Take a normal week in a mid-sized finance function. The close is on day four. A vendor has emailed new bank details. The auditors want the support for three reconciliations. The board pack needs commentary on a margin that moved for reasons nobody has written down yet. Most of the hours go on the first three, and the CFO's judgment is needed on the fourth. That is the case for AI in the office of the CFO, stated plainly: take volume off the people who should be spending their time on judgment, without loosening control.
This guide is the hub for our AI for CFOs series. Each section links to a fuller piece on the topic.
What AI in the office of the CFO is good at today
AI is reliable at a narrower set of tasks than most announcements suggest. It reads, classifies, matches and drafts well. It does not decide, approve or carry accountability, and it should not produce figures that have not come from the ledger.
| Area | What AI does reliably | What a person still does |
|---|---|---|
| Payables | Classify invoices and exceptions, suggest the cause of a match failure, flag likely duplicates, draft vendor queries | Approve payments, resolve disputes, approve vendor bank changes |
| Receivables | Suggest cash application, summarise a customer's history, draft reminders, rank the worklist | Agree payment plans, settle disputes, decide on credit holds |
| Close | Suggest reconciliation matches, flag unusual balances, draft flux explanations | Review, adjust, sign off reconciliations and journals |
| Reporting | Draft variance commentary from ledger figures, answer questions about the numbers | Decide what matters, approve what the board sees |
| Compliance | Track deadlines and changes, prepare checklists, chase owners | Prepare and sign filings, interpret rules, approve submissions |
Across the table, the pattern repeats. The software takes the first pass through high volumes of routine work and hands people the exceptions. We set out the operating model behind that in management by exception in finance.
Where finance teams actually are
On one point the surveys agree: use is widespread. They measure different things, so read them side by side rather than as one trend.
| Source | Sample | Finding |
|---|---|---|
| Gartner, November 2025 | 183 CFOs and senior finance leaders, May–June 2025 | 59% of finance leaders report using AI in their finance function; 91% report low or moderate impact initially |
| Gartner, June 2026 | Same 2025 survey | 84% have implemented or plan to implement AI; only 7% report a high or very high impact |
| KPMG, May 2026 | 1,013 senior finance leaders, 20 countries, March 2026 | Active use of AI in finance rose from 30% to 75% since 2024; 42% are fully assurance-ready for AI-enabled finance processes |
| The Hackett Group, March 2026 | Sample not disclosed | 33% of organizations already scaling AI in accounts payable; 77% cite a lack of AI talent |
| Deloitte CFO Signals, July 2026 | 200 North American CFOs at companies with US$1bn+ revenue | 19% of CFOs say they have greatest responsibility for AI governance |
Two points stand out for a CFO. Accounts payable is where most teams have gone first, which matches where the volume is. And the gap between use and measured impact is large. The cause is rarely the model. It is usually a pilot without a baseline, a process that was not redesigned around the tool, or a team that kept doing the old work alongside the new.
Choose use cases by volume, rules and risk
Start where three things are true. The work is high in volume. It follows written rules most of the time. And an error is caught before it costs money or reaches a regulator. Invoice exception triage, cash application suggestions, reconciliation matching and deadline tracking usually qualify. Judgement-heavy work such as forecasting assumptions or a revenue recognition position does not, at least not as the first project.
Before choosing, measure where the hours go. Most finance leaders are surprised by how much time sits in rework, chasing and reformatting rather than in the work itself. We describe a simple method in a time audit for the finance team and a scoring framework in AI use cases in finance.
Build on the systems you already run
A finance AI project should not start with replacing the ERP. Most of the value sits in a layer that reads from the systems already in place, shows the exceptions, assigns them to owners and keeps a record. The ERP remains the book of record.
This matters more than usual for SAP users. SAP's published maintenance strategy provides mainstream maintenance for SAP Business Suite 7, including SAP ERP 6.0, until the end of 2027, with optional extended maintenance from 2028 to the end of 2030 at a premium. Many finance teams will run an S/4HANA migration and an AI programme over the same period. A read-only layer above the ERP can carry across the migration without being rebuilt. We cover this in SAP ECC end of maintenance and AI in accounts payable on SAP.
The ERP vendors are adding their own AI features, and a CFO should use them where they fit. SAP has published an accounts payable assistant with agents for payment exceptions and bank account validation. Microsoft documents Copilot features in Dynamics 365 Finance for collections summaries and account reconciliation. The question is not whether to use them but where a cross-system view is still needed: across entities, across ERPs, or across finance and operational data. See AI in order-to-cash on Microsoft Dynamics 365 and multi-entity reporting across different ERPs.
Put the controls in first
Governance is easier to design before adoption than to retrofit after it. Settle four things before the first pilot leaves the test environment.
Access. Which tools may be used, by which roles, on which classes of data. Read-only by default.
Review. What must be checked by a named person before it leaves finance, whether that is a payment, a journal, a filing or a board paper.
Evidence. How prompts, inputs, outputs and approvals are logged, so that an auditor can follow what happened.
Change. Who approves a change to a prompt, a rule or a model that affects a control.
There are recognized frameworks to borrow from. NIST published its AI Risk Management Framework in January 2023 and a Generative AI Profile in July 2024. ISO/IEC 42001:2023 sets requirements for an AI management system. In the EU, the AI Act's obligations for general-purpose AI models have applied since 2 August 2025. The Digital Omnibus regulation of July 2026 moved the high-risk obligations to 2 December 2027 and 2 August 2028, depending on the category. On the audit side, PCAOB staff reported in 2024 that preparers' use of generative AI in financial reporting appeared secondary to operational and customer-facing uses. That is changing, and the conversation with the auditors should start before the first AI-assisted control goes live.
The series covers this in three parts: AI guardrails for finance data, an AI use policy for the finance function and AI and internal controls.
From pilot to production, and how to measure it
Most finance AI pilots stall for ordinary reasons. Nobody recorded the baseline, so nobody can say what improved. The success criteria were not agreed in advance. The pilot ran alongside the old process, so the team did both. Or no one owned the decision to scale or stop.
A pilot worth running has a fixed scope: one process, one data source, read-only access, a baseline measured beforehand, agreed success measures, a named control owner and a decision date. The measures should be ones a CFO already trusts: hours, error and rework rates, cycle time, and cash effects such as days sales outstanding or discounts captured. We set out the design in AI proof of concept in finance and the measures in measuring AI ROI in finance.
A twelve-month plan
| Quarter | Focus | Output |
|---|---|---|
| Q1 | Measure and govern | Time audit of the finance function; AI use policy approved; data classes and access agreed; auditors briefed |
| Q2 | First pilot | One high-volume process (for example AP exceptions or cash application) with a baseline, read-only access and a control owner |
| Q3 | Decide and extend | Scale or stop the first pilot on the agreed measures; start a second process in another area such as the close or compliance |
| Q4 | Embed | Controls documented and tested; roles and training updated; a quarterly report on hours, errors, cycle time and cash |
This plan is deliberately modest. A finance function that finishes the year with two processes running reliably under documented controls is ahead of most.
What AI in the office of the CFO will not do
It will not fix weak processes or poor data. An unreliable vendor master or a close without a reconciliation standard will produce faster errors, not fewer. Fix the foundations first or alongside.
It will not take accountability. The CFO still signs the accounts, the controller still owns the reconciliations, and the board still holds the finance function responsible for the numbers. AI changes how the work gets done. It does not change who answers for it. A separate piece on why finance teams distrust confident AI output covers that risk in more detail.
Frequently asked questions
What is AI in the office of the CFO?
AI in the office of the CFO is the use of AI tools in the finance function's own work, such as payables, receivables, the close, reporting and compliance. In practice it means software that reads, classifies, matches and drafts, while people review, approve and remain accountable.
Where should a CFO start with AI?
Start with one high-volume, rules-heavy process such as accounts payable exceptions or cash application. Measure the baseline first, give the tool read-only access to the existing system, agree the success measures and name a control owner before the pilot begins.
How many finance teams are using AI?
Gartner's survey of 183 finance leaders, published in November 2025, found 59% using AI in their finance function. KPMG's 2026 survey of 1,013 senior finance leaders reported active use rising from 30% to 75% since 2024. The surveys use different samples and definitions.
Why do so few finance AI projects show a large impact?
Usually because the pilot had no baseline, the process was not redesigned around the tool, or the team ran the old and new ways side by side. Gartner's June 2026 release reported only 7% of finance organizations seeing a high or very high impact.
Does AI in finance need new controls?
Yes. Any AI-assisted step that affects financial reporting needs a defined review, logged evidence and controlled changes to prompts and rules. Discuss the design with your auditors before the step goes live.
Do we need to replace our ERP to use AI?
No. Most of the value comes from a layer that reads from the existing ERP, shows exceptions and keeps a record, alongside the AI features the ERP vendor provides. The ERP remains the book of record.
Where CFOLogic fits
We work with CFOs to design and build AI-assisted tools and proofs of concept on top of the systems they already run, read-only by default, with a person approving anything that leaves finance. Public demos with fictional data show the approach: a P2P Command Center over a purchase-to-pay flow, a Compliance Command Center for multi-country deadlines, and Navigator for questions answered from the ledger. See also CFO Success Partners.