Build, buy or partner: how to choose your path to AI

When a company decides to move ahead with AI, it soon reaches a crossroads: buy a ready-made tool, build with an in-house team, or work with a specialist partner. For a while, building looked like the natural choice for large companies. The latest data tells a different story.

What changed between 2024 and 2025

Menlo Ventures, an investment fund that tracks the enterprise AI market in the United States, surveyed AI decision-makers two years running. In 2024, 47% of AI solutions were built in-house and 53% were bought. In 2025, 76% of AI use cases were bought (Menlo Ventures, 2024 and 2025). Because Menlo invests in AI companies, the figures should be read with that in mind. But the direction of the shift is clear.

The 2025 MIT NANDA study helps explain why. Across the 52 organisations interviewed, tools developed with external partners reached real use in around 67% of cases, compared with around 33% for those built in-house (Challapally et al., 2025).

Tools developed with partners reached real use about twice as often as those built in-house. The authors warn that the gap may reflect the organisations, not just the path.

Buying off-the-shelf software

Buying works well when the process is the same as at thousands of other companies: meeting transcription, writing support, basic customer service, reading standard invoices.

It goes wrong when the process is what sets you apart. If the way you plan routes, assess claims or manage stock is part of your advantage, a generic tool will force you to work like everyone else.

Building with an in-house team

Building in-house pays off when AI is at the heart of the business and there is enough work to keep a team busy for years.

The trouble starts when the team has to learn everything while delivering at the same time. MIT identifies a recurring problem with in-house builds: tools that do not learn from use or adapt to the company’s context, and that end up abandoned.

Working with a specialist partner

A partner makes sense when you need to reach production quickly on a process specific to your operation and you want to own the system at the end.

It goes wrong when the partner delivers a black box and disappears. Dependency is the main risk on this path.

The right questions

For each path, there is one question that settles it:

• Buy: does the tool adapt to our process, or do we adapt to it?
• Build: do we have AI work for the next three years, or only for the next six months?
• Partner: at the end of the project, do the code, the data and the know-how stay with us?

Most companies end up using all three: they buy what is generic, work with partners on the first bespoke systems and build a team as volume justifies it. Whichever path you take, it is worth measuring in the same way: how long until something is in production, used by the team, with measured results.