Your AI pilot is a license, not a strategy

Most companies we meet have three AI subscriptions, two pilots, and no number. That is not a strategy. It is a cost line waiting to be noticed.

Ask a leadership team what their AI strategy is and you will usually hear a list of tools. A license for the writing assistant. A pilot with a vendor everyone met at a conference. A team in one department doing something interesting that nobody outside that department fully understands. Everyone agrees it matters. Almost nobody can say what any of it has actually returned.

That is not a criticism of the people running these programs. It is close to where most of the market sits right now. But it is worth being honest that a stack of subscriptions is not a strategy, and a pilot with no metric attached to it is not a pilot. It is a license.

How this happens

The pattern is consistent enough to name. A vendor demo lands well in a leadership meeting. Someone senior wants to be seen moving on AI, reasonably, since the pressure to show progress is real. Procurement signs something. A small team starts using it enthusiastically, without anyone having measured what the process cost before the tool arrived. Six months later, the tool is still being paid for, the team likes it, and nobody in finance can say whether it is worth what it costs, because there was never a baseline to compare against.

Multiply that by three or four tools across different functions, each with its own champion, its own renewal date, and its own quiet assumption that it is obviously working, and you get the state most companies are actually in: real spend, real enthusiasm, no number anyone would defend in front of a board.

The question that fixes it

Which workflow, at what volume, costing what today, will cost what after the tool is in place? If nobody can answer that for a given tool, the tool is not deployed. It is installed.

That question sounds simple. It is rarely asked before the purchase, which is exactly the problem. Most organizations pick the tool first and try to build a business case backward from there, usually under pressure to justify a renewal that is already a few weeks away. Start with the process and the number instead. Choose the tool second. Doing it the other way round is how a company ends up paying for three overlapping things, with none of them clearly earning their keep.

There is a version of this that looks worse on paper but is actually healthier: a company running exactly one pilot, honestly labeled as an experiment, with a clear date to decide whether it gets killed or scaled. That beats five quiet subscriptions every time. The number of tools in use is not the signal worth watching. The number of tools with a named owner and a number attached to them is.

What working looks like

The pattern that actually works is narrower than most companies expect going in. One function. One high-volume, judgment-heavy workflow, the kind with enough repetition that a modest improvement is worth real money, but enough judgment involved that conventional software has not already automated it. A baseline measured honestly before anything gets built, cost, time, error rate, whatever the workflow's real currency is. A tool chosen for that specific job rather than for its logo or its funding round. One named owner accountable for the result. A short runway, sixty to ninety days, to get it into production with adoption included, not a technical pilot sitting unused on a server.

Then, and only then, the next workflow.

It is not an exciting way to run an AI program, and it will not make an exciting slide. It works, largely because the first thing a properly measured pilot tends to find is the licenses already being paid for that nobody is using, which is often enough on its own to fund the next one.