How to tell where AI is actually worth it
Not every problem is an AI problem. A practical way to spot the tasks where AI pays off, and the ones where it quietly costs you more than it saves.
Plenty of AI work fails not because the build was bad, but because it was aimed at the wrong thing. The model was fine. The task was a poor fit. So before spending anything on a build, it is worth knowing how to tell a good candidate from a bad one.
Here is the rough test we use.
Where AI tends to earn its place
Look for tasks that are:
- Repetitive and high volume. The same kind of work, done often. That is where even a small saving per item adds up, and where a tool gets used enough to become a habit.
- Language or pattern heavy. Drafting, summarising, classifying, pulling answers out of documents, tidying messy data. Work that is mostly about reading and writing is squarely in an AI’s strengths.
- Tolerant of a quick human check. The best fit is work where a person still casts an eye over the result before it matters. AI drafts, a human approves. That keeps you fast and safe at the same time.
- Bottlenecked by a scarce person. If one expert’s time is the constraint, handing them a tool that does the mechanical setup and write-up frees the part that genuinely needs them.
Where it usually is not worth it
Be wary of tasks that are:
- High stakes with no review step. If a wrong answer goes straight out the door with real consequences, the cost of the occasional confident mistake outweighs the speed.
- Starved of data. If the knowledge needed lives only in a few people’s heads and nowhere a tool can reach, fix that first. AI cannot answer from documents you do not have.
- Mostly judgement. Where the hard part is weighing trade-offs, reading a room, or owning a decision, AI can inform the call but should not make it.
- Rare and one-off. If something happens twice a year, the effort to build and maintain a tool rarely pays back.
The honest version
Most businesses have a handful of tasks that fit the first list well, and plenty that fit the second. The value is in being clear-eyed about which is which, and starting with the few that are genuinely worth it rather than the ones that simply demo well.
That clear-eyed read is where we start, before anyone writes a line of code.