Why most AI ROI calculations are wrong
Most AI ROI numbers you see in vendor pitches and conference keynotes are wrong. Not because the math is bad, but because they measure the wrong things.
Common mistakes:
- Counting only cost savings. "We'll need fewer analysts." This is both politically toxic and usually wrong. AI augments capacity — it rarely eliminates headcount in knowledge work.
- Using vendor benchmarks. "Customers see 40% productivity gains." Maybe — in a controlled demo. Your firm has compliance requirements, legacy systems, and change management overhead that the benchmark does not account for.
- Ignoring implementation costs. The licence is $40/user/month. The real cost includes training, workflow redesign, governance setup, and the productivity dip during the learning curve.
The ROI calculation in your readiness report needs to be conservative, specific, and based on your firm's actual numbers. Leadership will interrogate the assumptions. If they hold up, you get funded. If they don't, you lose credibility.
