An AI-first finance function is not finance with a few AI tools. It is a function redesigned so that machines do most of the transaction processing and first-draft analysis, and people concentrate on judgment, business partnering, control and exceptions.
What changes
- Processes: designed for straight-through processing with exception queues, not manual steps with occasional automation.
- Roles: fewer preparers; more reviewers, analysts, data and AI specialists, and controllers who design and test AI-enabled controls.
- Structure: shared services and centers of excellence shrink or shift toward exception handling and AI operations.
- Data: a governed finance data layer becomes core infrastructure.
- Controls: designed around agents from the start: limits, evidence, monitoring.
- Metrics: unit cost per transaction, touchless rates, close days, forecast accuracy, and decision speed.
How large companies are framing it
IBM has reported productivity savings from deploying AI across its own operations, including finance. In 2025 earnings remarks filed with the SEC, its CFO said productivity is a key driver of margin expansion "as we deploy AI at scale across IBM in areas including finance, supply chain, sales, HR, service delivery and customer support", and its CEO said IBM expected $4.5 billion of annual run-rate savings exiting the year. These are company-reported figures, not audited AI benefits, but they show how an AI-first operating model is described to investors: as run-rate productivity tied to margin.
A three-horizon roadmap
- Automate: transaction processing and reconciliations with AI and agents under controls.
- Augment: AI-assisted analysis, forecasting and reporting for every finance professional.
- Transform: redesign roles, structure and the service model around humans and agents.
Pitfalls
Redesigning roles before the technology works; removing experienced people who are needed to supervise AI; and letting the data foundation lag behind the ambition.