Build vs buy: when to make it, when to buy it

The build-versus-buy debate usually starts in the wrong place. Leaders ask whether to make a tool or purchase one, when the real question sits a level up: what actually differentiates this business — and is this it? Almost nothing does. Customers don't reward you for owning a clever expense-categoriser or a bespoke chatbot. They reward you for the handful of things you do better than anyone else. Build there, and only there.

Get that order right and most decisions answer themselves. Get it wrong and you end up funding engineering for a problem the market already solved years ago.

The default is buy

For the overwhelming majority of AI capability, the sensible default is to buy it off the shelf. The economics aren't close. A vendor amortises model costs, security, compliance and improvement across thousands of customers; you'd be carrying all of it alone. Buying wins on speed — weeks, not quarters — on predictable cost, and on the fact that the vendor, not you, stays up at night keeping it working as the underlying models change.

There's a useful discipline here: spend last. Reach for the capability already built into tools you pay for before you reach for anything new. AI is now bundled into the CRM, the helpdesk, the spreadsheet, the accounting package. Much of what firms scope as a project is sitting unused inside software they already own.

When configuring beats both

Between buying new and building from scratch lies the option most firms underuse: configuring and connecting what's already in place. The bought capability is rarely the bottleneck — the gap is that your systems don't talk to each other. The opportunity is to wire the data flow, set up the rules and integrations, and shape the off-the-shelf tool to your process.

This is where good consulting earns its keep. It's cheaper than building, more tailored than buying blind, and it leans on software you're already maintaining. If a problem can be solved by connecting two systems you own, that is almost always the right answer.

When it's worth building

Building is the right call in a narrow set of cases — and all the conditions need to hold at once. It's worth building when the capability is core differentiation, not a back-office utility; when it runs on proprietary data or a process competitors can't replicate; and when, after a genuine look, no tool on the market does the job. Meet all three and a build can create real, defensible advantage. Meet only one or two and you're about to spend a lot to reinvent something you could have rented.

The hidden cost of build

The sticker price of a build is the smallest part of it. The real bill is what comes after: maintenance, model drift as the underlying AI is updated beneath you, security and data-governance obligations, and — most expensive of all — the people you need on hand to keep it running. A bespoke system isn't a purchase; it's a standing commitment.

The track record should give any leader pause. The Standish Group's long-running analysis of tens of thousands of software projects found only around 31% succeed outright — the rest are late, over budget, cut back, or abandoned. And the newest wave is no kinder: Gartner expects more than 40% of agentic AI projects to be scrapped by the end of 2027, citing runaway costs and unclear value. Build is not just expensive; it is the option most likely to fail.

A simple test

Before you commit to building anything, run the decision through five yes/no questions. Mostly "no" means buy or configure — and save the budget for where it counts.

• Is this capability genuinely core to how we compete — or is it plumbing?
• Does it depend on data or a process a competitor couldn't simply buy too?
• Have we honestly checked that no off-the-shelf tool — or one we already own — does the job?
• Can we fund not just the build, but the people and upkeep to run it for years?
• Could we get most of the value by connecting systems we already have instead?

If you're answering "yes" to the first four and "no" to the last, you may have a real case to build. Anything short of that, and the market has already made the thing for you — cheaper, faster, and someone else's problem to maintain.

Sources: Gartner, "Over 40% of Agentic AI Projects Will Be Canceled by End of 2027" (2025); The Standish Group, CHAOS project-outcome analysis (2020); published total-cost-of-ownership analyses comparing SaaS and custom software (industry, 2025–26). Cost figures are illustrative and indicate relative scale, not a quote for any specific build.