When AI earns its place (and when a spreadsheet wins)

Almost every product now carries an "AI powered" badge, and almost none of them explain what that means for you. The label has stopped carrying information. So the useful question is not whether something uses AI. It is whether AI is the right tool for the specific job in front of you, or whether something simpler would do the same work better and cheaper.

Here is the test we actually use.

Four questions before any model

1. Is the input messy language or documents, rather than clean rows and columns?
2. Is the volume high enough that people cannot keep up by hand?
3. Does a rough first draft cost you less than having no draft at all?
4. Will a person still review the output before anything happens because of it?

Answer yes to most of these and AI probably earns its place. Answer no, and a simpler tool is likely the honest choice.

When AI earns its place

Language models are good at work that has no fixed rules and too much volume for people to keep up with. Sorting a flood of inbound messages by what they are actually about. Turning a long document into a first draft summary that someone then checks. Finding the one relevant paragraph across thousands of files that were never organized. In each case the model does the reading and the routing, and a person keeps the judgment.

When a spreadsheet wins

If the rules are fixed, the volume is low, and the data is already structured, you do not need a model. You need a clear formula and a layout people can trust. The same holds when compliance requires the same input to produce the same output every time. A model that is right most of the time is the wrong tool for a job that demands right every time.

When neither is the answer

Sometimes the problem sits upstream of both. Two systems that will not talk to each other. A data model that records the wrong things. A process with a missing step that everyone patches by hand. No model and no spreadsheet fixes a broken pipe. Fix the plumbing first, then decide what belongs on top of it.

What to watch for in a pitch

A few signals that a vendor is selling the label rather than the tool. There is no clear point where a person reviews the output. They can tell you how the system succeeds but not how it fails. Or the fine print quietly lets them train on your data. Any one of these is worth a hard question before you sign anything.

How this shows up in our work

On a scoping call, this same test is how we talk ourselves out of work. When the honest fix is a form, a routing rule, or a dashboard, we say so, because selling you a model you do not need is a reliable way to never hear from you again. AI is a tool, not a default.

The right answer is often boring on purpose.