AI is an operating decision, not a technology decision

Most organisations approach AI the way they approach software: pick a vendor, buy seats, run a pilot, measure adoption. Twelve months later there is a licence bill, a small group of enthusiasts and no change in how the business actually runs.

That happens because the decision was framed wrong. AI is not a technology decision. It is an operating decision.

The difference in practice

A technology decision asks: which tool do we buy, and who gets access?

An operating decision asks: which piece of work stops being done the way it is done today, who owns the new version of it, and what number moves when it works?

The second question is harder because it implicates process ownership, headcount, quality control and accountability. It is also the only one that produces results.

Start from throughput, not from capability

The wrong starting point is a demo. The right starting point is a bottleneck.

Pick the process where volume is capped by a person's hours: quoting, first-line customer replies, supplier follow-ups, pricing updates, document preparation. Then describe the current work precisely enough that someone new could do it. If you cannot write it down, you cannot automate it, and the AI conversation is premature.

Only then ask what a model can absorb: drafting, extracting, classifying, summarising, checking. Keep judgement, exceptions and relationships with people.

Design for the failure case first

Every automated workflow will be wrong sometimes. The workflows that survive in real businesses are the ones where being wrong is cheap and visible:

• A human approves anything that leaves the company or costs money.
• Every output is logged with its input, so an error can be traced.
• There is a defined threshold at which the system escalates instead of guessing.
• Someone owns the workflow by name, the way they would own a team member's work.

Without those four things, you have not automated a process. You have added an unmonitored employee.

Measure the operating change, not the usage

Seat adoption is a vanity metric. The questions worth reporting to a board are simpler:

• How many quotes did we send this month, and how long did each take?
• What did we stop doing entirely?
• Where did quality get better or worse?
• What capacity did this free, and what did we spend it on?

If none of those numbers moved, the pilot succeeded technically and failed commercially.

The uncomfortable part

Treating AI as an operating decision means leadership cannot delegate it to IT and check back in a quarter. The change is in how work is defined and owned, and that belongs to the people who own the operation.

The companies pulling ahead are not the ones with the best models. They are the ones that were willing to redesign a process, put a name against it, and measure what changed.