Most Government AI Pilots Should Fail
Why Safe Bets Are the Riskiest Strategy in Government AI
Government AI pilots are designed to succeed on paper while delivering nothing in practice. If your pilot portfolio has a 100% success rate, you're not innovating - you're wasting money on theatre.
The Success Theatre
Government AI pilots have an extraordinary success rate. Almost all of them "succeed" - on paper. The pilot report lands on someone's desk, ticks all the boxes, and then nothing happens. The technology never makes it to production. The vendor moves on. The problem remains.
This isn't a technology problem. It's an incentive problem. Pilots are designed to succeed because failure has consequences and success doesn't require actual change. A pilot that "proves the concept" is a success even if the concept never gets implemented.
Pilot Health Check
Run the diagnostic. Six vital signs that tell you whether your AI pilot is alive or on life support.
Does the pilot have a defined success metric?
Is there a named decision-maker who will kill it if it fails?
Has anyone used the output in a real decision yet?
Is the team building toward production or just proving a concept?
Does the vendor have skin in the game (outcome-based pricing)?
Could you explain what the pilot achieved to a minister in one sentence?
The Sunk Cost Machine
The hardest decision in government AI isn't starting a pilot. It's killing one. Every month that passes makes it harder to pull the plug. "We've already invested £200k." "The vendor says they're close." "It'll look bad in the annual review."
The money spent is gone either way. The question is whether you keep spending more on something that isn't working, or redirect those resources to something that might. Most departments choose to keep spending. It's the safer career move.
The Silent Death of a Pilot
Watch £600k disappear one 'let's give it another month' at a time. How long will you let it run?
You're overseeing a 12-month AI pilot. Budget: £600k. The vendor is confident. Your job: decide when enough is enough.
The Portfolio Approach
Venture capital figured this out decades ago. Most investments fail. That's the point. You spread your bets across a portfolio, knowing that the few that succeed will more than pay for the failures. Government AI needs the same thinking.
A portfolio with a 100% success rate isn't a sign of good management. It's a sign that you only picked safe bets - and safe bets in AI mean incremental improvements that don't change anything. The real transformations come from the risky bets that occasionally pay off spectacularly.
The Failure Portfolio
You have £2M. Pick your pilots. Then see what safe bets actually deliver vs. taking real risks.
Ready to rethink your pilot strategy?
We help government teams design AI pilot portfolios that balance risk with genuine transformation. Let's talk about what your portfolio should look like.
Get in Touch