Build, buy, or wait

A short framework for AI decisions. The right answer is "wait" more often than anyone selling something will admit.

Every AI use case comes down to three options: build it, buy it, or wait. This is the framework we use in feasibility studies. It's deliberately simple.

Buy when the problem is generic and the data isn't yours

Meeting transcription. Email drafting. General coding assistance. Standard OCR. These are solved problems where your data adds little and a vendor's scale adds a lot. Building here wastes good engineers. The only questions are cost, data handling, and whether the vendor will exist in three years.

Build when the edge is in your data or your process

If what would make the system good is something only you have, like ten years of claims history, your product catalogue, or the way your underwriters actually decide, a generic product won't capture it. These are the cases worth building, and they're narrower than most pitches suggest. One test: if a competitor bought the same vendor product, would you still have an edge? If yes, build.

Wait when the capability is moving faster than you can deploy it

Nobody sells this option, which is why it's undervalued. Some problems that need a custom system today will be a checkbox in a general product in eighteen months. If the cost of waiting is low and the problem isn't core, waiting is a real decision. Use the time to get the data and the evaluation in order, so that when you move, you move fast.

The mistake we see most

Companies build what they should buy, because building feels like strategy. And they buy what they should build, because the vendor demo was good. Same fix either way: decide where your actual edge is before the conversation about tools begins.