": "10 + 65*0.10 + 5 + 15 + 5 = 41.5; (100-41.5)/100 = 58.5%", "condition": "For the same ten narratives and ten-page pack, 65 baseline hours are spent on recurring configured variance analysis/first drafts, while an already efficient template requires five assembly hours. Preserve all 15 human context/check/correction hours plus final approval. Task shares are analyst assumptions to be replaced by a local time study; PwC 90% still applies only to the named eligible stage.", "source_ids": [ "planning-pwc_cfo_advisor" ] }, "additional_fast_configuration_sensitivity": { "purpose": "Show the PwC result without using it as the whole-pack rate or an empirical cap.", "baseline_eligible_variance_preparation_hours": 45, "eligible_task_saving_pct": 90, "all_other_work_hours_retained": 55, "after_hours": 59.5, "saving_pct": 40.5, "arithmetic": "45*0.10 + 55 = 59.5; 40.5% saved", "assumption": "An alternative task mix dedicates 45 baseline hours to configured variance analysis and first drafts; all 55 other hours, including corrections and approvals, stay unchanged. Do not add this result to the Docker scenario.", "source_ids": [ "planning-pwc_cfo_advisor" ] }, "previous_range_percent": [ 20, 45 ], "remaining_hours_per_100": [ 60, 76 ] }, "AI-06": { "title": "Finance information support", "range_percent": [ 30, 70 ], "baseline": "Find metric definitions/approved values, locate datasets, retrieve policies and precedents, prepare cited answers, check authority/current version and applicability, and resolve unsupported questions. Count all requester and finance-owner human work once. No downstream transactions or decisions; ongoing knowledge maintenance is separate.", "evidence_strength": "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.", "blocks": [ { "work": "Locate authoritative material: metric definitions/approved values, datasets, policies and precedents", "baseline_hours": 60, "coverage_fraction": [ 0.7, 0.95 ], "successful_fraction_of_covered_work": [ 0.85, 0.95 ], "successful_work_reduction_fraction": [ 0.8, 0.95 ], "basis": "FM estimates 6–10 minutes saved per search, without a baseline duration. This reference assumes a successful manual lookup takes 10 minutes and accepted assisted lookup takes 2 or 0.5 minutes: 8 or 9.5 minutes saved. Those timings, coverage (70%/95%) and success on covered work (85%/95%) are assumptions. They are not FM-measured percentages or precision/recall conversions. Uncovered and unsuccessful work retains its baseline time; all existing checking is separate.", "source_ids": [ "planning-fm_knowledge_2026" ], "after_hours": [ 31.44, 8.5575 ], "reduction_percent": [ 47.6, 85.7375 ] }, { "work": "Assemble and communicate a concise cited answer", "baseline_hours": 20, "coverage_fraction": [ 0.7, 0.95 ], "successful_fraction_of_covered_work": [ 0.85, 0.95 ], "successful_work_reduction_fraction": [ 0.6, 0.9 ], "basis": "Assumed residual human preparation of 40% or 10% for accepted grounded drafts. Template and recurring-answer conditions drive the high case. No cited source measures these exact writing reductions; excluded Noy/Zhang result is not reused.", "source_ids": [ "planning-amazon_regulatory_rag_2026", "planning-tr_copeland_buhl_2026" ], "after_hours": [ 12.86, 3.755 ], "reduction_percent": [ 35.7, 81.225 ] }, { "work": "Check source/version and applicability; resolve unsupported questions and approve answers", "baseline_hours": 20, "reduction_percent": [ 0.0, 0.0 ], "basis": "Preserve all existing human checking, judgment and escalation labor. This is an assumed baseline share, not an observed finance service time study. Count requester and finance-owner work once.", "after_hours": [ 20, 20 ], "source_ids": [] }, { "work": "Additional prompting, failed attempts and AI-specific correction", "baseline_hours": 0, "added_hours": [ 4, 1 ], "basis": "Explicit assumed extra labor per 100 baseline human hours, including attempts that fall back to the original work. In the factors above, unsuccessful/uncovered work retains its full baseline labor. Separate corpus maintenance is excluded.", "after_hours": [ 4, 1 ], "source_ids": [], "reduction_percent": null } ], "source_ids": [ "planning-fm_knowledge_2026", "planning-amazon_regulatory_rag_2026", "planning-tr_copeland_buhl_2026" ], "assumptions": [ "Baseline task hours, eligible coverage and remaining human-work budgets are scenario assumptions.", "Published task results inform the scenario; no source establishes the complete workflow range.", "Same work volume and required output; manual fallback and corrections are included. Separate upkeep is excluded." ], "lower_case": "70% coverage and 85% success on preparation work; accepted lookup 80% faster and cited-answer preparation 60% faster; all 20 existing checking/escalation hours retained; add 4 failure/correction hours. Net 31.7 hours saved.", "upper_case": "95% coverage and 95% success on preparation work; accepted lookup 95% faster and cited-answer preparation 90% faster; all 20 existing checking/escalation hours retained; add 1 failure/correction hour. Net 66.6875 hours saved.", "reasoning": "Re-estimated from retained references with explicit before/after human-hour budgets.", "outside_band": [ "Poor coverage, failed retrieval, a weak manual baseline or lengthy numerical/source reconciliation can yield less than 30% or negative savings.", "A service dominated by simple well-covered metric/dataset/policy lookups with very short retained checking can exceed 70%; report its measured task mix rather than raising a universal cap." ], "uncertainties": [ "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." ], "validation_needed": [ "Time matched reference questions across metrics, datasets, policies and precedents, including ambiguous, restricted, conflicting and unsupported cases.", "Record total active human minutes, including requester prompting, source verification, finance-owner escalation, failures and corrections; report model wait separately.", "Estimate time-weighted content coverage, successful-answer rate and stage reductions independently, then compare equal answer accuracy/currentness.", "Measure recurring corpus/access/update maintenance separately." ], "research_file": "reestimation-2026-10-25/model.json", "estimate_status": "analyst_scenario", "reference_review": "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.", "calculated_percent": [ 31.7, 66.6875 ], "scenario_method": "before_after_hours", "scenarios": { "constrained": { "conditions": "70% coverage and 85% success on preparation work; accepted lookup 80% faster and cited-answer preparation 60% faster; all 20 existing checking/escalation hours retained; add 4 failure/correction hours. Net 31.7 hours saved.", "remaining_hours": 68.3, "saved_hours_per_100": 31.7 }, "fast": { "conditions": "95% coverage and 95% success on preparation work; accepted lookup 95% faster and cited-answer preparation 90% faster; all 20 existing checking/escalation hours retained; add 1 failure/correction hour. Net 66.6875 hours saved.", "remaining_hours": 33.3125, "saved_hours_per_100": 66.6875 } }, "direct_lookup_high_speed_example": { "status": "Illustrative assumption; exact before/after not observed in cited studies", "before_human_minutes": 10, "before_preparation_minutes": 9.5, "before_existing_check_minutes": 0.5, "after_human_minutes": 1, "after_preparation_minutes": 0.5, "after_existing_check_minutes": 0.5, "calculated_saving_percent": 90, "conditions": "One well-covered, unambiguous question; current authorized source; short answer; accepted on first attempt; source checking takes 30 seconds in both methods. A 20-minute baseline with the same 1-minute assisted total would save 95%.", "interpretation": "90%+ task-level savings are entirely possible under these assumptions. The mixed-workflow scenario is lower because of its assumed work mix, coverage and full retained checking, and does not impose a 70% cap on every service." }, "previous_range_percent": [ 15, 35 ], "remaining_hours_per_100": [ 33.3125, 68.3 ] }, "AI-07": { "title": "Tax monitoring", "range_percent": [ 20, 50 ], "baseline": "Monitor five official sources for one tax regime across two jurisdictions, including checks with no relevant change; triage and read updates, extract dates, draft summaries and obtain expert applicability approval. Same update volume/coverage and approved output. No filing, autonomous advice or policy implementation.", "evidence_strength": "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.", "blocks": [ { "work": "Check the five official sources, including no-change checks; detect and triage updates", "baseline_hours": 30, "coverage_fraction": [ 0.8, 1.0 ], "successful_fraction_of_covered_work": [ 0.9, 0.98 ], "successful_work_reduction_fraction": [ 0.8, 0.95 ], "basis": "Analyst operational assumptions, without an empirical scan-time anchor. Low: feeds/readable exports cover 80% of baseline scanning work, 90% succeeds and accepted automation removes 80% of its human time. High: all five sources work reliably, 98% succeeds, accepted scan/triage needs 5% of its former time. Human spot checks and false-positive review remain in residual scan time; no-change checks belong to the denominator.", "source_ids": [ "planning-amazon_regulatory_rag_2026" ], "after_hours": [ 12.72, 2.07 ], "reduction_percent": [ 57.6, 93.1 ] }, { "work": "Read relevant updates, extract publication/effective dates and draft cited summaries", "baseline_hours": 30, "basis": "Apply the adjacent reported 32% research saving to the entire 30-hour read/extract/draft block in the constrained case. The fast case assumes 65% less time, consistent with the conditional same-output interpretation of Copeland Buhl’s 2–3-times-faster research claim. Both are scope transfers to monitoring, not observed monitoring effects. No additional coverage or success discount is applied to these real-world research results.", "source_ids": [ "planning-tr_cocounsel_corporate", "planning-tr_copeland_buhl_2026" ], "after_hours": [ 20.4, 10.5 ], "reduction_percent": [ 32.0, 65.0 ] }, { "work": "Check legal applicability, effective dates and completeness; correct and approve summaries", "baseline_hours": 40, "reduction_percent": [ 0.0, 0.0 ], "basis": "Retain all expert review and existing correction labor. Forty hours is an assumed monitoring-workflow share, not a measured result. The PLAT benchmark supplies accuracy context rather than a review-hours factor. No filing or autonomous advice is included.", "after_hours": [ 40, 40 ], "source_ids": [] }, { "work": "Additional failed-attempt, false-positive and AI-specific correction labor", "baseline_hours": 0, "added_hours": [ 3, 1 ], "basis": "Assumed extra human work per 100 baseline hours, including investigating missed or misclassified updates and checking failures. Unsuccessful/uncovered work also keeps its full baseline labor above. Ongoing source-feed maintenance is separate.", "after_hours": [ 3, 1 ], "source_ids": [], "reduction_percent": null } ], "source_ids": [ "planning-tr_cocounsel_corporate", "planning-tr_copeland_buhl_2026", "planning-amazon_regulatory_rag_2026", "planning-tax_plat" ], "assumptions": [ "Baseline task hours, eligible coverage and remaining human-work budgets are scenario assumptions.", "Published task results inform the scenario; no source establishes the complete workflow range.", "Same work volume and required output; manual fallback and corrections are included. Separate upkeep is excluded." ], "lower_case": "Scanning has 80% coverage, 90% success and 80% accepted-task saving; read/draft work uses the reported 32% research effect without another coverage discount. Retain all 40 expert-review hours and add 3 correction hours.", "upper_case": "All source checks are covered, with 98% success and 95% accepted-task saving; the research block assumes 65% less time. Retain all 40 expert-review hours and add 1 correction hour.", "reasoning": "Re-estimated from retained references with explicit before/after human-hour budgets.", "outside_band": [ "Source failures, weak archives, heavy legal interpretation or long review/correction can put results below 20% or below zero.", "More than 45% is possible when routine source checks/drafting dominate baseline time and review is a smaller share; 40% retained review here creates a scenario-specific limit, not a universal tax-monitoring cap." ], "uncertainties": [ "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." ], "validation_needed": [ "Time scan work on all five sources, including no-change days and irrelevant updates; compare against the actual existing alert workflow.", "Use the same historical and live update sets and require equivalent recall; missing a relevant update is not a saving.", "Time human triage, date extraction, drafting, applicability approval, rejected summaries and corrections separately.", "Track sources/updates that fall back to manual processing and record the extra failed-attempt labor.", "Measure source-feed upkeep, test updates and classification maintenance separately." ], "research_file": "reestimation-2026-10-25/model.json", "estimate_status": "analyst_scenario", "reference_review": "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.", "calculated_percent": [ 23.88, 46.43 ], "scenario_method": "before_after_hours", "scenarios": { "constrained": { "conditions": "Scanning has 80% coverage, 90% success and 80% accepted-task saving; read/draft work uses the reported 32% research effect without another coverage discount. Retain all 40 expert-review hours and add 3 correction hours.", "remaining_hours": 76.12, "saved_hours_per_100": 23.88 }, "fast": { "conditions": "All source checks are covered, with 98% success and 95% accepted-task saving; the research block assumes 65% less time. Retain all 40 expert-review hours and add 1 correction hour.", "remaining_hours": 53.57, "saved_hours_per_100": 46.43 } }, "previous_range_percent": [ 10, 35 ], "remaining_hours_per_100": [ 53.57, 76.12 ] }, "AI-08": { "title": "Audit evidence preparation", "range_percent": [ 25, 35 ], "baseline": "Retrieve the fixed change-ticket sample, extract evidence for the same three test criteria, prepare source-linked workpapers, independently verify evidence and reach the auditor's conclusion. Same sample, audit coverage and evidence standard at both endpoints.", "evidence_strength": "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.", "blocks": [ { "work": "Locate tickets and gather supporting evidence", "baseline_hours": 20, "reduction_percent": [ 25.0, 50.0 ], "after_hours": [ 15, 10 ], "basis": "Assumed retrieval budget: 20 hours before, 15 with partial assistance or 10 with consistent indexed evidence. The 50% fast saving is a scenario assumption; the document-processing case does not separately time retrieval.", "source_ids": [ "accounting-microsoft_sox", "accounting-provectus_johnson_lambert" ] }, { "work": "Extract approval, testing and date attributes; populate rule checks", "baseline_hours": 30, "reduction_percent": [ 50.0, 50.0 ], "after_hours": [ 15, 15 ], "basis": "Use the adjacent case’s 50% document-processing result as the reference extraction assumption: 30 hours becomes 15. Independent source verification stays in the final block; this result does not measure the entire audit workflow.", "source_ids": [ "accounting-microsoft_sox", "accounting-provectus_johnson_lambert" ] }, { "work": "Draft and edit source-linked workpapers and annotations", "baseline_hours": 20, "reduction_percent": [ 40.0, 50.0 ], "after_hours": [ 12, 10 ], "basis": "Assumed workpaper preparation: 20 hours becomes 12 with substantial editing or 10 with consistent templates and source links. No published writing-time percentage is transferred.", "source_ids": [ "accounting-microsoft_sox", "accounting-provectus_johnson_lambert" ] }, { "work": "Independently verify original evidence, resolve exceptions and reach the conclusion", "baseline_hours": 30, "reduction_percent": [ -10.0, 0.0 ], "after_hours": [ 33, 30 ], "basis": "All thirty baseline auditor-review hours remain at the upper endpoint. Lower adds three hours to resolve omitted/conflicting attributes and annotations. Prepopulated pass/fail fields do not determine the final audit conclusion.", "source_ids": [ "accounting-afm_controls" ] } ], "source_ids": [ "accounting-microsoft_sox", "accounting-provectus_johnson_lambert", "accounting-afm_controls" ], "assumptions": [ "Baseline task hours, eligible coverage and remaining human-work budgets are scenario assumptions.", "Published task results inform the scenario; no source establishes the complete workflow range.", "Same work volume and required output; manual fallback and corrections are included. Separate upkeep is excluded." ], "lower_case": "Partly indexed evidence and substantial editing leave 15 retrieval + 15 extraction + 12 drafting + 33 independent-review/correction hours: 75 remain, 25 saved.", "upper_case": "Indexed evidence, consistent formats and reusable workpapers leave 10 retrieval + 15 extraction + 10 drafting + 30 independent-review hours: 65 remain, 35 saved.", "reasoning": "Re-estimated from retained references with explicit before/after human-hour budgets.", "outside_band": [], "uncertainties": [ "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." ], "validation_needed": [ "Time matched manual and assisted preparation on the same fifty historical cases, with held-out adjudication", "Record attribute omissions, wrong evidence links, false passes and missing/conflicting approvals separately", "Measure preparer and independent auditor labor including correction and source re-performance; retain the same evidential standard", "Test whether preparation fits the reference 35–42 hour budget per 100 baseline hours, while preserving the full auditor-review and correction effort." ], "research_file": "reestimation-2026-10-25/model.json", "estimate_status": "analyst_scenario", "reference_review": "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.", "calculated_percent": [ 25, 35 ], "scenario_method": "before_after_hours", "scenarios": { "constrained": { "conditions": "Partly indexed evidence and substantial editing leave 15 retrieval + 15 extraction + 12 drafting + 33 independent-review/correction hours: 75 remain, 25 saved.", "remaining_hours": 75, "saved_hours_per_100": 25 }, "fast": { "conditions": "Indexed evidence, consistent formats and reusable workpapers leave 10 retrieval + 15 extraction + 10 drafting + 30 independent-review hours: 65 remain, 35 saved.", "remaining_hours": 65, "saved_hours_per_100": 35 } }, "near_automatic_sensitivity": { "label": "Near-automatic evidence preparation", "saving_percent": 65, "before_hours": [ 20, 30, 20, 30 ], "after_hours": [ 1, 2, 2, 30 ], "conditions": "Complete indexed evidence and reliable template population reduce preparation from 70 to 5 hours. All 30 independent-review hours remain. This is an unvalidated automation scenario, beyond the quantified case reference." }, "previous_range_percent": [ 20, 35 ], "remaining_hours_per_100": [ 65, 75 ] }, "AI-09": { "title": "Investor briefing", "range_percent": [ 25, 55 ], "baseline": "One quarterly script and eight peer-call summaries from approved digital sources; include preparation, editing, fact/quotation checks and publication approval. Exclude event logistics, website, CRM and the rest of IR.", "evidence_strength": "Analyst work-share scenario. Recent references remain, but none establishes these whole-workflow endpoints.", "blocks": [ { "work": "Gather approved source packs and organize current materials", "baseline_hours": 15, "reduction_percent": [ 0, 20.0 ], "basis": "Analyst assumption: retrieval assists organization but approved input preparation remains.", "source_ids": [ "ir-procurement-q4-workflow-2026" ] }, { "work": "Draft and edit the earnings script", "baseline_hours": 30, "reduction_percent": [ 45.0, 80.0 ], "basis": "Analyst transfer below the 88.9% first-draft TEI reference to allow tone changes, contextual editing and different tool quality.", "source_ids": [ "ir-procurement-q4-forrester-2025" ] }, { "work": "Read and synthesize the eight peer calls", "baseline_hours": 35, "reduction_percent": [ 50.0, 80.0 ], "basis": "Analyst transfer capped at the TEI's 80% peer-summary reference; low endpoint retains more transcript reading.", "source_ids": [ "ir-procurement-q4-forrester-2025" ] }, { "work": "Verify figures, claims and quotations; complete disclosure approval", "baseline_hours": 20, "reduction_percent": [ -25.0, 0 ], "basis": "Analyst safeguard: no approval saving, and up to five extra hours of checking per 100 baseline workflow hours.", "source_ids": [ "ir-procurement-q4-workflow-2026" ] } ], "source_ids": [ "ir-procurement-q4-forrester-2025", "ir-procurement-murphy-q4", "ir-procurement-q4-workflow-2026" ], "assumptions": [ "Manual drafting and peer synthesis remain substantial baseline work; source packs and transcripts are already available in readable form.", "Users check source links and retain existing disclosure approval. Additional fact-checking can increase the review block by 25% at the low endpoint.", "The 15/30/35/20 baseline allocation is a transparent analyst scenario, not a measured industry distribution.", "The upper endpoint uses smaller drafting savings than the consultant's 88.9% first-draft reference and allows no final-review saving." ], "lower_case": "", "upper_case": "", "reasoning": "Drafting and peer summaries shrink; source preparation and disclosure checks remain. Task anchors: sponsored consulting model of 36→4 h for a first script draft and 10→2 h per peer summary.", "outside_band": [ "AI drafts require substantial rewriting or repeat source checking", "Existing templates and summaries already remove most preparation labor", "Expansion to more peers or richer output absorbs released time" ], "uncertainties": [ "Share of current time spent on final review versus first drafts and peer summaries", "Whether baseline research already uses summaries or transcripts", "Correction burden for financial claims and executive tone" ], "validation_needed": [ "Time one comparable earnings cycle and separately log preparation, first drafts, peer synthesis, edits and final review; compare equal outputs and eight peers." ], "research_file": "source-review-2026-10-25/review.json", "review_block_index": 3, "estimate_status": "analyst_scenario", "reference_review": "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.", "calculated_percent": [ 26.0, 55.0 ], "remaining_hours_per_100": [ 45.0, 74.0 ], "review_sensitivity": { "review_before": 20, "review_after": 35, "calculated_percent": [ 16.4375, 44.6875 ], "range_percent": [ 15, 45 ] } }, "AI-10": { "title": "Sourcing and bid comparison", "range_percent": [ 15, 35 ], "baseline": "One sourcing category, one RFP and six bids; include requirements, drafting, extraction/comparison, verification, clarification, negotiation and award approval. Exclude broader category strategy, supplier onboarding and contract migration.", "evidence_strength": "Analyst work-share scenario. Recent references remain, but none establishes these whole-workflow endpoints.", "blocks": [ { "work": "Agree requirements and evaluation criteria", "baseline_hours": 15, "reduction_percent": [ 0, 10.0 ], "basis": "Analyst assumption: light template/retrieval help; stakeholder decisions remain.", "source_ids": [ "ir-procurement-bcg-procurement-2025" ] }, { "work": "Draft and revise the RFP", "baseline_hours": 25, "reduction_percent": [ 30.0, 50.0 ], "basis": "Analyst transfer informed by BCG's 40% tender-creation and approximate 50% drafting cases; not an exact pooled estimate.", "source_ids": [ "ir-procurement-bcg-procurement-2025" ] }, { "work": "Extract terms and prepare a six-bid comparison", "baseline_hours": 35, "reduction_percent": [ 30.0, 50.0 ], "basis": "Analyst transfer from the approximate 50% offer-comparison case; preserve readable evidence links and human interpretation.", "source_ids": [ "ir-procurement-bcg-procurement-2025" ] }, { "work": "Verify prices, omissions and competitive claims; resolve clarifications", "baseline_hours": 15, "reduction_percent": [ -20.0, 0 ], "basis": "Analyst safeguard: no review saving; up to three extra checking hours per 100 current workflow hours.", "source_ids": [ "ir-procurement-bid-scoring-volatility-2026" ] }, { "work": "Negotiate and approve the award", "baseline_hours": 10, "reduction_percent": [ 0, 0 ], "basis": "Outside automated authority; no saving credited.", "source_ids": [ "ir-procurement-bid-scoring-volatility-2026" ] } ], "source_ids": [ "ir-procurement-bcg-procurement-2025", "ir-procurement-gatx-levelpath-2025-july", "ir-procurement-bid-scoring-volatility-2026" ], "assumptions": [ "RFP drafting and bid comparison together represent 60% of this modeled workflow; all shares are analyst judgments.", "The tool creates a source-linked comparison, while buyers retain commercial judgment, negotiations and award decisions.", "RFP and comparison savings are bounded below the most dramatic runtime/capacity claims and around the BCG task references.", "At the low endpoint, checking AI extraction and omitted terms increases the review block by 20%." ], "lower_case": "", "upper_case": "", "reasoning": "RFP drafting and bid comparison shrink; requirements, checking, negotiation and award approval remain. Consulting cases report roughly 40–50% task improvements, with incomplete timing methods.", "outside_band": [ "Differentiated bids require extensive expert interpretation", "Unreadable files, ambiguous units or missing exclusions produce lengthy correction loops", "Existing e-sourcing templates already remove most draft/comparison effort" ], "uncertainties": [ "Current shares of drafting/comparison versus negotiation and clarification", "Bid complexity and document quality", "Time needed to verify price normalization and omitted commercial terms" ], "validation_needed": [ "Time a six-bid event using matched category complexity and output quality; include stakeholder, buyer and legal labor, plus corrections and negotiation." ], "research_file": "source-review-2026-10-25/review.json", "review_block_index": 3, "estimate_status": "analyst_scenario", "reference_review": "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.", "calculated_percent": [ 15.0, 31.5 ], "remaining_hours_per_100": [ 68.5, 85.0 ], "review_sensitivity": { "review_before": 15, "review_after": 30, "calculated_percent": [ 8.8235, 25.9412 ], "range_percent": [ 5, 30 ] } }, "AI-11": { "title": "Shared finance AI foundation", "range_percent": null, "baseline": "Shared identity, retrieval, logging and evaluations for one approved vendor-hosted product.", "evidence_strength": "Architecture/work evidence, no calibrated reuse percentage", "blocks": [], "source_ids": [ "ir-procurement-aws-finance-foundation", "ir-procurement-microsoft-controls-2025" ], "assumptions": [], "lower_case": "", "upper_case": "", "reasoning": "Application savings are counted on the application rows. This foundation enables those projects and can reduce repeated setup; it adds central maintenance work.", "outside_band": [], "uncertainties": [], "validation_needed": [ "A delivery-effort reuse estimate would require logs for common setup versus application-specific implementation across multiple comparable deployments; it is a separate metric from operating workflow savings." ], "research_file": "source-review-2026-10-25/review.json", "estimate_status": "indirect", "reference_review": "Keep Indirect. Retained sources describe architecture and work; none measures saved implementation labor across comparable deployments. Application workflow savings should remain on application rows." } }, "source_date_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.", "source_audit": "source-review-2026-10-25/review.json" }