The first AI project sets the tone for every one after it. Choose well and the business asks for more. Choose badly and “we tried AI” becomes the reason nothing else gets funded. This is the scoring method we use in assessments. You can run it yourself in an afternoon.
List the work, not the technology
Start from a list of recurring tasks, written the way the people doing them would describe them: “check supplier invoices against purchase orders”, “answer where-is-my-order emails”. Aim for fifteen to thirty. Avoid entries like “use a chatbot”; that is a solution looking for a problem.
Score each task on five questions
- Volume. How many times a week does it happen? Below twenty, the saving rarely covers the build.
- Rules. Could a capable new hire learn it in a day from written instructions? If it takes years of judgment, keep it for later.
- Inputs. Does the information arrive digitally and in a few consistent shapes? Paper and phone calls add cost.
- Cost of a mistake. If an error is cheap and visible, automate fully. If it is expensive, automate the preparation and keep a person on the decision.
- Owner. Is there one person who wants this fixed and has authority over the process? Without an owner, projects drift.
Score each from one to five and add them up. Your first project is the highest score where the cost of a mistake is low. It is rarely the most exciting item on the list, and that is fine.
Set the measure before you build
Write down today's numbers: hours per week, turnaround time, error rate. Agree what result would make the project a success. Without a baseline, every outcome is an argument.
Keep the first build small
Three to four weeks, one workflow, one team. A small win in production teaches you more about your data, your systems and your people than a six-month programme on slides.