An AI adoption checklist for mid-sized businesses

Most failed AI projects fail before any model is built. These are the questions worth answering first.

Start with a process, not a technology

The projects that work begin with a named process that is high volume, rules heavy and already documented. The projects that stall begin with a decision to use AI and a search for somewhere to put it.

Write down the process as it runs today, including the exceptions. If it cannot be written down, it cannot be automated reliably, and the first piece of work is documentation rather than engineering.

Check the data before promising anything

Ask where the data lives, who owns it, how far back it goes and how clean it is. A model trained on three months of inconsistent records will produce confident, useless output.

A short data assessment ahead of any build is cheap and regularly changes the plan. It is better to learn in week one that the data is not there.

Agree what success means in numbers

Set the metric and the threshold before the work starts: accuracy on a held-out sample, hours saved per week, cost per processed item. Then test against it honestly.

Without a threshold, every result can be presented as progress, and nobody can say whether the project worked.

Decide where a human stays in the loop

For anything that changes data, moves money or reaches a customer, decide up front which steps need approval and what happens to low-confidence results.

A review queue for uncertain cases is not a sign of a weak model. It is what makes the output safe to use.

Plan for the second year

Models drift as the business changes. Someone needs to own monitoring, retraining and the cost of running the thing once the project team has moved on.

If no owner is named, the model quietly degrades until it is switched off, and the organisation concludes that AI did not work.

A short checklist

Is the process documented, including exceptions? Is the data available, owned and good enough? Is there a written success metric and threshold? Is it clear where a person approves? Is there a named owner for the year after launch? Five yes answers is a project worth starting.

Want this applied to your business?

Answer four quick questions and we will turn them into a brief, ready to send to [email protected].