Is this a problem for AI?

Start with the problem, not the technology. Write it in one sentence without mentioning AI: "Our support team spends two hours a day routing emails to the right queue."

Then ask five questions:

Is it repetitive and high-volume? AI pays off on tasks done thousands of times, not twice a year.
Are the rules too fuzzy to write down? If a simple rule works ("anything from the billing address goes to billing"), use the rule.
Is there data or a clear definition of a good result? You need examples of past decisions, or at least a way to judge an output.
How bad is a mistake? Drafting an email reply tolerates errors that approving a loan does not. The higher the cost, the more human review you need.
Who does the work today, and what happens to them? Plan for how the job changes, not only for the tool.

Good fit	Poor fit
Sorting, tagging and routing text	Decisions needing a guaranteed, explainable rule
Drafting and summarising for a person to check	Rare, one-off tasks
Finding patterns in large amounts of data	Problems with no data and no way to judge output
Transcribing, translating, extracting fields	High-stakes decisions with no human review

If a problem fails most of these questions, the best AI project may be no AI project.

Check your understanding

Further reading