Benchmark Report 03 · Finance function
Demand and delivery run in opposite directions across the finance calendar.
Accounts Payable has the highest demand of the eleven functions and the least growth left in it.
FP and A has the second highest demand and the widest gap between wanting AI and running it, at 39 points.
One function breaks the pattern. Investor Relations is the only entity in this series where adoption exceeds demand.
11 functions · revenue $100M to $5B · window Aug 2024 to Jul 2026 · Q3 2026 edition+39point gap on FP and A, the widest of the eleven and the function CFOs most want AI for.
0.4 daysis how far the median month-end close moved between 2015 and 2024, from 6.4 to 6.0 days.
-9Investor Relations, the only negative gap anywhere in the series. Adoption arrived ahead of demand.
The index
Eleven functions, scored on demand, adoption, readiness and momentum.
Autonomy ceiling is this report's own diagnostic: how far a function can realistically go toward unattended execution.
It is the reason AP and Expenses can be pushed hard and Treasury and FP and A cannot.
| # | Finance function | DI | AD | RI | MI | CAII | Gap | Autonomy |
|---|---|---|---|---|---|---|---|---|
| 1 | Accounts Payable | 96 | 71 | 84 | 58 | 82.3 | +25 | 78 |
| 2 | Financial Close | 89 | 62 | 71 | 66 | 75.0 | +27 | 62 |
| 3 | Expense Management | 69 | 64 | 79 | 67 | 69.3 | +5 | 74 |
| 4 | Accounts Receivable | 78 | 58 | 72 | 61 | 69.1 | +20 | 70 |
| 5 | FP&A | 83 | 44 | 58 | 69 | 64.9 | +39 | 35 |
| 6 | Financial Reporting | 74 | 55 | 63 | 60 | 64.7 | +19 | 55 |
| 7 | Compliance & Internal Audit | 71 | 49 | 52 | 72 | 60.7 | +22 | 41 |
| 8 | Tax | 61 | 47 | 60 | 77 | 58.2 | +14 | 52 |
| 9 | Treasury | 64 | 41 | 55 | 70 | 55.9 | +23 | 38 |
| 10 | Investor Relations | 42 | 51 | 66 | 74 | 52.7 | -9 | 44 |
| 11 | Inventory & Cost Accounting | 58 | 38 | 47 | 63 | 50.3 | +20 | 46 |
| Cohort mean | 71.4 | 52.7 | 64.3 | 67.0 | 63.9 | +18.6 | 54 |
Finding 01
The functions CFOs want most are the ones least able to run unattended
FP and A carries the second highest demand and an autonomy ceiling of 35, the second lowest. Treasury sits at 38.
Both are judgement functions where a wrong answer does not announce itself.
The two with the highest ceilings, AP at 78 and Expenses at 74, carry the narrowest paths to further gain. Where AI can safely run alone, the work is mostly done.
Demand against production, by function
Accounts Payable+25
Financial Close+27
Expense Management+5
Accounts Receivable+20
FP&A+39
Financial Reporting+19
Compliance & Internal Audit+22
Tax+14
Treasury+23
Investor Relations-9
Inventory & Cost Accounting+20
Demand
In production
Bar: the addressable gap
Investor Relations runs the other way, in green. Production sits ahead of demand by nine points, the only negative gap in the series.
Finding 02
A decade of close automation moved the median close by four tenths of a day
APQC has tracked median monthly close cycle time across roughly 2,300 organisations since 2015.
Over that decade, close automation software became standard and the median went from 6.4 days to 6.0.
Tooling without process change buys almost nothing. Teams that pair automation with a redesigned close finish inside six days at 71 percent, against 23 percent of those that do not.
Median monthly close, days
6.66.15.6
201520172019202120222024
The y axis starts at 5.6 days rather than zero, which is what makes the flatness visible at all. On a zero baseline these six bars look identical.
Measured check
A live search pull confirms two findings outright and puts a warning on the Tax number.
Where the measurement agrees
FP and A returns zero agentic search volume across the entire five year window. It is the only function of the eleven with a true zero on the agent pattern.
That is independent support for the widest gap in the report. People search for AI in FP and A. Almost nobody searches for an agent to run it.
Investor Relations measures as the fastest growing of the eleven, rising from an index of 0.04 in 2024 to 44.5 in 2026, matching its modelled momentum score of 74.
Where it disagrees
Demand for Tax, Compliance and Treasury is inflated by search terms that are not about the CFO office at all.
Measured against narrow phrasing, "AI for tax" runs 26.6 times higher than "AI for corporate tax", and "AI for compliance" 14 times higher than "AI for financial compliance".
On CFO-specific phrasing the order becomes corporate tax, financial compliance, accounts payable, treasury management, then FP and A a distant fifth.
Finding 03
Autonomy ceiling explains the deployment order better than budget does
Nothing here correlates with adoption as cleanly as the autonomy ceiling.
The four functions above 60 average 64 percent in production. The four below 45 average 42 percent.
Control requirements set the order in which finance automates. A function where an error reverses cheaply gets an agent. One where an error compounds silently gets a pilot that never graduates.
Autonomy ceiling, ranked
Green: unattended execution realistic
Amber: under 50, not currently realistic
Momentum
Tax leads on growth, and it is the only function whose deadline is written into law.
| Finance function | 24 month growth | Peak quarter | Since peak | Fastest growing query |
|---|---|---|---|---|
| Tax | 6.8x | Q2 2026 | -2% | ksef france e invoicing readiness |
| Investor Relations | 6.1x | Q2 2026 | -4% | ai earnings call qa preparation |
| Compliance & Internal Audit | 5.9x | Q2 2026 | -7% | ai controls testing pcaob evidence |
| FP&A | 5.6x | Q1 2026 | -17% | ai driver based rolling forecast |
| Treasury | 5.5x | Q1 2026 | -11% | ai cash forecast accuracy 13 week |
| Expense Management | 5.3x | Q2 2026 | -6% | detect ai generated fake receipt |
| Financial Close | 5.1x | Q1 2026 | -14% | ai agent month end close reconciliation |
| Inventory & Cost Accounting | 4.9x | Q1 2026 | -12% | ai landed cost standard cost variance |
| Accounts Receivable | 4.8x | Q1 2026 | -13% | ai cash application unapplied cash |
| Financial Reporting | 4.6x | Q4 2025 | -18% | ai variance commentary flux analysis |
| Accounts Payable | 4.4x | Q4 2025 | -22% | ai agents for invoice exception handling |
France went live on 1 September 2026. Poland reached all VAT-registered businesses in April 2026, and Belgium went live for all business to business invoicing on 1 January 2026.
Accounts Payable sits last on momentum while carrying the highest demand, which is what a mature category looks like.
Function notes
Eleven functions, with the benchmark, the failure mode and the finding.
Accounts Payable82.3
Gap +25 · autonomy ceiling 78
Benchmark
Average cost to process an invoice $9.40, best in class $2.78. Industry average touchless rate 32.6 percent, best in class 49.2 percent.
What breaks
The ceiling on a touchless rate is set by purchase order discipline and supplier master data, not by the parser.
Finding
Highest demand of any function and the lowest momentum. The remaining value sits in the exception queue, not the extraction step.
Financial Close75.0
Gap +27 · autonomy ceiling 62
Benchmark
APQC median cycle time 6.4 days across roughly 2,300 organisations. Gartner estimates 62 percent still use spreadsheets as a primary close tool.
What breaks
The close is a coordination problem wearing an accounting costume. Most delay is waiting on somebody else's number.
Finding
A decade of close automation moved the median by four tenths of a day. That is the strongest argument here that tooling without process change buys nothing.
Expense Management69.3
Gap +5 · autonomy ceiling 74
Benchmark
Average expense report costs about $58 and 20 minutes. 19 percent contain errors costing an extra $52 and 18 minutes each.
What breaks
AI generated receipts went from 0 percent of flagged fakes in March 2025 to 70.8 percent by mid May 2026.
Finding
The only function where AI is both the fastest growing threat and the only workable control. Manual receipt review stopped being viable.
Accounts Receivable69.1
Gap +20 · autonomy ceiling 70
Benchmark
DSO 39 days and touchless payment 92 percent. 67 percent of mid-market CFOs name cash application their top AR pain point.
What breaks
Collections is not the bottleneck. Cash application is. Unmatched payment becomes unapplied cash, then a phantom receivable.
Finding
The highest share of workflow intent queries of any function. AR searchers arrive closer to a live project than anyone else.
FP&A64.9
Gap +39 · autonomy ceiling 35
Benchmark
65 percent of CFOs raised technology budgets by 20 percent or more, yet only 28 percent use AI in forecasting.
What breaks
Forecast quality is bounded by the driver model, and most driver models live in one analyst's head.
Finding
The widest gap in the report. FP and A is the function CFOs most want AI for and least often have, on the weakest realised impact.
Financial Reporting64.7
Gap +19 · autonomy ceiling 55
Benchmark
62 percent of finance organisations still use spreadsheets as a primary reporting tool. 88 to 90 percent of spreadsheets contain at least one error.
What breaks
Variance commentary is written under deadline by the person least able to check it, then reviewed by the person least able to challenge it.
Finding
Highest Copilot share outside FP and A, because reporting output lands in Word, Excel and PowerPoint rather than in the ERP.
Compliance & Internal Audit60.7
Gap +22 · autonomy ceiling 41
Benchmark
Organisations spend an average $2.3 million a year on SOX. Generative AI adoption among Chief Audit Executives doubled from 15 to 40 percent.
What breaks
AI makes 100 percent population testing feasible, but PCAOB standards were written around sampling and approval has not followed.
Finding
Third highest momentum and third lowest readiness. Document full population testing as an enhancement, not a replacement.
Tax58.2
Gap +14 · autonomy ceiling 52
Benchmark
More than 30 countries are expected to mandate e-invoicing by 2030. Belgium went live January 2026 with no phase-in, France September 2026.
What breaks
A clearance model does not fail at audit, it fails at delivery. A tax data problem becomes an AR problem within one billing cycle.
Finding
Highest momentum of the eleven, and the only function whose timeline is set by statute. Statute is a better sponsor than a business case.
Treasury55.9
Gap +23 · autonomy ceiling 38
Benchmark
73 percent name cash forecasting their top priority, and 59 percent name data quality rather than technology as the barrier.
What breaks
Treasury moves money, which puts every agent in the payment path behind the highest control bar in finance.
Finding
Most likely department to catch a fraud attempt at 83 percent, least likely to use AI to do it at 17 percent. The clearest unclaimed control case.
Investor Relations52.7
Gap -9 · autonomy ceiling 44
Benchmark
51 percent of IR professionals have embedded AI, up from 30 percent in 2024 and under 10 percent in 2023. Only 2 percent have no plans.
What breaks
Everything an IR team publishes is a regulated disclosure. The AI can draft and rehearse, but accountability never moves.
Finding
The only function in the series where adoption exceeds demand. Nobody searched for it. It arrived with the communications platform.
Inventory & Cost Accounting50.3
Gap +20 · autonomy ceiling 46
Benchmark
Auditors consistently identify inventory as the balance sheet area most prone to accounting issues. Shrinkage typically runs 1.5 to 3 percent.
What breaks
Net variance hides gross error. A standard cost roll that nets small can contain two large offsetting errors.
Finding
Lowest index score of the eleven. Inventory sits at the ERP and plant floor boundary, so the data an agent needs is in a system finance does not own.
Work it out for your estate
What the functions you have not automated cost
The gap column shows which parts of the calendar are unclaimed across the segment.
Open the AI Cost Estimator
Four inputs: level of finance AI, ERP complexity, industry and annual revenue.
Method
Sourced anchors
How this was measured, and what it cannot tell you.
Demand
Eight keyword groups per function, chained to a constant anchor and rebased across the cohort.
Narrow check
Broad function nouns were re-run against CFO-specific phrasing, and the divergence is reported rather than smoothed away.
Panel
412 organisations. Production means posting or preparing real transactions in a real period, not a pilot.
Limits
Search interest measures attention, not spend. The panel is self selected, so adoption reads as an upper bound.
Ardent Partners, AP Metrics That MatterAPQC close and accounts payable benchmarksBilltrust 2026 AR Benchmark ReportPYMNTS Intelligence, mid-market CFO surveyLimelight 2026 FP and A Trends ReportKPMG 2025 SOX SurveyIIA 2025 Chief Audit Executive surveyNasdaq Global IR Issuer PulseGBTA Foundation expense benchmarksACFE 2026 Report to the NationsAFP 2025 Treasury Benchmarking and 2026 Payments FraudPublished e-invoicing mandate calendar
Every figure marked traces to one of these, and each is re-verified at the quarterly refresh.
Get Started
See what a Finance AI platform does on the ERP you already run.
This index measures where finance AI demand sits and how little of it is running.
ChatFin is built for the half that converts first: reconciliation, cash application, AP matching and close.
Every action runs through your existing ERP and lands in a log a controller can sign off.
- ✓See AI agents working directly with your ERP data
- ✓Explore workflow automation for your specific finance operations
- ✓Get a custom deployment roadmap and implementation timeline
Request a Demo
Confidential & Private
Source: ChatFin CFO Office AI Intent Index, Q3 2026 edition, finance function cut. 11 functions, revenue segment $100M to $5B, window August 2024 to July 2026. Panel of 412 finance organisations, where production means posting or preparing real transactions in a real period.
Demand figures are modelled priors checked against a measured Google Trends pull. Figures are directional and refreshed quarterly.