If most organisations can't show a return on AI — only 39% report any EBIT impact, and just 4 of the top 50 banks saw realised ROI in 2025 — then measuring ROI is itself the competitive skill.
Why AI ROI is hard to see
- Benefits are often diffuse (minutes saved per task) and hard to roll up.
- The full cost — inference, monitoring, prompt upkeep, integration — is rarely tracked against the benefit.
- Perception misleads: METR showed people feel faster while being slower. Vibes aren't ROI.
A practical approach
- 1.Pick one workflow with a baseline — current cost, time and quality, measured before AI.
- 2.Instrument the AI version — its true running cost and its measured output.
- 3.Compare like for like, including the maintenance tail, not just the launch.
- 4.Tie it to a P&L line — revenue, cost or risk — or admit it's a bet, not a return.
If you can't draw a straight line from the AI feature to a number that matters, you don't have ROI — you have a hope.
Sources
- McKinsey — The State of AI 2025
- METR — Developer productivity RCT
- Coinlaw — AI in Banking Statistics 2025
Written by ivector
Start a project →