How to tell if AI actually fits your business

Every week, someone asks us to "add AI" to their business. Almost as often, the most useful thing we can say is: not yet, or not here.

That's an unusual sales pitch for an AI company. But the fastest way to waste a budget is to point a model at a problem that didn't need one — and the AI industry has gotten very good at encouraging exactly that.

So before we write a line of code, we run a short, honest test. Here's how it works.

The wrong question

The wrong question is "where can we use AI?" Ask it and you'll get a list of places AI is possible — which is everywhere, and therefore useless.

It leads to demos that impress in a meeting and quietly get abandoned three weeks later, because they automated something nobody actually struggled with.

The right question: where's the seam?

The better question is: where does AI change the shape of the work, not just its speed?

We call that the seam — the single point in a workflow where the bottleneck isn't effort, it's something only judgment, language, or pattern-recognition can unblock. A few signs you've found one:

• A person spends hours turning messy, unstructured input (calls, documents, emails) into structured output.
• The task is repetitive but not rule-based enough for ordinary automation to have already solved it.
• The cost of the work isn't the doing — it's the waiting, the queue, the "I'll get to it tomorrow."

When the seam is real, AI doesn't just make the task 20% faster. It removes the task.

Three signs AI genuinely fits

1. There's language or perception in the middle. Transcribing a consultation, reading a contract, triaging a support inbox — work that lives in words or images is where modern models earn their keep.
2. You have the data, or you generate it as a byproduct. If the raw material already flows through your business, you're most of the way there. If you'd have to manufacture it, be skeptical.
3. "Good enough, reviewed by a human" is acceptable. AI is probabilistic. If a 95%-correct draft that a person checks is valuable, you're in great shape. If you need 100% with no human in the loop, tread carefully.

Three signs it doesn't (yet)

• The real problem is a process problem. If the workflow is broken, AI will just break faster. Fix the process first; sometimes that's the whole project.
• A spreadsheet or a script would do. If deterministic rules cover the case, you don't need a model — you need an afternoon.
• Nobody owns the outcome. If no one inside the business is accountable for the result the AI produces, it won't get adopted, no matter how good it is.

The hard part of AI isn't the model. It's the ten thousand product decisions that turn a model into something people use every day.

How we run discovery

Our first engagement with anyone is a discovery call, not a build. We map the workflow, find the seam (or tell you there isn't one yet), and put a number on the payoff. You leave with a concrete plan and a clear answer — even when the answer is "don't."

If it does fit, the next step is small: the smallest thing that works, in production, in front of real users, fast. Everything compounds from there.

If you're staring at a workflow and wondering whether AI belongs in it, that's exactly the conversation we like to have. Tell us what you're working on — we'll tell you straight.