Finance departments are not losing interest in artificial intelligence, but they are attaching more conditions to the next round of spending as systems move from document processing into decisions affecting approvals, exceptions, and payments.

A study commissioned by Basware and conducted by Forrester Consulting found that 76% of surveyed finance and accounts payable decision makers expect to increase AI investment during the next 12 to 24 months. However, 68% said further spending would require demonstrable returns.

The survey covered 231 enterprise decision makers and finance professionals in the UK, France, Germany, and the United States during the first quarter of 2026. It is a relatively small, vendor funded snapshot rather than an independent census of corporate finance, and its conclusions should be read within that limitation.

Even so, the results reinforce a pattern visible across enterprise AI programmes. Organisations are moving beyond whether a model can perform a task and asking whether it can operate dependably inside a controlled process with defined authority, records, escalation, and financial accountability.

Accounts payable offers measurable starting points

Basware’s study found that 67% of respondents had moved beyond pilots into targeted AI use within day-to-day accounts payable operations. Invoice data capture was the most frequently identified area of benefit, followed by invoice matching and fraud or risk analysis.

Those uses are not a sudden change in finance technology. Optical character recognition, rules based matching, workflow automation, and anomaly detection have operated in accounts payable for years, while newer models can process more varied documents and communications.

Agents extend the proposition by investigating exceptions, recommending actions, or working across several systems. The economic attraction is that outcomes can be counted through processing time, touchless invoice rates, duplicate payments, exception volumes, early payment discounts, fraud losses, and transaction cost.

Return expectations in the survey were relatively restrained. Only 7% expected payback within six months, while 20% anticipated six to 12 months and 35% expected value to emerge over 13 to 24 months. Those periods allow for integration, data work, process redesign, control testing, and adoption.

Autonomy changes the finance control environment

An AI system extracting an invoice number operates within a different risk category from one deciding whether an exception can be resolved or a payment approved. As authority expands, finance teams need transaction limits, segregation of duties, evidence requirements, human review, and clear conditions under which a system must stop.

Only 39% of respondents reported operating an AI centre of excellence at scale, while 46% believed their organisations had achieved an effective balance between governance and innovation. Policies and executive support are therefore not being translated consistently into day-to-day operating models.

Basware describes its preferred approach as governed autonomy, dividing authority between advisor, collaborator, and operator roles. The terminology supports its own product strategy, although the underlying principle is widely applicable: systems should gain authority through measured performance rather than receiving extensive permissions at deployment.

Auditability becomes particularly important when an automated process combines an invoice, purchase order, supplier history, employee communication, and fraud score. A later reviewer must be able to establish which data was used, what the system recommended, whether a person intervened, and why the final action was allowed.

European regulation adds pressure from another direction. The EU’s VAT in the Digital Age programme and national electronic invoicing mandates are increasing the volume of structured transaction data while changing reporting requirements. Better data can support automation, although platforms must keep pace with differing national rules and implementation schedules.

The survey found that 65% of respondents believed major or urgent improvement was needed to adjust to new financial regulation, while 64% prioritised stability and compliance over raw innovation when choosing AI. Those preferences are unsurprising in a function responsible for cash, tax, suppliers, audit, and financial statements.

Finance departments appear willing to increase AI spending where benefits can be demonstrated, but a successful pilot is no longer sufficient. The next phase will favour systems that act within enforceable limits, explain their work, and remain dependable when an invoice is incomplete, the supplier is unfamiliar, or the payment should not proceed.