I’ve been watching companies pour millions into AI projects for decades now. The promise is always the same: “This will revolutionise how we make decisions.” The reality? Most of these projects end up as expensive proof-of-concepts that never see production.
Here’s what I’m seeing go wrong – and more importantly, what actually works.
The Problem: Falling in Love with the Algorithm
Recently, I spoke with a claims director who’d spent 18 months building a sophisticated machine learning model to predict fraud. The model was technically brilliant with high accuracy in testing. But it’s been sitting unused for six months because nobody could figure out what adjusters should actually do with a “73% probability” score.
Decades ago, I spoke with a marketing director who had a brilliant model to predict that someone was “85% likely to churn” but had no tools or insights to retain them. Predictably, discounts were the easy answer, leading the industry to a race to the bottom. (See also: Commercial Churn Modelling.)
This is the classic mistake: starting with the technology instead of the decision.
The Insight: Start with the Action, Not the Algorithm
The most successful AI implementations I’ve seen follow what I call the “Decision-First Principle”:
- Map the actual decision – What choice does someone need to make? By when? With what constraints?
- Identify the decision bottleneck – Is it lack of information, too much information, or unclear trade-offs?
- Design the intervention – What would change the quality or speed of that decision?
- Then build the analytics to support that intervention
The Action: Three Questions Before Your Next AI Project
Before investing in any AI initiative, ask:
- Who makes the decision this AI is supposed to improve?
- What would they do differently if they had perfect information?
- How will we measure whether decisions actually improved?
If you can’t answer all three clearly, you’re not ready for AI – you’re ready for a conversation about decision-making. (Give me a call – I can help.)
The Impact: From Proof-of-Concept to Profit Center
Companies that start with decisions rather than data science see dramatically different outcomes. One client increased their underwriting speed by 40% not because they built a better model, but because they redesigned the decision process around what underwriters actually needed to know.
The AI was still sophisticated - but it was sophisticated for a purpose.
The Bottom Line: Great AI Projects Look Simple
The irony? The best AI projects often look deceptively simple. That’s because they solve real problems rather than interesting technical challenges.
What’s your experience with AI projects that worked vs. those that didn’t? I’m curious about the patterns you’re seeing.
About me: I help organisations turn complex data into clear decisions and commercial outcomes. My focus is on enabling better decision-making and unlocking new value through data-driven innovation – especially where the stakes are high and the problems are difficult and poorly defined.
Follow me on LinkedIn for more insights.