We make our living putting AI into real businesses, so it probably sounds strange to hear us argue against it. But the fastest way to lose a client's trust is to sell them a clever tool they did not need. So here is the honest version, the one we give over coffee before anyone signs anything.
The first place AI does not belong is anywhere the rules are already clear. If a task has a fixed set of conditions and a known answer for each one, you do not want a model guessing at it. You want plain logic. A model that approves a refund "most of the time" is worse than a simple rule that approves it every time the criteria are met. Save the judgment for places that actually need judgment.
The second is anywhere the cost of a quiet mistake is high and nobody is watching. AI is brilliant at producing a confident answer. It is much weaker at telling you, loudly, when it is unsure. If an error can slip into a contract, a payment, or a compliance record without a human ever seeing it, that is not a workflow for autonomous AI. It might still be a workflow for AI that flags and suggests, with a person making the call. The difference is everything.
When the data is not ready
The third, and the one we run into most, is when the underlying data is a mess. People hear that AI can "read anything" and assume it will tidy up years of inconsistent records for free. It will not. Point a model at scattered, contradictory, half duplicated data and it will give you scattered, contradictory answers, just faster and with more polish. The polish is the dangerous part, because it hides the rot. Sort the data first. Then talk about AI.
There is also a quieter category: tasks that are genuinely small. We have talked clients out of building an AI feature because a one hour change to a form, or a saved filter, solved the actual problem. Not everything that could use AI should. If a checklist or a template gets you ninety percent of the way there, build the checklist.
And finally, do not reach for AI when what you really want is to avoid a hard conversation. We have seen teams try to automate around a broken process or an unclear responsibility, because the tooling feels easier than the meeting. It is not. The model will simply industrialise the confusion.
None of this means we are pessimistic about AI. We are the opposite. We are confident enough about where it earns its place that we are relaxed about naming where it does not. That is also why every engagement starts with a Discovery Sprint of 2 to 4 weeks. By the end of it we have a working prototype and, just as often, a clear list of the things we decided not to build. Both are wins.
Facing something similar in your business?
Talk it through with our AI guide, or send the team a note. We will tell you straight whether and how we can help.