Somewhere between the board presentation and the first sprint, most AI projects lose the thread. Not because the technology doesn't work. Not because the team isn't capable. But because no one agreed on what problem they were actually solving.
The single most reliable predictor of AI project failure is a vague problem statement dressed up as an ambitious vision.
The scoping trap
When we audit failed AI initiatives, the pattern is almost always the same: the project was defined by its technology, not its outcome. "We want to implement an AI chatbot." "We want to use machine learning for our data." These are solutions looking for problems — and they reliably produce expensive pilots that never make it to production.
The companies that succeed with AI start from the opposite direction. They identify a specific, measurable pain point — a process that takes 40 hours a week that shouldn't, a decision that requires three people to coordinate that could be automated, a piece of customer communication that's inconsistent because it's written by hand. Then they ask: could AI address this? What would "solved" look like? How would we measure it?
Three questions before any AI project
- What specific task or decision are we trying to improve — and by how much?
- What does the before and after look like for the person doing this work today?
- How will we know in 30 days if this is working?
If you can't answer all three with specificity, you're not ready to build. You're ready to scope — which is a different, and more valuable, exercise.
What good scoping looks like
Good AI scoping is a structured audit of your workflows to find the highest-leverage opportunities. It's not a brainstorm. It's not a vendor demo. It's a methodical process of mapping how work actually flows through your organisation, identifying where AI can genuinely reduce friction or improve quality, and ranking those opportunities by impact and feasibility.
The best AI implementations are almost boring to describe. They just make something faster, cheaper, or more consistent than it was before.
The output of a good scoping exercise isn't excitement — it's clarity. A written spec, a defined outcome, and a build plan with a fixed timeline and cost. That's when AI projects succeed.