Your First AI Automation: Five Questions That Pick the Right Process

The first AI automation a company ships sets the tone for everything after it. If it works, the next project gets budget and goodwill. If it stalls, AI becomes "that thing we tried". In our projects, the deciding factor is rarely the model. It is the choice of process.

These five questions are the filter we use in discovery workshops. A process that passes all five is a strong first candidate.

1. Does the work repeat every week?

Automation pays back through repetition. A task that happens a few times a year is rarely worth a system, however annoying it is. Look for work that arrives in a steady stream: invoices, support emails, intake forms, weekly reports.

2. Can you describe a good result?

If two experienced colleagues would disagree about what a correct outcome looks like, an AI system will struggle too. Pick work where "done well" can be written down, even if the steps to get there vary. That definition later becomes your test set.

3. Is the input mostly text or documents?

Language models are strongest where traditional automation broke down: reading emails, pulling fields from PDFs, classifying requests and drafting replies. If the work is already structured data moving between two systems, a classic integration may be cheaper and more reliable.

4. What happens when the system is wrong?

Every AI system makes mistakes, so design for them. A good first process has a cheap failure mode: a person reviews the output, or errors are easy to spot and fix. Avoid starting where one wrong answer reaches a customer or a regulator unchecked.

5. Can you measure the before and after?

Write down today's numbers before you build anything: how many items arrive per week, how long each one takes, how often something goes wrong. Without a baseline, even a successful project turns into a debate about opinions.

Three traps we see often

• Starting with the most visible process. The process everyone talks about in leadership meetings is often the most political and the hardest to measure. Visibility can come with the second project.
• Automating a broken process. If the manual version is unclear or full of exceptions, automation only makes the confusion faster. Simplify the process first, then automate it.
• Skipping the people who do the work. The team that handles the process today knows every edge case. Involve them early: they become the system's first users and its best testers.

Putting it together

Score your candidate processes against the five questions and start with the one that passes most clearly, not the one with the biggest theoretical upside. A smaller win that ships in weeks beats an ambitious project that is still in testing next quarter.

For a quick first estimate, describe the process in the ROI calculator, or browse 30 common workflows for examples by business area.