Nobody needs an AI strategy.

If you run operations at a company doing somewhere between $30M and a billion, you have already sat through the meeting. Someone presented a maturity model. There was a roadmap with three horizons on it. Everyone agreed AI was going to be significant for the business, a steering committee was proposed, and eleven months later the only thing that has actually changed is that two people in finance are pasting things into a chatbot and hoping nobody asks about it.

This is not a failure of ambition. Almost every company in this bracket has adopted AI somewhere — a pilot, a bundled feature in a renewal, a tool a manager expensed. Very few are running any part of the business on it. The gap between those two states is enormous, and almost nobody names what actually causes it.

The bottleneck isn't the technology

Here is the uncomfortable part: the models are fine. They have been fine for a while. For the things an operations team actually wants — summarize this, extract that, answer a question about last quarter, route this request to the right person — the capability has been sitting there, commoditized, for longer than most strategy decks acknowledge.

What is not fine is that the information those models would need to be useful is locked inside systems that were never built to be asked questions.

Your AI problem is almost always a data-access problem wearing a different hat.

And "data access" sounds abstract until you look at what it means concretely in a real operations business:

• A core platform your business has run on for a decade that has no meaningful API — or has one, technically, that no one at the vendor can explain.
• A reporting module that will happily produce a beautiful PDF and offers no other way to get the same numbers out.
• Four spreadsheets that are load-bearing, one of which only one person fully understands, and that person is on holiday.
• A status field that means three different things depending on which regional office typed it, because nobody ever wrote down what it was supposed to mean.
• Ten years of scanned documents that contain the answer to a question you get asked weekly.

None of that is a strategy problem. All of it is plumbing. And plumbing is unglamorous enough that it rarely survives contact with a steering committee.

The trap that follows

Once a company does correctly diagnose this as a data problem, the recommendation that usually arrives is worse than the disease: consolidate onto one platform, or build a central data platform first. Rip out the old thing. Migrate everything. Then, on the far side, everything will be connected and the AI work becomes easy.

Sometimes that is genuinely the right call. Far more often it is a two-year commitment that delivers its value entirely at the end, if it delivers at all — and it asks an operations team to absorb enormous disruption before receiving anything. Meanwhile the report still takes three hours every Monday. It took three hours before the migration was proposed and it will take three hours for the two years the migration is running.

There is also a quieter cost. A platform migration is a decision that has to be defended, which means it becomes hard to stop once started, even when the evidence says stop.

What we'd do instead

Pick the single process that costs your team the most hours. Not the most strategic one — the most expensive one in wall-clock time. Make it work end to end against the systems you already own. Ship it. Then pick the next one.

That approach is unfashionable and it has four things going for it.

• It produces evidence. In weeks, not quarters, you know whether this works in your business rather than in a case study about someone else's.
• It is reversible. If it turns out badly you have spent a small amount of money and changed nothing structural. Try reversing a platform migration in month fourteen.
• It tells you what your data is actually like. Every organization believes its data is worse than average. You cannot find out which parts are genuinely bad from a workshop. You find out by trying to use it for something real.
• It pays before it finishes. The hours come back immediately, on the first process, while the rest is still ahead of you.

Where this argument stops

It would be dishonest to pretend this scales indefinitely. Solve enough processes one at a time and you will eventually accumulate connections that want governing properly — and at that point the platform conversation becomes real rather than theoretical.

The difference is that you will arrive at it having already banked value, knowing precisely which systems matter, and holding actual evidence about the state of your data. That is a far better position from which to spend serious money than a maturity model and a hunch.

A test you can run in one sentence

If you want to know whether any of this applies to you, don't convene anything. Just try to finish this sentence: "The thing that eats the most time around here is ______."

If you can name it immediately — and most operators can, because they have been irritated by it for years — then you do not have a strategy problem. You have a specific, boring, solvable problem, and the fastest route to an answer is to go and solve that one thing.

If you genuinely cannot name it, that is worth knowing too. It usually means the cost is spread thin across a lot of small frictions rather than concentrated in one place, and the honest answer is that AI is not your highest-value next move.