Build vs buy AI - why your leadership does not understand either choice

The VP of Engineering wants to build. The CFO wants to buy.

Both are wrong. Not for the reasons they think.

Companies burn through decision frameworks trying to figure out whether to build custom AI or buy off-the-shelf. They create weighted scoring models, compare total cost of ownership, and analyze time-to-value. Then they pick one, invest heavily, and watch nearly half their AI pilots die before reaching production. RAND Corporation put a number on it: more than 80% of AI projects fail at twice the rate of non-AI IT projects. And 92% of companies are investing in AI but only 1% achieve full maturity.

The problem isn’t the technology choice. Your middle managers have no idea how to use AI, and they’re the ones who determine whether anything you build or buy actually gets used.

The layer everyone skips

What actually happens in most companies follows a pattern. The C-suite gets excited about AI, reads analyst reports, attends conferences, and mandates adoption.

Entry-level employees start experimenting immediately because they have nothing to lose.

But the middle layer, the people who actually run your operations, they’re stuck. Nearly half face pressure from above to deliver on initiatives they don’t fully understand while reassuring those below about job security. These are the managers who make or break your AI investment, regardless of whether you built it or bought it. And 46% of tech leaders cite AI skill gaps as a major obstacle to getting anything done.

An HBR analysis of organizational barriers tells the same story: a lack of clear AI strategy is most commonly cited as the biggest barrier to adoption. But dig deeper and you find the real issue: middle management doesn’t understand how AI works well enough to integrate it into their daily operations. They’re being asked to lead a change they don’t fully grasp.

So they do what makes sense to them. They maintain the status quo.

Why build vs buy frameworks keep failing

The typical build vs buy analysis looks at cost, time, customization needs, and competitive advantage. All rational criteria. And 76% of AI use cases are now deployed via third-party or off-the-shelf solutions anyway, suggesting the market has already made up its mind about buying.

What the analysis doesn’t look at: whether anyone in your organization can actually explain to a new employee how the AI tool helps them do their job better.

A mid-size company I followed spent eight months on this decision. They built a detailed scoring model. They interviewed vendors and mapped their technical requirements. They chose to build, invested significantly in custom development, and produced something technically solid.

Six months after launch, adoption sat at 11%.

The AI worked perfectly. The problem was that managers didn’t trust it because they didn’t understand it. When employees asked questions, managers couldn’t answer. When edge cases appeared, they defaulted to the old manual process. The custom AI sat there, performing flawlessly for the tiny fraction of work anyone would actually send its way.

This pattern is everywhere. Only 11% of organizations are actively using AI agents in production. The rest are stuck in pilot programs, abandoned after cost overruns, or quietly shelved when real expenses surfaced.

That company could have bought an off-the-shelf solution and hit exactly the same adoption rate for a fraction of the cost. 71% of tech teams choose off-the-shelf specifically to accelerate time-to-value. The build vs buy decision was the wrong question to begin with.

What P&G and AXA figured out

P&G created something most companies skip entirely. An AI mentorship program where junior employees teach senior leaders how the technology actually works.

Not in a formal training session. Not in a webinar. Through structured reverse mentoring where a 24-year-old data analyst sits down with a VP and shows them, hands-on, what AI can and can’t do.

Their results: reporting time dropped from hours to minutes. But more importantly, the executives understood why and how the AI reached its conclusions. They could answer questions from their teams and make informed decisions about when to use the AI and when not to.

AXA Insurance ran a similar program starting in 2014. After six reverse mentoring sessions, 97% of participants recommended it. Why? Senior leaders finally understood the technology well enough to champion adoption across their teams.

Linklaters, the law firm, reported that 100% of participants endorsed their reverse mentoring program as an effective way to build shared understanding around new technology. When managers understand the tools, they use the tools. I think that probably sounds obvious. So why do so few companies actually sequence it this way before making the build vs buy call?

The sequence that changes the outcome

Stop asking build or buy first. Start by asking whether your management layer understands AI well enough to successfully deploy anything.

If they don’t, start an AI mentorship program immediately. Pair junior employees who use AI naturally with the middle and senior managers who make adoption decisions. Give them three months of structured learning.

Then make your build vs buy decision.

Something real shifts when you do this in the right order. When your VP of Operations understands how AI works, they can tell you whether an off-the-shelf solution will actually fit your workflow. When your Director of Customer Success has hands-on experience with AI limitations, they can specify what custom features would actually deliver value versus what just sounds good in a requirements document. The build vs buy decision becomes dramatically clearer when the people making it actually understand the technology.

BearingPoint’s research on middle management AI adoption points in the same direction. Give middle managers time and tools to become confident AI users before asking them to lead others. Otherwise you get resistance masquerading as legitimate concerns about the technology. Only about 20% of organizations achieve enterprise-level impact from AI initiatives. Most fail due to weak data foundations, inadequate governance, and poor integration. All symptoms of leadership that doesn’t understand what they deployed.

What to add to your framework right now

If you’re choosing between build and buy right now, add one more criterion: which option includes a real plan to make your managers capable AI users?

A custom-built solution with no adoption plan will fail. An off-the-shelf platform with no internal champions will fail just as surely. Both cost money. Both waste time. 85% of companies miss AI cost forecasts by more than 10%. That gap is where AI projects go to die, regardless of whether you built or bought.

The companies actually getting results with AI, built or bought, are the ones that invested in reverse mentoring first. They created an AI mentorship program that gave their decision-makers real hands-on experience before asking them to mandate adoption from above.

The spreadsheet comparison between building and buying matters less than you think. Most Fortune 500 firms now settle on a blend, buying vendor platforms for governance and compliance while building the last mile of custom workflows. What actually matters is whether the people running your operations can confidently use AI in their daily work and teach others to do the same.

Fix that first. Then the build vs buy decision becomes obvious.