Most enterprise AI use cases are now bought rather than built, and the research mostly supports that default. But buying is not always right. A custom tool wins when the work is specific to your business, when your data or workflow is unusual, or when the tools on the market cannot meet your security requirements.
Short answer: buy when an existing tool already covers most of what you need and the process is the same as everyone else's. Configure or extend a bought tool when it gets you most of the way but needs your data or systems. Build when AI is part of how you compete, when your workflow is unusual, or when nothing on the market fits. Whichever you choose, make sure you own what is built for you.
The research leans toward buying
Three sources point the same way.
Menlo Ventures' 2025 survey of 495 US enterprise AI decision makers found that "76% of AI use cases are purchased rather than built internally." In 2024 the split was close to even: "47% of AI solutions were built internally, 53% purchased." Menlo is a venture capital firm, so read its numbers with that in mind.
MIT NANDA's July 2025 report, The GenAI Divide, found that "external partnerships with learning-capable, customized tools reached deployment ~67% of the time, compared to ~33% for internally built tools." The authors are careful about this. The figures come from interviews with 52 organizations, are self-reported, and "the correlation between external partnerships and success does not necessarily prove causation."
Gartner made a similar call in March 2025, forecasting that CIOs "will reduce POC and self-development efforts, focusing instead on GenAI features from existing software providers." In July 2024 it predicted that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025, citing "poor data quality, inadequate risk controls, escalating costs or unclear business value."
Note what the MIT finding actually says. The tools that did best came from external partners but were "learning-capable" and "customized." The report also says generic enterprise tools lose to "deeply customized alternatives."
Build and buy are not the only two options
The real choice is a spectrum. It helps to name three points on it.
| Option | What it means | Good fit when |
|---|---|---|
| Buy | Use an off-the-shelf AI product as it comes | The task is common and the tool already fits most of the need |
| Configure and extend | Connect a model or product to your data, prompts and systems | The tool fits partly, but needs your documents, CRM or rules |
| Build | A custom application designed around your users and workflow | The workflow or data is unusual, or AI is part of your edge |
The middle option is where much of the practical work happens. OpenAI's own guidance on improving model accuracy says "prompt engineering is typically the best place to start." The next step is retrieval augmented generation, which "is used to give the model access to domain-specific context to solve a task." Fine tuning, continuing to train a model "on a smaller, domain-specific dataset," comes later and only for specific problems. In other words, most custom AI does not start with training a model. It starts with good instructions, your data, and connections to your systems.
When buying wins
Buy when the process looks the same at every company: meeting notes, general writing help, standard support ticket triage, transcription. Vendors spread their development cost across many customers, ship updates you would never fund yourself, and carry the maintenance.
Buy also when speed matters more than fit. McKinsey's 2024 State of AI survey found that highly customized or proprietary models were "1.5 times more likely than off-the-shelf, publicly available models to take five months or more to implement."
When building wins
A custom tool earns its cost in a few situations.
- The workflow is yours alone. Quoting with complex pricing rules, underwriting checks, or a review process no vendor has modeled.
- AI is part of your product or your edge. If customers choose you because of how the AI works, renting the same tool as your competitors gives that away.
- Your data is unusual or sensitive. Internal documents, legacy systems, or records that cannot leave your environment.
- Off-the-shelf tools fail your security review. Hosting, access control or audit needs that a SaaS product cannot meet.
- Per-seat pricing stops making sense. At high volume, a tool built on model APIs can cost less than licenses for every user. Run the numbers for your case.
Hidden costs on both sides
Buying has costs that do not show on the price page: integration work, data you must export, features you cannot change, and a roadmap someone else controls. Building has its own: evaluation, monitoring, model updates and someone to maintain it after launch. Gartner analyst Rita Sallam said in 2024 that with generative AI, "costs aren't as predictable as other technologies."
There is also a risk in the middle. In September 2026 Gartner predicted that "by 2028, 70% of enterprises will abandon agentic AI built by vendor forward-deployed engineering," warning that such engagements "often fail structurally before they fail technically." If someone builds for you, insist on owning the code, prompts and configuration, on documentation, and on a handover your team can run.
Five questions to decide
- Does an existing tool cover most of what you need today, without workarounds?
- Is this process the same at every company, or is it how you win?
- What data does the AI need, and can it leave your systems?
- Who will own, monitor and update the tool after launch?
- What does each path cost over three years, including integration, usage and maintenance?
If you are not sure which processes deserve AI at all, start with an inventory. Our guide on how to run an AI automation audit walks through ranking work by hours and fit.
How Lean Discovery Group helps
Our rule of thumb is simple: buy when a tool already fits 80 percent of your needs, and build when AI is part of your competitive advantage, when your data or workflow is unusual, or when off-the-shelf tools cannot meet your security requirements.
When building is the right call, our custom AI development team designs and ships AI applications, AI features inside existing products, and internal tools. Our process runs through discovery, design, build, evaluation, and launch and support, and you own all code, prompts and configuration we create. If you want help making the decision first, our AI consulting services start with an automation audit, and you can browse all of our AI automation services. Book a discovery call to talk it through.
Frequently asked questions
Should I build or buy an AI tool?
Buy when an existing tool covers most of your needs and the process is common. Build when the workflow or data is unusual, when AI is part of your competitive edge, or when no tool meets your security requirements.
Is it cheaper to build or buy AI?
Buying is usually cheaper to start. Building can cost less at high volume or when license fees grow with headcount. Compare three-year totals, including integration, usage and maintenance.
Do bought AI tools succeed more often than custom ones?
In MIT NANDA's 2025 interviews, customized tools from external partners reached deployment about twice as often as internal builds. The authors note the results are self-reported and do not prove cause.
What is the middle option between building and buying?
Configure and extend: connect a model or product to your data and systems with prompts and retrieval augmented generation. OpenAI's guidance treats prompting as the place to start and fine tuning as a later step.
Who owns a custom AI tool built by an agency?
It depends on the contract. Ask for ownership of the code, prompts and configuration, plus documentation and a handover, so you are not locked in to the builder.
Sources
- Menlo Ventures, "2025: The State of Generative AI in the Enterprise," December 9, 2025
- MIT NANDA, "The GenAI Divide: State of AI in Business 2025," July 2025 (copy hosted by Webber Wentzel)
- Gartner, "Gartner Forecasts Worldwide GenAI Spending to Reach $644 Billion in 2025," March 31, 2025
- Gartner, "Gartner Predicts 30% of Generative AI Projects Will Be Abandoned After Proof of Concept By End of 2025," July 29, 2024
- Gartner, "Gartner Predicts 70% of Enterprises Will Abandon Agentic AI Built by Vendor Forward-Deployed Engineering by 2028," September 29, 2026
- OpenAI, "Optimizing LLM Accuracy," OpenAI API documentation
- McKinsey, "The state of AI in early 2024," May 30, 2024
