Invoice processing
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: Analyst work-share scenario. Recent references remain, but none establishes these whole-workflow endpoints.
Reference task model
The reference workload starts with 100 manual staff-hours. Organization settings adjust the hours below at the same work volume. Work shares and task reductions are assumptions informed by the cited evidence. Negative values add checking work.
| Operating work | Before staff-hours | Lower saving assumption | Higher saving assumption | Staff-hours saved lower case | Staff-hours saved higher case |
|---|---|---|---|---|---|
Invoice intake and extractionAssumption and referencesAnalyst-selected baseline share and task rates; combined document extraction can remove much rekeying, but leave format exceptions and source checking. Sources: 1 | 35 | 50% | 80% | 17.5 | 28 |
| Supplier lookup and accounting coding | 25 | 0% | 60% | 0 | 15 |
Prepare the ERP draft billAssumption and referencesDraft assembly removes duplicate typing; rates are transfer assumptions, not published stage measurements. Sources: 1 | 15 | 50% | 80% | 7.5 | 12 |
Check drafts, correct exceptions and approveAssumption and referencesApproval 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. Sources: 3 | 25 | -30% | 0% | -7.5 | 0 |
| Reference workload total | 100 | 17.5 | 55 |
17.5–55% before outward rounding to 15–55%. These are scenario endpoints, not confidence limits.
How much does the work mix matter?
The work shares above have not been measured. If independent checking takes 40 of the original 100 hours, instead of 25, the same task-saving assumptions produce 5–45%.
Other work shares shrink proportionately so the baseline remains 100 hours. The additional 15-hour review share is an illustrative sensitivity, not a measured allocation. An organization with a different work mix needs its own baseline.
Citations
Deloitte: Hands off the task, eyes on the outcome
Consultancy-authored anonymous customer deployment · 2026-10-19
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.
Evidence period: Not stated. Describes an implemented Zora AI deployment after earlier ERP/process/RPA work; measurement dates and model version are absent.
Microsoft customer story: Pamarco Technologies
Vendor-authored named customer deployment · 2026-09-18
Pamarco reports about 95% success coding freight invoices. It estimates much larger time savings after full adoption, including automatic posting.
Evidence period: Deployment and measurement dates not stated; 2026 case describes current ERP-connected Copilot Studio/MCP use.
Choi and Xie: Human + AI in Accounting — 2025 framed experiment only
Peer-reviewed randomized framed experiment · 2026-04-16
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.
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.
How to interpret this range
Conditional planning scenarios, not measured population bounds or confidence intervals.
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.
- 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.
Savings include conventional extraction, matching, retrieval and integration delivered with each scoped project. The separate marginal contribution of GenAI is not established.
Separate costs: Implementation and training, Separate project/platform upkeep, Upstream remediation outside the project, Additional output and newly worthwhile work.
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.
Convert to monthly hours: 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.
What would improve the estimate
- 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.
- 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.
Reviewed 2026-10-25 · Reference model data · Research notes