Start with the process, not the tool
AI works best in finance where the process is already well understood: the inputs are known, the rules are documented and the outcome can be checked. Where a process is undocumented or inconsistent across entities, AI tends to automate the inconsistency. The first step is usually to standardise and document the process, which also improves it before any automation is added.
Where AI helps finance teams today
- Accounts payable. Reading invoices in any format, matching them to purchase orders and receipts, flagging duplicates and routing exceptions to the right person.
- Accounts receivable and cash application. Matching incoming payments to open invoices, even with incomplete remittance details, and suggesting the next action on overdue accounts.
- Record to report. Preparing reconciliations, explaining variances in plain language and drafting commentary for the month-end pack.
- Controls and audit. Testing whole populations of transactions instead of samples, highlighting unusual entries and assembling evidence for auditors.
- Planning and forecasting. Producing a first forecast from history and drivers, so analysts spend their time on judgement rather than data gathering.
From pilots to agents
Many finance teams are moving from single AI features to AI agents that complete a sequence of steps, such as receiving an invoice, checking it, posting it and chasing a missing approval. Agents can take on more of the routine work, but they also act with more independence. That makes clear limits, approvals and audit trails more important, not less.
Keep people accountable
Finance leaders remain accountable for the numbers, whatever produced them. A sound design keeps a named person responsible for each process, sets thresholds above which a human must approve, and makes it easy to see why the system made each suggestion. Auditors will ask how an AI-assisted process is controlled; the answer should already be documented.
Controls to build in from the start
- Approval thresholds: automatic handling below agreed limits, human approval above them.
- A complete audit trail: what the system saw, what it suggested, who approved it and when.
- Segregation of duties that still holds when part of the work is automated.
- Data protection: clear rules on which data may be sent to which AI service, and where it is processed.
- Ongoing monitoring: exception rates, error rates and drift, reviewed every month.
Skills matter as much as software
AI changes the finance role rather than removing the need for qualified people. Teams need professionals who understand accounting and controls, can judge the exceptions the system raises, and can explain the process to auditors. That is why many organisations pair AI adoption with a dedicated finance capability, often in a capability centre, where qualified accountants and auditors run the process and improve it continuously.
A practical first 90 days
- Days 1-30: choose one high-volume process, such as invoice processing; document it, measure today's cycle time, error rate and exceptions, and agree the controls.
- Days 31-60: introduce AI assistance with human review of every output; compare results against the baseline.
- Days 61-90: set approval thresholds for automatic handling, put monthly monitoring in place, and decide whether to extend to the next process.
Our finance and audit practice runs accounts payable, receivable, record to report and audit support with ACCA-qualified professionals, and introduces AI-assisted steps with the controls above. Finance leaders can read more on how we work with CFOs.