Five questions every executive should ask before investing in AI

Before you buy another platform, run another pilot, or hire another data scientist or AI Adoption Strategist, answer these five questions first.

The research on why AI initiatives fail is unusually consistent. RAND, BCG, McKinsey and MIT are studying different companies with different methods, and they keep landing on the same handful of causes — almost none of which are about the model. These five questions map onto those failure modes. If you can answer all five before you sign anything, you've ruled out most of the ways this goes wrong.

1. What decision or workflow does this change — and who owns that outcome today?

RAND's top-ranked cause of failure is that business and technical stakeholders never align on what problem is being solved, so the model ends up optimized for a metric no one acts on. If you can't name the specific decision or task that will be done differently, and the person accountable for that result, you don't have a project yet — you have a technology looking for a home.

2. Do we have the data — and the legal right to use it — at the quality this needs?

Inadequate or inaccessible data is one of RAND's leading failure causes, and it's typically discovered after the budget is committed. Gartner names data quality as a top reason generative-AI projects are abandoned after proof of concept. Ask what data the system needs, whether it exists in usable form, who owns it, and whether you're permitted to use it for this purpose.

3. Where does this run after the demo — and who maintains it then?

A proof of concept proves almost nothing about production. RAND identifies missing infrastructure — no established way to deploy, monitor and maintain a model — as a distinct failure cause. The 2015 paper "Hidden Technical Debt in Machine Learning Systems" made the point early: the model is a small fraction of the system that has to surround it. Know which team owns the thing in month six.

4. How will we measure value — and what's the kill criterion?

Only about 13% of organizations have a clear KPI structure for tracking the value of their generative-AI work, per McKinsey. Gartner cites "unclear business value" as a leading cause of abandonment. Define the metric that says this is working before launch — and the threshold at which you stop.

5. Whose work changes — and who's accountable for adoption?

BCG's 10–20–70 rule holds that 70% of the effort in a successful AI initiative is people and process, not algorithms or infrastructure. The MIT analysis found pilots that combined internal domain knowledge with outside expertise succeeded far more often than technology-led builds. Someone senior has to own the change in how people work — not the model, the behavior.

If the answers aren't there

The gap isn't in your vendor shortlist. Fix that first; the technology decision is the easy part.