When AI is the wrong tool

A meaningful share of the projects I get asked to scope do not need AI. The buyer thinks they do because the problem is hard, the team is busy, and AI is the topic at every leadership meeting. That is not a good reason.

Here is the short list of patterns I run through before saying yes.

The output is deterministic. If the answer to the question is the same every time given the same inputs, you do not need a model. You need a function. SQL, Python, an Excel macro, anything. Adding an LLM in the middle introduces non-determinism into a problem that did not have any.

The data is small and structured. A few thousand rows of clean tabular data, a clear schema, and a question with a known shape: this is a regression or a rule. Not a transformer. The "AI" framing is mostly social pressure at that point.

You already have a tool that does this. I have been pulled into projects where the actual fix was upgrading the existing reporting tool to its current version. Tableau got better. QuickBooks added the report. The team did not know because they were already shopping for the AI version.

The bottleneck is not the analysis, it is the decision. Sometimes the data is fine, the report is fine, and the actual problem is that the people who need to act on the report do not. Building a smarter dashboard does not fix a culture issue. A dashboard rebuild is a symptom.

The thing has to be auditable. Regulated industries, clinical decisions, anything where someone might end up in front of a court. If you cannot explain why the system did what it did, you cannot ship it, full stop. There are AI patterns that meet this bar; most off-the-shelf chatbot patterns do not. Knowing the difference is part of the job.

The honest version of this conversation takes thirty minutes. If your problem is in the list above, you will leave the call with a clearer picture and probably a vendor recommendation. I would rather lose the engagement than build the wrong thing.

If your problem is not in the list, we keep talking.