Build vs buy: a framework for the AI tooling decision

The build-vs-buy question comes up in nearly every AI engagement, and the default answer (buy an existing tool rather than build custom) is correct more often than founders' instincts suggest, because most AI use cases aren't actually unique to the business asking about them.

The framework I use has three questions, in order: is this capability core to the business's actual competitive advantage, not just useful but differentiating, does an existing tool solve 80% of the requirement out of the box, and does the team have the ongoing capacity to maintain a custom system after the initial build.

A "no" on the first question usually ends the conversation in favor of buying: a genuinely undifferentiated capability, like document summarization or basic lead enrichment, is rarely worth the maintenance burden of a custom build, however satisfying it might be to build.

The exceptions are real but narrower than expected: proprietary data creating a genuine moat, or a workflow specific enough that no existing tool's assumptions match it closely enough to be worth the integration effort. Most teams overestimate how often their situation is actually the exception.