Start with the process, not the model

Every week someone asks us which AI model they should be using. It's the wrong first question. The model is the easy part; the hard part is choosing work where automation actually pays off.

Before we build anything, we run every candidate task through five questions.

1. How often does it happen?

A task that happens 200 times a week is a far better candidate than one that happens twice a month, even if the monthly one is more painful. Frequency is what turns minutes into hours.

2. How long does it take each time?

Multiply frequency by duration and you have a baseline. "We get around 80 invoices a week and each takes about six minutes" is eight hours a week, or roughly 400 hours a year. Now you know what the prize is.

3. Are the rules mostly clear?

Good candidates have inputs that vary but decisions that don't: read this document, find these fields, put them there. If the decision depends on judgment only one senior person has, start somewhere else.

4. What does a mistake cost?

A drafted reply that a person reviews is low risk. Moving money is not. High-risk tasks can still be automated, but they need review steps, and that changes the savings math.

5. Where does the data live?

If the information is locked in a system with no API and no export, the integration work can outweigh the savings. Accessible data makes everything cheaper.

Putting it together

Score each task against those five questions and a short list appears quickly. In most businesses the same winners come up again and again: document intake, inbox triage, recurring reports and status updates. None of them is glamorous. All of them add up.

If you'd like a second opinion on your list, that's exactly what our free audit call is for.