Build, buy or partner: how to source AI capability

Four routes, chosen one capability at a time

There are four ways to get an AI capability. You can build it with your own staff. You can buy a product that already does the job. You can bring in a delivery partner to build it with you or for you. Or you can have a partner build a whole team, run it and transfer it to you, which is the build-operate-transfer (BOT) model.

Make the choice per capability. A single company-wide policy forces the wrong route on some of the work. Most organisations end up with a mix: products for common needs, a partner for specialised projects and in-house people for the work that sets them apart.

Four questions for each capability

Ask the same four things of every capability on your list. First, how central it is: whether it shapes what customers pay you for, or supports the business the way payroll does. Second, how fast it is needed: whether a customer, a competitor or a cost you carry today sets a deadline. Third, whether you can hire the talent: whether the people exist in your market, would join you, and would have enough interesting work after the first project to stay. Fourth, what you want to own at the end: the code, the data pipelines, the models, the team, or only the result.

Write each answer down in a sentence. The discipline matters more than any scoring scheme. If two executives give different answers on how central a capability is, settle that before you talk to any supplier.

• How central the capability is
• How fast it is needed
• Whether the talent can be hired
• What you want to own at the end

When to build in-house

Build in-house when the capability is central, you expect to keep improving it for years, and it depends on data or knowledge only you have. Owning the team means owning the learning: every release teaches your people something a supplier would otherwise keep.

Building needs conditions. You need enough continuing work to keep a team occupied, since a single project does not justify permanent hires. You need a leader who can judge technical quality. And you need time, because hiring, onboarding and a first release all come before any value. If the capability is central and the need is urgent, building is still the destination, and you may need a partner or BOT to get there.

• Central to what customers pay for
• Depends on your own data and knowledge
• Years of continuing work ahead
• Technical leadership already in place

When to buy a product

Buy when the need is common across many companies and a product already meets it. Document search, meeting notes and standard support chat are examples where most companies have much the same requirement. A product gives you speed and a supplier who carries the maintenance.

Check four things before you sign. See how much of your process must bend to fit the product. Confirm that it connects to the systems you run. Find out what happens to your data inside it, and whether that meets your privacy and compliance requirements. And work out how you would leave: what you can export and in what form. A customisable platform sits between buying and building, with a working base and changes made for your processes.

• Fit with your process
• Connection to your systems
• Data handling
• Exit and export terms

When to use a delivery partner

Use a delivery partner when the work is specific to you, the need is near, and you cannot or should not hire a permanent team for it. A partner brings people who have done similar work and can start without a hiring cycle. It suits a defined project, a first use case that tests whether AI pays in your business, or specialised skills you need for a limited period.

The risk is dependence. Reduce it in the contract and in the working method. Own the code, the documentation and the data pipelines. Have your people sit in the work, even part-time, so knowledge moves across as it is created. Ask how the partner hands over, and ask to see what a handover pack contains.

• Work specific to your business
• A near deadline
• Skills needed for a limited period
• Ownership of code and documentation
• A planned handover

When a build-operate-transfer centre makes sense

BOT answers a particular combination: the capability is central, you want to own a team at the end, and you cannot hire that team quickly yourself. A partner recruits the people, sets up the centre, runs delivery, and transfers the team and operations to you at an agreed point. You get working output during the operate phase and a permanent capability after it.

It makes sense at scale. A centre needs enough work for a full team with its own leads, and a parent organisation ready to receive it. For a handful of people, hire directly or use a partner. Define the transfer before you start: which people move, what conditions trigger it, and what support follows. Data Prophets works on this model and has built and handed over two centres, each with 80+ people.

Putting the answers together

Read your four answers in this order. If the need is common and a product fits, buy. If the work is specific to you and sits outside your core, use a partner and keep ownership of the outputs. If it is central and you have the time and the hiring reach, build. If it is central and you lack the time or the hiring reach, start with a partner for a single capability, or BOT for a whole team, with ownership passing to you by design.

Then test the decision against the end state. Picture the capability a few years on and ask who maintains it, who improves it and who holds the knowledge. If the answer makes you uneasy, change the route or change the contract. Revisit the decision when the facts change: a product matures, a project becomes central, or hiring gets easier.

• Common need, product fits: buy
• Specific, outside your core: partner
• Central, time and talent available: build
• Central, team needed fast: BOT

What to do next

List the AI capabilities you want over the next year. Answer the four questions for each in one sentence. Mark each as build, buy, partner or BOT, and note what you will own at the end. If you are unsure where to start, our free AI readiness report gives a written view within one business day.