Build, buy or partner is usually settled by whoever is most persuasive in the room, then defended as a technology decision. It is a strategy decision about where your advantage really sits, and answering that first makes the rest straightforward.
The short answer
- Build where the capability is a genuine source of advantage and rests on data or process only you have. Buy where it is a solved commodity. Partner where you need capability now and intend to own it later.
- The real cost of building is not the first version. It is owning, securing and maintaining it for years, which is where most build cases quietly understate the bill.
- Buying trades control for speed, and its central risk is lock-in, so the exit terms matter as much as the features.
- Partnering is a bridge, and it only works if it transfers capability on a defined timetable rather than renewing indefinitely.
- The decision is rarely uniform. Different layers of the same system point to different answers.
Decide by advantage, not by preference
The engineers often want to build, the finance team often wants to buy, and the integrators always want to partner. None of those instincts is a strategy. The question that resolves it is where your competitive advantage actually lives. Capability that differentiates you, and that depends on data, process knowledge or customer relationships that competitors cannot copy, is worth owning. Capability that every organisation needs and nobody wins on is a commodity to acquire as cheaply and reliably as possible. Most of an enterprise AI system is the second kind, and a small, decisive part is the first.
That is why the decision is rarely uniform across a whole system. The underlying model may be bought, the data pipeline that feeds it built, and specialist help brought in to stand up the first release. Deciding build-buy-partner for the system as a single lump is the mistake; deciding it layer by layer, against advantage, is the discipline.
What each option really costs
| Build | Buy | Partner | |
|---|---|---|---|
| You gain | Control, fit, and ownership of the advantage | Speed, and a supported product | Capability now, without hiring first |
| You give up | Time, and a long maintenance commitment | Control, and independence from the vendor | Margin, and some control of the roadmap |
| The hidden cost | Years of owning, securing and updating it | Lock-in, and the price of leaving | Dependence that outlasts the plan |
| Right when | The capability is a durable advantage | The capability is a commodity | You need it now and will own it later |
The build case that understates the bill
Building looks cheapest at the point of decision because the number on the slide is the cost of the first working version. The real cost is the decade after it: the team that maintains it, the security work that never stops, the retraining as data drifts, and the key-person risk when the two people who understand it move on. A build case that counts only the initial delivery is comparing an incomplete number against a vendor's fully loaded one, and it will win the argument and lose the budget. Build deliberately, and only where the ownership buys a real advantage, because everything you build you own forever.
The buy decision that hinges on the exit
Buying is the right answer for commodity capability, and its central risk is dependence on the vendor. The features get all the attention in the sales process, and the terms that matter most are the ones about leaving: whether you can get your data out in a usable form, whether the model or its outputs are portable, and what switching would actually cost in year three. A buy decision made on the demo and signed without the exit terms is how a commodity purchase becomes a strategic dependency. Weigh the cost of getting out before you weigh the cost of getting in.
The partnership that has to end
Partnering is the right bridge when you need capability faster than you can hire it and intend to own it eventually. It fails when the transfer never happens and the arrangement renews year after year, so the capability lives permanently outside the organisation. A partnership worth entering has capability transfer written into it: named people learning named things on a defined timetable, with the internal team taking the work over by an agreed date. Without that, you have bought a dependency and called it a partnership, and the choice of whether to hold capability in-house is a decision in its own right, examined in why pilots stall between demonstration and production.
The constraint
The binding limit is clarity about which parts of the system are your advantage and which are commodity, rather than the build-buy-partner choice itself. Deciding sourcing before that map exists means building commodities you should have bought and buying the very capability that would have differentiated you, because the deciding question is where your advantage sits, and only that answer tells you what is worth owning.
How to run the decision
Take the system apart into its layers, and for each one ask whether it is a source of advantage or a commodity, whether the data underneath it is uniquely yours, and how long you are prepared to own it. Build the advantage layers, buy the commodity layers, and partner only where you need a bridge with a transfer date attached. A decision made that way is defensible in front of a board and stable over the life of the system. A decision made by whoever argued hardest tends to be re-litigated the first time the bill or the lock-in becomes visible.
Sources
- RAND Corporation, The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed, 2024, on capability and ownership as determinants of success.
- Gartner press releases on generative and agentic AI project outcomes, 2024 and 2025, on cost and value as cancellation drivers.