Buy, build, or wait

Every AI capability a business wants can be obtained in one of three ways: buy a product that does it, build something specific, or wait for the market to make it cheaper and better.

All three are legitimate. What is not legitimate is choosing by disposition — the firm that always builds, the firm that always buys — rather than by looking at the particular capability in question.

Buy

Buying wins whenever the capability is genuinely generic. Transcription, translation, note-taking, general-purpose assistants. Thousands of businesses need the same thing, a vendor amortises the cost across all of them, and no internal build competes.

The failure mode is buying a generic product for a specific need and then spending eighteen months and considerable goodwill bending your process to fit it. The warning sign is a procurement conversation dominated by workarounds. If the demo requires you to describe how you would change your operation, you are buying the wrong thing.

Build

Building wins where the value comes from something specific to you: your knowledge, your process, your data, your judgement. Nobody sells a product that knows how your firm prices unusual jobs, because that market has one customer.

The failure mode is building something a vendor already sells, usually because it seemed cheap to prototype. It generally is cheap to prototype. The cost is not the build; it is the four years of maintenance, the person who has to understand it, and the upgrade you now have to do yourself while the vendor's customers get theirs for free.

Ask what you are actually buying when you build: a capability, or a permanent maintenance obligation with a capability attached.

Wait

Waiting wins when a capability is clearly on a trajectory — improving fast, falling in price, with several credible vendors moving towards it. Committing early to something that will be a commodity in eighteen months buys you eighteen months of advantage in exchange for a system you will have to unpick.

The failure mode is waiting as a permanent posture. Deliberate waiting has a review date and a trigger. Waiting without one is not a strategy, it is a habit, and the businesses that lose ground are rarely the ones that chose to wait — they are the ones that never noticed they had.

A rough test

Describe the capability in one sentence, then ask how many other businesses would write that same sentence.

• Thousands, in nearly identical words — buy it, and change your process to fit rather than the reverse.

• A handful, and only in your sector — buy the underlying pieces, build the thin layer that makes them yours.

• Only you, because it depends on how your business actually works — build it, and budget for owning it.

• You cannot write the sentence yet — wait, with a date in the diary.

The answer often differs across a single project. A knowledge system might buy the model, build the retrieval and interface, and wait on the agent layer. Treating buy, build and wait as one decision for a whole initiative is what produces the worst outcomes in all three directions.