You Tried AI. It Didn't Work. Now What?

I know why you’re reading this

You’re here because you already tried. You spent €30K, €80K, €200K. You got a demo that looked great. Then the vendor disappeared, or the model broke, or nobody used it.

You’re not sure if the problem was AI, the vendor, or you. Let me answer that: it wasn’t you. It was the approach.

I’m not going to tell you to “try again with the right partner” - that’s what every consultant says. I’m going to tell you exactly why it failed, how to diagnose it, and whether it’s worth trying again.

The autopsy: What actually went wrong

Most failed AI projects die in one of four ways. I’ve audited enough of them to recognize the patterns:

Death #1: The demo theater. The vendor showed you a demo that worked on curated data. Your data is different. The model was never going to work on your real data. The vendor knew this. You didn’t.

Death #2: The science experiment. The project was “exploratory.” No success metrics, no deadline, no production plan. It drifted until someone lost interest. Nobody declared it a failure - it just faded away.

Death #3: The orphan. The AI worked. Nobody used it. The UI was clunky, the workflow wasn’t integrated, the team didn’t trust it. The vendor delivered “the model.” The model wasn’t the product.

Death #4: The money pit. The project was supposed to cost €50K. It’s at €120K and counting. Every milestone reveals new “complexity.” The vendor keeps billing. You keep paying because you’re already invested.

Which one sounds like yours?

The question you’re actually asking

You’re not asking “does AI work?” You’ve seen it work - in demos, in competitors, in the news.

You’re asking: “Can I make it work? Can my company? With my data and my budget and my team?”

That’s a different question. And it deserves a different answer than “yes, AI is transformative.”

The honest answer: it depends on your data, your problem, and your team. Not on AI.

Why your first attempt failed (and it wasn’t your fault)

Most AI vendors sell you AI. They don’t sell you a production system. They don’t audit your data first. They don’t assess your team’s ability to maintain it. They build a model, show you it works on clean data, and call it success.

That’s not a failure of AI. It’s a failure of the approach. The approach that skips the foundation.

Your data wasn’t ready. Your team wasn’t ready. Your success metrics weren’t defined. None of that is your fault - nobody told you to check. The vendor certainly didn’t.

Should you try again? A decision framework

Don’t try again if: your data is still a mess, your team hasn’t changed, and you’re doing it because of board pressure. Same conditions = same result.

Try again if: you’re willing to start with an audit (not a project), you have a specific business problem (not “we need AI”), and you’re willing to walk away if the audit says no.

The difference between your first attempt and a successful one isn’t a better vendor. It’s a better starting point. Start with diagnosis, not with building.

How to not get burned again

Rule 1: No one builds anything until you’ve audited your data. If a vendor wants to start building before auditing your data, walk away.

Rule 2: Define success before you start. “Reduce manual review by 40%” not “explore AI capabilities.” If the vendor won’t commit to a metric, they’re not confident in their work.

Rule 3: You own the code, the data, and the documentation. If the vendor keeps any of it hostage, you’re building a dependency, not a capability.

Rule 4: Phased engagement. Audit first. Build second. Deploy third. If a vendor wants a single €200K contract for everything, they’re not accountable to results - they’re accountable to signing.

Rule 5: If they can’t explain what they’re building in plain language, they don’t understand it well enough. Or they’re hiding something.

What I do differently

I start with an audit. Two weeks. You get a diagnosis: what’s broken, what AI can do, what it costs, what to fix first.

If the audit says AI won’t work for you, I tell you. I’ve turned down projects. Not because I don’t want the work - because building the wrong thing is worse than building nothing.

If the audit says AI will work, we build in phases. You see results at each stage. You can stop at any point. No blank checks.

The real cost of not trying again

Your competitors are figuring this out. Not all of them - 80% will fail too. But the 20% who get it right will have a cost structure and capability advantage you can’t match.

The question isn’t “should we try AI?” - it’s “can we afford to be the company that didn’t figure it out while everyone else did?”

But trying the same way will get the same result. Try differently. Start with an audit.

FAQ

How do I know you’re not just another vendor? You don’t. But I’ll let you audit me the same way I audit your data. Ask me hard questions. If I can’t answer them clearly, don’t hire me.

What if we already have a failed project - can you salvage it? Maybe. I’ll audit it and tell you. Sometimes the foundation is there. Sometimes it’s cheaper to start over.

What if my board has lost faith in AI? Show them this article. Or let me present the audit findings to them directly. Data rebuilds faith faster than promises.