Skip to main content

The Build vs Buy Question for AI

Enterprise AI: build custom, buy off-the-shelf, or something in between? The decision framework we use with every client conversation.
10 February 2023·5 min read
Isaac Rolfe
Isaac Rolfe
Managing Director
enterprise-aistrategyprocurement
The question comes up in every enterprise AI conversation: should we build our own AI capability, buy an off-the-shelf solution, or do something in between? It's the right question. And the answer is almost always "it depends" - but it depends on specific, knowable things.

Why This Question Is Different for AI

Build vs buy isn't new. Enterprises have been making this call for decades with every category of software. But AI introduces wrinkles that change the calculus:
The technology is moving extraordinarily fast. What you build today may be obsoleted by a foundation model update in six months. What you buy today may not keep pace with what's possible tomorrow.
Data is the differentiator, not the model. The competitive advantage in enterprise AI isn't which model you use. It's the data you feed it, the processes you wrap around it, and the domain expertise you encode. This shifts the build vs buy equation significantly.
Integration is where the value lives. A standalone AI tool is a toy. An AI system integrated into your workflows, your data, your decision processes - that's where enterprise value materialises.

The Decision Framework

Buy When:

The problem is generic. If you need AI-powered email drafting, document summarisation, or meeting transcription, buy it. These are solved problems with competitive markets and mature products. Building custom solutions for generic problems is a waste of engineering time.
Speed matters more than differentiation. If you need to demonstrate AI capability quickly - to a board, to customers, to your team - buying gets you there faster. You can always build later.
You lack AI engineering capability. Building custom AI requires specific skills: ML engineering, data engineering, prompt engineering, evaluation methodology. If you don't have these skills and aren't ready to invest in them, buying is the pragmatic choice.

Build When:

The problem is specific to your domain. If your competitive advantage depends on AI understanding your specific processes, terminology, or data, off-the-shelf tools will always be generic. A claims processing AI for a New Zealand insurer has different requirements than a generic document analyser.
Data is your moat. If you have proprietary data that creates genuine competitive advantage, building AI systems that leverage that data keeps the advantage in-house. Feeding your proprietary data to a third-party tool raises both strategic and security questions.
Integration depth matters. If the AI needs to operate deeply within your existing systems - reading from multiple data sources, writing to operational systems, triggering workflows - custom builds give you the integration flexibility that off-the-shelf products rarely provide.

Compose When:

This is the option most enterprises miss. You don't have to choose between fully custom and fully off-the-shelf. The emerging pattern is composition: using foundation models (GPT-4, Claude, open-source models) as building blocks, combined with your own data, your own orchestration logic, and your own integration layer.
This is what we think will become the dominant enterprise pattern. You buy the intelligence (the model). You build the context (your data, your processes). You compose them into something that's tailored to your business without requiring you to train models from scratch.
67%
of enterprise AI budgets allocated to 'buy' in early 2023, with 'compose' emerging as a third option
Source: Gartner, Emerging Technology Survey, Q1 2023

The Questions to Ask

Before making the build vs buy decision, get clear on:
  1. Is this problem generic or specific? Generic problems have commercial solutions. Specific problems need custom work.
  2. Where does our competitive advantage come from? If AI is the advantage, build. If AI supports the advantage, buy or compose.
  3. What data does this need? If it's your proprietary data, think carefully about where it goes.
  4. How deeply does this integrate? Surface-level integration favours buying. Deep integration favours building.
  5. Do we have the team? Be honest. Custom AI development requires specific, scarce skills.

Our Position

We're seeing the "compose" pattern win most enterprise conversations, and we think that's right. The foundation models are too good and improving too fast to compete with on training. But the enterprise wrapper - data integration, governance, process alignment, domain specificity - that's where custom work creates value.
Build the wrapper. Buy the intelligence. Compose the solution.