Build, buy, or partner? Every company that spends on AI answers this question, though few answer it deliberately, and mistakes are expensive in both directions. Our Simon-Kucher Elevate break down how to make the right call, capability by capability – and why this deserves a fresh look every year.
In the spring of 2023, Bloomberg unveiled BloombergGPT: a 50-billion-parameter language model trained on four decades of proprietary financial data. A serious feat of engineering, built by one of the few companies in the world with the data and the talent to pull it off. Within months, researchers were reporting that OpenAI's GPT-4, a general-purpose model anyone could rent for cents per query, matched or beat it on many financial tasks.
Morgan Stanley made the opposite bet that same year. It built no model at all. It licensed one from OpenAI and pointed it at the one asset no competitor could copy: a library of roughly 100,000 internal research reports. The assistant built on top of that library went out to the firm's advisors in September 2023. A year later, Morgan Stanley reported that 98 percent of its advisor teams were using it, and that the share of the research library advisors could actually reach in their daily work had risen from about 20 percent to 80.
The two firms operate in the same industry – and faced the same question at the same moment: which AI capabilities should a company own, and which should it rent? They reached opposite conclusions, and at the time each decided, each was right.
What makes the comparison uncomfortable is that most companies could not reconstruct their own reasoning if you asked them. The make-or-buy question gets answered somewhere in every organization, in IT budgets, in procurement, in a department's software subscriptions, and in many of them nobody ever treats it as the strategic decision it has become.
Where the money goes when nobody decides
The sums involved have left the rounding-error stage some time ago. Gartner expects worldwide AI spending to reach 2.59 trillion dollars in 2026, up 47 percent from the year before. Boards have largely stopped debating whether to fund AI. What they rarely discuss is what all that funding is supposed to own once the invoices are paid.
Simon-Kucher's Global Pricing Study 2025 gives a sense of the gap. 72 percent of companies said they already use artificial intelligence in pricing. Look closer, and most apply it to basic tasks such as error reduction. Among companies not using AI at all, 54 percent point to missing expertise or resources. Nearly everyone is spending. But if you ask them about the strategy, only few can say why they own one capability and rent another.
Try a simple test in your own company: ask where last year's AI budget actually went. In many organizations the honest answer is a list of generative AI tools. This means licenses, a couple of pilots, an integration project, perhaps a retainer with an agency.
A list of tools cannot tell you, though, which of these were meant as temporary rentals and which were supposed to grow into something the company owns – and that distinction is where the money is made or lost. Every one of the three paths can carry a solid business case on its own. Whether the case holds depends on whether the path fits the capability it is attached to.
Where each path wins
The three paths answer different situations:
Buy an existing tool when the capability does not set you apart: Document extraction, transcription, standard forecasting – the market solved these problems years ago, and building your own version means spending your scarcest people on something competitors handle with a purchase order. One caveat deserves more attention than it usually gets: Read the contract like an owner! Who holds the data, and what would leaving cost in two years?
Build when the capability is deeply tied to your business: Netflix has estimated that its recommendation engine, developed in-house and trained on viewing behavior only Netflix can see, is worth around one billion dollars a year in retention. A genuine build case rests on the same three elements: proprietary data, direct margin impact, and an advantage that compounds over time rather than expiring with a license.
Partner with an AI company where neither solution fits: Morgan Stanley did not need to own a language model; it needed one embedded in its research library, and it needed that quickly. Co-developing an industry data platform follows the same logic, as does bringing in a specialist team to stand up a capability that your own people then run.
In all three cases, an informed decision is made at the level of the capability rather than the company. The same organization will sensibly buy its infrastructure, build its edge, and partner where speed matters more than ownership, all at the same time.
Four questions that settle the decision
None of this requires a six-month assessment. A leadership team can work through its AI portfolio in a single structured session, capability by capability, using four central questions:
- What specific business outcome are we trying to achieve? "Reduce quote turnaround from three days to four hours" is an outcome. "We need a large language model" is a purchase in search of a justification.
- How critical is this capability to how we win? If competitors can buy the same tool tomorrow, it will not differentiate you, however impressive the demo looks. If it runs on data only you have, ownership starts to pay.
- Can we realistically deliver it? Most build ambitions fail at this question, and rarely because of the model itself. The data sits fragmented across ERP, CRM, and departmental spreadsheets, the systems were never designed to talk to each other, and the people who could fix all of that are already committed elsewhere. Bloomberg could answer yes here. Most companies cannot, and pretending otherwise is where the notorious 18-month chatbot projects come from.
- Which path, and with what commitment? Choose one and fund it properly. Three half-funded paths cost more than one funded well.
Then write the answers down. A documented rationale is what turns scattered spending into decisions you can defend in the next budget round.
Decisions age fast
The Bloomberg story has a second half. The decision to build was defensible in early 2023; by 2024, the market had turned it into a buy case. This says something about a market that moves faster than any capability plan, which is why buying or building works better as a portfolio you review than as a call you make once.
Vendors consolidate, prices fall, your own data matures, and a partnership that made sense last year becomes a purchase order this year. Measure the return on investment for each capability separately, revisit the portfolio annually, and keep your exits open. Whether you can take your data with you, and what switching would cost, together decide how expensive a change of mind will be.
Choose deliberately – with Simon-Kucher Elevate
There is no universal answer to AI buy or build. The companies that handle it well are recognizable by something modest: they can explain, capability by capability, why they own this and rent that.
At Simon-Kucher Elevate, our specialists advise on AI solutions, build them, and integrate them along all three paths. We support clients on every side of this decision from atechnology-agnostic and neutral position – which means we have no reason to steer you toward any one of them.
Talk to our experts about your AI buy or build portfolio and bring the four questions along. One session is usually a strong starting point to see where your budget has been deciding for you.
