Ask a leadership team where AI should go first and you usually get the most visible process. Ask the people doing the work and you get the most repetitive one. The second answer is almost always the better project.
The three questions that rank any process
- Volume — how many times a month does this happen? Low volume cannot repay integration cost, however irritating the task is.
- Variability — do the inputs arrive in a predictable shape? Predictable means automation; variable means AI earns its place.
- Consequence of error — what does a wrong output cost? This determines how much review the process needs, and therefore whether the economics survive.
Where to look
The strongest candidates are rarely on anyone's roadmap. They surface as: work that exists only because two systems do not talk; anything a person does by copying between screens; queues that grow predictably every Monday; requests that get triaged by whoever is available; and the reports somebody rebuilds by hand every month.
The opportunities that look attractive and stall
Customer-facing systems with high error cost, processes that are undocumented, anything depending on data quality nobody has audited, and work whose rules exist only in one experienced person's head. Each is genuinely valuable and none is a first project — the risk arrives before the organisation has learned how to run these systems.
Why conventional prioritisation misfires
Vendor workshops rank opportunities by what the product does well. Internal wishlists rank by departmental influence. Neither examines data readiness or error cost, which are the two variables that actually determine whether a project reaches production.
Knowing the three questions is easy. Applying them honestly across forty processes, with access to the data behind each one and no stake in the answer, is the work — and it is why we run assessment as a separate engagement that can conclude you are not ready.
What this means for your business
Spend a fortnight mapping processes against volume, variability and error cost before choosing anything. The ranked list is usually surprising, and it is far cheaper than discovering the ranking through a failed build. Our guide to AI versus automation covers how to tell which tool each candidate actually needs.