The Build-vs-Buy-vs-Skip Scorecard We Use for Every AI Evaluation

Every AI decision a business faces comes down to one of three answers: build it, buy it, or skip it. Most owners only seriously consider one of the three — usually “buy,” sometimes “build” after a good demo — and almost nobody puts “skip” on the table. That's exactly backwards. Here's the scorecard for choosing well.

First: is this even a job worth doing?

Before build-vs-buy, there's a gate: is automating this job worth anything? Name the job in one sentence, then put a number on what doing it manually costs today — the hours, the errors, the delay. If you can't, you're not ready to evaluate tools; you're ready to go measure. A tool that automates a job that didn't matter is just a more efficient way to waste money.

Buy — the default, and usually right

If a vendor already does this well and the price is fair, buy it. Don't build what you can rent. The test for “buy” is whether the job is common enough that someone has built a mature tool for it, and whether that tool fits how you work without heavy customization. Accounting, email, scheduling, payroll, most CRMs — these are solved problems. Nobody should write their own general ledger.

There's a real example worth borrowing here. A bookkeeping operation ran its core accounting on SAP — bought, emphatically not built, because the general-ledger problem is solved and owning that code would be madness. But the leadership reporting SAP didn't surface cleanly? That got built on top. Same operation, two different answers, each correct for its layer. That's the judgment the scorecard is meant to force.

Build — rarer than it feels in the moment

Build only when two things are both true: the job is genuinely specific to how you operate, so no off-the-shelf tool fits; and the value is big enough to justify owning a thing you'll have to maintain forever. The trap is building because it's interesting, or because a tool is 90% there and you want the last 10%. Custom software is a liability you carry indefinitely — worth it sometimes, but the bar is high, and most “we should build this” impulses don't clear it.

Skip — the answer a vendor will never give you

The most undervalued option. Skip when the job isn't worth automating, when no tool is ready, or when your data or process isn't in shape to support it yet. “Not yet” is a real answer, and it's free. A skipped automation costs nothing; a wrong build or buy costs money, time, and the credibility of the next AI project you propose. Reaching for “skip” without guilt is a sign of a mature evaluation, not a failed one.

The scorecard, in one pass

1. Name the job and what doing it manually costs. No number, no decision.
2. Buy if it's a solved problem and a tool fits without heavy customization.
3. Build only if it's specific to you AND the value justifies owning it forever.
4. Skip if the job's not worth it, the tool's not ready, or you're not ready.

Run every candidate through those four and most of the noise clears. The businesses that get AI right aren't the ones that build the most or buy the most — they're the ones that can tell the three apart, and aren't afraid to choose the one that costs nothing.