We can't tell if AI is actually working for us

Common questions from Omaha leaders

Why is AI ROI so hard to measure?

Because most teams measure activity (token spend, login frequency, prompts per day) instead of outcomes (cycle time, error rate, customer outcomes, employee capacity). Activity metrics get gamed within a week and produce noise about whether AI is working. Outcome metrics — measured against pre-AI baselines on the same workflow — produce signal.

What's the realistic ROI range for AI?

Wide. Deloitte's Q4 2024 GenAI survey found 74% of respondents say their most advanced GenAI initiative is meeting or exceeding ROI expectations, with 20% reporting ROI in excess of 30%. But the McKinsey 2025 data is sobering: only 39% of organizations report any EBIT impact at the enterprise level, and only ~6% are 'AI high performers' attributing 5%+ EBIT impact. The narrow gap between 'we use AI' and 'AI changed our business' is wide for most mid-market companies.

Should we kill our AI initiative if we can't measure ROI?

Almost never the right move. The more common diagnosis: you're measuring the wrong thing. Switch to workflow-level outcome metrics (cycle time on the specific workflow AI is touching, error rate before vs. after, hours of human time freed) and watch what actually changes. If those numbers don't move after 90 days, then yes — but the diagnostic almost always finds 'we never measured the right thing' before it finds 'AI doesn't work here.'

What about hard EBIT impact — when does that show up?

Usually only after workflow redesign at scale. McKinsey 2025 found AI high performers are nearly 3x more likely to have fundamentally redesigned individual workflows. Tool adoption (Copilot for the team) produces small-to-moderate gains. Architectural adoption (workflow redesigned around AI as a participant) is what produces EBIT impact. The gap between the two is usually 12–24 months and a different operating model.

How do we set realistic ROI expectations with leadership?

Three levels. Tier 1 (3 months): one workflow shipped with measurable cycle-time or error-rate improvement. Tier 2 (12 months): cumulative gains across multiple workflows, measurable employee capacity reallocated. Tier 3 (24+ months): EBIT-level impact tied to workflow redesign at scale. Promising tier 3 in 90 days is how trust gets eroded.