There's a lot of noise about AI right now, and almost none of it tells a business owner what to actually do. So let's lower the volume.

Most coverage of AI sits at one of two extremes. On one end: breathless promises about how everything is about to change overnight. On the other: vague anxiety about jobs, ethics, and machines that don't quite work the way the headlines suggest. Neither end is useful if you run a real business, employ real people, and have a finite budget for experimentation.

The middle, the part that is useful, is much quieter, and it sounds something like this: a small number of tasks in your business are eating disproportionate time, costing disproportionate money, or producing inconsistent results. Those tasks are where AI earns its keep. Not all of them. Not even most of them. But the ones it does are often transformative, and you don't need a technical background to find them.

The "expensive, repetitive, inconsistent" filter

When we work with a client on AI, we don't start by picking a model. We start by walking through their week and listening for three things:

  • What's expensive? A senior person doing something a junior could, if only the junior had the same context.
  • What's repetitive? Work that follows roughly the same shape every time, but doesn't quite fit a rule.
  • What's inconsistent? A judgement call where two competent people reach different answers, and there's no audit trail of how they got there.

If a task hits two of the three, it's worth a closer look. If it hits all three, it's almost certainly a candidate for AI augmentation, and almost certainly worth doing properly rather than tactically.

The most expensive AI is the AI you never used. The second most expensive is the AI you bolted onto something that wasn't broken.

Augmentation, not replacement

The word matters. We don't talk about replacing people with AI because, in practice, that's almost never the right framing. The work we ship keeps the human firmly in charge of the decision and uses AI for the heavy lifting around it, reading the inputs, structuring the comparisons, flagging the confidence, citing the sources.

This isn't just an ethical choice. It's an operational one. A system that operates as a black box is a system nobody trusts, which means a system nobody uses, which means a system that doesn't return its investment. Trust is a feature, and the most undervalued one in the market right now.

What "doing it properly" actually means

Three commitments, very practical:

One: pick the right task. The single biggest mistake in AI deployments is picking the wrong target, usually something flashy rather than something that compounds. Spend more time here than feels reasonable.

Two: test the option, not the brochure. The model that wins a benchmark is rarely the model that wins your task. We routinely test 3–5 systems against the actual job, with the actual data, and the answer is almost never the one we'd have guessed.

Three: design for trust. Confidence ratings. Source citations. Human-in-the-loop checkpoints. Audit trails. The boring stuff nobody puts on a launch slide is exactly the stuff that decides whether the tool gets used a month later.

If you take one thing away

You don't need to learn how AI works under the hood. You don't need to hire a team. You don't need to pick a vendor first.

What you need is to look at your business with the three-part filter, expensive, repetitive, inconsistent, and find the one task that hits all three. That's your starting point. Everything else, model choice, deployment, governance, change management, flows downstream from getting the target right.

No technical background required.
Just a willingness to work smarter.

If reading this has surfaced a candidate task, we'd be glad to talk it through. We won't sell you anything you don't need; often the conversation ends with us telling you the problem isn't quite ready for AI yet, and what would make it ready. That's part of the work.