Quick Answer
The UK build-vs-buy AI decision turns on three questions: is the capability generic (buy) or domain-specific (consider build), do you have proprietary data that sharpens a custom model, and do you have the ML engineering capability to sustain what you build? The pragmatic default for most UK mid-market businesses is hybrid — buy foundations, build thin high-value layers on top.
Key Takeaways
- 01Generic capabilities (email drafting, meeting summaries) should almost always be bought
- 02Domain-specific capability tied to proprietary data is the build case to evaluate
- 03Ongoing operational cost is typically 25–35% of initial build cost per year
- 04Hybrid (buy foundations, build thin high-value layers) is the default for UK mid-market
- 05UK ML engineering talent is scarce — build only if you can sustain the team
The build versus buy decision has become one of the most consequential choices UK business leaders make in 2026. Build the wrong thing and you burn budget and credibility. Buy something that looks like a fit but cannot flex to your business and you inherit limitations that slow you down for years. The framework below helps UK executives make the decision with clear eyes.
Start by characterising the capability you are considering. Is it a generic business function like meeting summarisation or email drafting, where the best commercial tools are already excellent? Or is it a domain-specific capability tied to your proprietary data, customer relationships or operating model? Generic capabilities should almost always be bought, because the market will out-invest any individual business. Domain-specific capabilities are where the build case becomes interesting, particularly where the capability could create defensible competitive advantage.
The second dimension is data. Do you have, or can you realistically acquire, high-quality training and evaluation data that a commercial vendor would not have access to? If yes, you have the raw material for a genuinely differentiated build. If not, a commercial product trained on a much broader dataset will almost certainly outperform anything you could build with limited data. The third dimension is organisational capability: do you have or can you hire the machine learning engineers, data scientists and AI product managers required to build and sustain the capability? Many UK organisations underestimate the ongoing operational cost of owning AI systems.
A pragmatic default is the hybrid approach. Buy commercial platforms for the foundations, whether that is OpenAI, Anthropic, Azure AI or managed services from specialist UK partners. Build thin, high-value layers on top where your data and domain expertise create real differentiation. This hybrid pattern is how most of the leading UK adopters we see in 2026 are organising their AI estate. It gives you the speed of buy with the defensibility of build, and it keeps your technology strategy aligned with your commercial strategy rather than your internal ego.
Frequently Asked Questions
FAQ
Common questions
When three things are true simultaneously. One: the capability is tied to proprietary data or workflow that commercial products don't capture. Two: getting it right would create genuine competitive advantage that compounds over years. Three: you have or can sustainably hire the ML engineering capability to run it in production. If any one of the three is missing, buy commercial tooling and redirect the energy to something where all three conditions are met. Most UK mid-market businesses have one or two genuine build candidates — not ten. The organisations that over-build typically underestimate operational cost and end up with expensive legacy systems within three years.
Initial build for a well-scoped custom capability typically runs £50–250K depending on complexity. Ongoing operational cost — model updates, evaluation, monitoring, incremental improvement, incident response — runs 25–35% of initial build cost per year. For a £150K build, plan £40–50K per year ongoing for at least three years. UK ML engineering salaries are £80–140K for mid-senior engineers, with strong competition from finance and big tech. Sustaining a team of two or three engineers costs £300–500K per year fully loaded, which only makes sense if the custom capability is producing clear annual value above that threshold.
Commercial foundations (OpenAI, Anthropic, Azure OpenAI, Google Vertex AI, or a managed platform like WayaNerd Growth at £299/month) handle the generic layers — language understanding, content generation, data extraction. On top of those foundations you build thin, high-value layers: your retrieval system over proprietary data, your evaluation framework for domain-specific quality, your integration with unique internal systems, and your user experience for the specific workflow. The commercial foundation gets the economies of scale; the thin build layer captures your unique advantage. Most UK hybrid deployments put 70–80% of code into the thin layer and rely on commercial vendors for the heavy foundation work.
For most UK SMEs, most of the time. Off-the-shelf AI — WayaNerd's Starter and Growth plans, Microsoft Copilot, Zapier AI, HubSpot AI features, document automation vendors — covers 70–80% of realistic AI value for under-100-employee organisations. Build is rarely right at that scale because operational burden outweighs incremental value. Buy-only is also right for mid-market organisations in commoditised verticals where proprietary data isn't a real advantage. The build case strengthens as organisations grow past 500 employees and into sectors with regulatory complexity or proprietary data at scale. Below that, buy wins on both cost and risk.
Three practices. One: keep the prompts, retrieval data and evaluation framework in your own tooling rather than the vendor's — if the vendor becomes problematic, you can port to another foundation model in weeks rather than rebuilding from scratch. Two: insist on data portability in the contract — you should be able to export your customer records, conversation logs and training data in a standard format. Three: maintain relationships with at least two foundation-model providers (for example OpenAI and Anthropic, both with EU endpoints) so you're never dependent on a single vendor's pricing or terms. WayaNerd deployments are designed to be portable by default — data, prompts and retrieval stay in your own cloud tenancy where possible.
Keep reading
When Off-the-Shelf AI Isn't Enough: UK Custom Development Guide
A guide for UK business leaders on when off-the-shelf AI reaches its limits and how to approach custom AI development without losing control of cost.
10 min read
StrategyUK AI Strategy: Building Your 2026-2028 Roadmap
A structured approach to building a three-year AI roadmap for organisations, balancing quick wins with strategic transformation.
11 min read
StrategyAI Readiness for UK Businesses: A 5-Step Assessment
A five-step readiness assessment that UK business leaders can use to honestly evaluate whether their organisation is ready to scale AI in 2026.
9 min read