The CFO’s guide to evaluating AI ROI

As a CFO who now builds AI systems, I see the same problem from both sides: technology teams struggle to articulate business value, and finance teams do not know how to evaluate AI investments.

The hidden costs nobody mentions

Most AI business cases dramatically underestimate implementation costs. They account for software licensing but forget about data preparation (typically 60–70% of project effort), change management, ongoing maintenance, and the opportunity cost of internal resources.

A practical ROI framework

I break AI ROI into four categories: direct cost savings (staff time, error reduction), revenue enablement (faster decisions, better forecasting), risk reduction (compliance, audit trail), and strategic optionality (platform for future capabilities).

Time to value matters more than total value

A project that delivers £50k of value in 3 months is often more valuable than one promising £200k over 18 months. Quick wins build confidence, fund further investment, and prove the approach works.

The maintenance question

Every AI system requires ongoing attention. Budget 15–20% of initial implementation cost annually for maintenance.

When AI is not the answer

Sometimes the best ROI comes from simpler automation — connecting existing systems, eliminating manual data entry, or fixing broken processes.

The board presentation

When presenting AI investments to the board, lead with the business problem, not the technology. Show the cost of inaction. Present conservative estimates with clear assumptions.