AI Pilots: Why Organisations Keep Picking the Wrong Things

 Over the last two years we’ve watched hundreds of organisations begin their AI journey. Almost all of them follow the same pattern: they get excited, gather a small team, and immediately start hunting for a “great pilot”. Something new and clever. Something that proves they’re ahead.

And almost every time, they run straight into the same two risks.

The first is Commodity Risk. This is where a team picks a problem that feels unsolved but is actually being tackled by everyone else, including OpenAI, Google, Anthropic and every VC-funded startup you’ve never heard of. They spend months designing, prototyping and polishing something only to discover that ChatGPT, Claude or Gemini quietly released a new feature that does the whole thing out of the box.

This happens far more often than people realise.

The second is Overtake Risk. This is where a team genuinely finds something interesting, a future-facing idea, but the capability of consumer AI moves so quickly that, by the time they finish building their customised version, the foundation models have already leapfrogged it. They’ve built a bespoke version of something that is now a free feature in the apps their staff already use.

And they’re back at square one.

So you get this organisational hesitancy: “Maybe we should wait. Maybe we should see what the next model does.”

And sometimes that’s absolutely the right instinct. In fact, for a huge amount of “AI ideas”, waiting six months is exactly the right move, because the capability will either become commoditised or simply get absorbed into standard tools.

But, and it’s a BIG BUT! There’s a problem.

If all you do is wait, you never actually explore the frontier. You never develop fluency. You never stumble across the niche areas where your business actually has something unique, the parts that rely on your workflows, your data, your processes, your domain expertise. Those are the places where AI can deliver real, defensible advantage.

This is the frustrating paradox of AI adoption:

Move too early, and you risk wasting time on something soon made obsolete.

Wait too long, and you miss the window where you could have built something no-one else can.

This is why picking your pilots matters.

The real opportunities aren’t in building generic assistants or replicating the next chat interface. They’re in weaving AI into the parts of your organisation that only you understand, the messy processes, the specialist judgement, the flows of knowledge that outsiders never see. These aren’t areas foundation models will replace any time soon, because they don’t have your context, your people or your data.

And that is your defensible layer.

So rather than chasing “the clever idea”, start with your strengths. Start with the work your organisation knows better than anyone else. Start with the pain points that are genuinely yours.

That’s where you’ll find pilots that survive the next model update.

That’s where you’ll avoid the commoditisation trap.

And that’s where you’ll create something that actually moves the needle.