The Person Who Already Does It Well

Every AI project I have ever seen succeed in a small business had one thing in common, and it is not the thing you would expect.

It was not the model. It was not the budget. It was not the founder's enthusiasm or the team's technical literacy. The single most consistent predictor of success was whether the team had identified, very early, the one person who already did the task well, and built the AI around their judgment.

I want to make this concrete, because it is the most under-used lever in AI deployment inside SMBs.

In every team I have ever worked with, there is one person who is unusually good at the task the AI is being asked to do. The person whose customer emails get the most positive responses. The person whose pitches close. The person whose code reviews catch the bugs everyone else misses. The person whose research notes are the ones the founder actually reads.

They are not always the senior person. They are not always the loudest. Sometimes they are the most junior, sometimes the most senior, often someone in the middle who has been quietly excellent for years without anyone naming it. They are the person whose work, if you could clone it, would lift the whole team. They are the local genius at this specific task.

If you are about to build an AI feature, your first job is to find this person. Your second job is to spend a lot of time with them.

Almost nobody does this. There are three reasons, and they're all painful to admit.

The local genius doesn't know they're the local genius. They have been doing the task for so long that it feels obvious. They cannot articulate what they do differently. They will tell you, sincerely, that anyone could do it. They are wrong, but they believe it, and so does everyone around them.

The leadership doesn't know they're the local genius either. Performance metrics inside SMBs are crude. The local genius produces better outcomes on each individual task, but their work doesn't show up in dashboards. The person who produces three times the volume gets the recognition. The person who produces work three times better often does not.

The team running the AI build prefers to start from scratch. It is intellectually easier. You can read papers, evaluate models, design prompts. You can build something that feels new. Sitting with a forty-seven-year-old customer success manager for three days to extract how she thinks about angry-customer emails feels less like AI work. It is, in fact, exactly the work that determines success.

Once you have identified the local genius, you have to extract their judgment in a form the AI can use. This is a craft, and it is not technical. It is interview craft, observation craft, and writing craft.

You sit with them. You watch them work. You ask them, in detail, what they're thinking when they make a choice — not in the abstract, but in the moment, on the specific email or proposal or decision. You write down what they say, and then you show them the writing and ask them to correct it. You repeat this for days, not hours, until you have produced something they would sign off on as a fair representation of how they do the work.

That document — that articulated judgment — is the most valuable artefact in your AI deployment. It is the difference between a generic system that produces generic output and a system that produces output indistinguishable from your best person.

The first thing that changes is the output. AI that has been trained on the local genius's judgment produces work that feels recognisable to the team. They critique it the way they would critique an early draft from a junior — this is the right shape but the tone is off here — rather than the way they would critique a generic AI output — this just isn't us. The system becomes part of the team's standards rather than a foreign object inside them.

The second thing that changes is adoption. The team adopts the tool because they recognise it. They see their best person in the output. They trust it the way they trust her, because in a real sense it is her work, scaled. Compare that to a tool that has been built from generic best practices and you can feel the difference within the first week of rollout.

And the third thing that changes is the local genius's career. They have become, formally, the standard-setter for that workflow in the business. Their value goes up. They get recognised for what they have always been doing. They stay. They mentor. The AI deployment, instead of devaluing your best people, has the opposite effect — it makes the depth of their judgment visible, transferable, and economically central to the company.

Find the local genius. Spend the days extracting their judgment. Build the AI around it. It is the cheapest and most overlooked move in the entire stack.