Where Does AI Fit? Start With the Bottleneck, Not the Technology

Every rescue project I have been called into started with the same sentence, delivered with real conviction: "We want to use AI."

It is a solution shopping for a problem. And it is the single most expensive way to answer the question that actually matters, which is: where does AI fit in this business?

Start with the bottleneck, not the technology

The companies that get genuine value from AI do something far less exciting than issuing an AI mandate. They start with the bottleneck.

Not the strategic-sounding one that goes in the board deck. The real one. The thing that is slow, expensive, or quietly consuming three people's weeks. The report that takes four days to assemble. The intake queue where every request waits on one overloaded person. The reconciliation nobody wants to own.

Only once that is named do you ask what fixes it. Sometimes the honest answer is AI. Frequently it is a cleaner data source, a rewritten process, or deleting a step that no one could justify when finally asked. I own and act as CTO for three companies, so I make this call in my own businesses every week. I have killed more AI ideas than I have shipped. Not because AI does not work, but because the bottleneck was somewhere else.

The technology is the last question, not the first one. Any AI advisor who leads with the tool rather than the constraint is selling you something.

The test that separates good AI fit from expensive disappointment

Once you have found the bottleneck, there is a single distinction that predicts whether AI will earn its keep there. Ask what kind of judgment is happening at that gate.

Repetitive judgment is the same decision made thousands of times against rules you could almost write down. Sorting, drafting, matching, summarizing, triaging, extracting. The work is real but the reasoning is patterned. This is where AI is genuinely excellent, and where most of the unglamorous, high-ROI wins live.

Real judgment is the call that depends on context nobody documented, a relationship with history, or a consequence someone has to personally own. Here AI is not merely weak. It is confidently wrong, and the confidence is the dangerous part. A system that fails loudly is a nuisance. A system that fails smoothly and plausibly, at scale, is a liability you will discover months late.

35+ years in software and security has taught me that nearly every failed AI budget I have reviewed was aimed at the second category while a pile of untouched work sat in the first.

"Not yet" is a legitimate answer

There is a trap hidden inside the question "where does AI fit in our business?" The trap is assuming that it fits somewhere, right now.

Sometimes, after you actually look, the answer is: not yet. Your data is scattered across four systems that disagree with each other. The process changes every month. The task you want automated has eleven exceptions that only live in one person's head. Point AI at that and you have funded an expensive, faster version of the same mess, plus a new maintenance burden.

That is not a reason to sit the technology out. It is a sequence. Fix the input, then automate it. Companies that skip the first half spend a year discovering it the hard way, and often conclude that AI does not work for their industry, when what actually failed was the foundation underneath it.

Constraint beats capability

Years ago I built AI for sports video, finding marketable moments in football matches using facial recognition. Run against every player in the league, it produced garbage. At one point it confidently identified a player who had been dead for twenty years.

The fix was not a better model or more training data. It was constraint. We limited recognition to the twenty-two players actually on the pitch. Accuracy went through the roof overnight.

That is the whole lesson of AI fit compressed into one story. The gain did not come from more AI. It came from someone with enough experience to narrow the problem until the technology could win. As an AI consultant, that narrowing is most of the value I bring, and it rarely looks impressive from the outside.

What this looks like in practice

If you want a practical starting point this week, do three things. Name your top bottleneck honestly, without dressing it up. Classify the work sitting in it as repetitive judgment or real judgment. Then check whether the inputs feeding it are clean enough to automate, or whether that is the actual first project.

Do that and you will have a defensible answer to where AI belongs in your business, along with a short list of places it clearly does not. Both halves are worth having. The second half is what saves the budget.

I was employee #2 as Senior Director of Technology at an AI startup, and every meaningful win there came from narrowing scope rather than widening it. That has not changed as a fractional CTO. Point AI at the right narrow task with clean constraints and results arrive in weeks. Point it at everything and you have a science project with a burn rate.

If you want a second opinion on where AI actually fits in your business, I take a few clients at a time. Book your free 15-min AI Fit Call →