By the time a company calls us, they've usually already made the decision to do something with AI. What they haven't decided is what, or whether their organisation is in a position to absorb it. That gap between the decision and the readiness is where most projects fail.

Here are five patterns we see consistently. None of them are unusual. Most organisations going through their first AI project will recognise at least three.

1. Treating AI as a software purchase

The framing matters more than it seems. Companies that approach AI as a procurement exercise — evaluate vendors, pick the best one, buy licences, deploy — consistently underestimate the organisational work required. They end up with a product they can't integrate, data that doesn't fit it, and a team that wasn't involved in selecting it.

AI implementation is an operational change project that happens to involve software, not the other way around. The technology is often the least difficult part. Process redesign, change management, and team training take longer and require more internal capacity than most organisations plan for.

2. Starting with "what can AI do?" instead of "what are we trying to fix?"

AI capabilities are genuinely impressive right now. That makes it easy to start from the technology and work backwards — to ask what AI can do and then look for applications. This almost always leads to solutions that are technically possible but organisationally irrelevant.

The useful starting point is a specific, measurable problem. Not "we want to improve customer service" — but "we receive 400 support tickets per week, 60% of which are answered by the same 12 FAQ responses, and our average first response time is 14 hours." That framing immediately clarifies whether AI is the right tool and how you would know if it worked.

3. Underestimating data readiness

This is the most common reason AI projects stall after they've started. The assumption going in is that data exists and is usable. The reality is that data exists but is inconsistent, incomplete, or siloed in systems that don't communicate with each other.

We ask clients to estimate how long it would take to pull a clean dataset of their last 12 months of a specific process. If the answer involves manual exports from multiple systems and significant cleaning time, the data isn't ready. That's not a disqualifier, but it needs to be scoped and costed before implementation begins.

4. Not involving the people who will use it

AI systems designed without input from the people who will work alongside them tend to fail in predictable ways. They optimise for the wrong things. They create workflows that are technically correct but practically unusable. The team doesn't trust them and finds workarounds.

The operations team that processes invoices manually every month knows things about edge cases, exceptions, and informal rules that don't appear in any documentation. That knowledge is essential input to building a system that handles the actual process rather than a simplified version of it.

5. No defined metric for success

Without a defined metric agreed before work begins, you can't tell whether the project succeeded, and you can't make a rational decision about whether to continue investing in it.

The metric doesn't have to be complex. "Processing time for this task drops from 120 hours per month to under 20 hours" is clear, measurable, and directly tied to the business problem. It also creates accountability — if the system isn't delivering that improvement, something needs to change.

What this looks like in practice

The organisations that run successful first AI projects share a few characteristics. They start with a specific, bounded problem. They involve the people closest to the process from the beginning. They audit their data before making any technology decisions. And they treat the first project as a learning exercise rather than a transformation programme.

None of that requires a large budget or a dedicated AI team. It requires clarity about what you're trying to solve and a realistic view of what you're starting with.