Build, buy and foundations
Most organisations buy more AI than they build. This module helps you decide which is which — and what foundations you need either way.
Learning objectives
- Choose between buy, build-on-platform, build-custom and partner
- Explain why data foundations matter
- Outline key platform decisions
- Evaluate AI vendors on data, security, quality and exit
4.1 Build, buy or partner?
| Option | Best when | Watch out for |
|---|---|---|
| Buy (AI features in existing software, AI assistants) | The need is common across companies; speed matters | Lock-in, limited customisation, data terms |
| Build on a platform (model APIs, cloud AI services) | You need integration with your data and processes | Skills, ongoing maintenance, usage costs |
| Build custom (fine-tuned or own models) | It's a core differentiator and you have unique data | Cost, talent, time, responsibility for everything |
| Partner | You lack skills now but want to build capability | Knowledge transfer, dependency |
A useful rule: buy for parity, build for differentiation. Most organisations mostly buy and build on platforms, and custom-build rarely.
4.2 Data is the foundation
Most AI value depends on your own data: documents for an assistant to answer from, records for a model to learn from, context for agents to act on. Before scaling, check that data is findable, accessible, good enough in quality, and governed — with clear ownership, access rules and lawful basis for use. See Data Quality & Preparation.
4.3 Platform choices
- Model access — which providers and models, through which channels (direct API, cloud marketplace)
- Retrieval and integration — connecting models to your documents and systems (see RAG and AI Agents)
- Security and identity — single sign-on, permissions, logging, data residency
- Operations — evaluation, monitoring and cost control (see LLMOps)
Avoid betting everything on one model: capabilities and prices change quickly. Designing so you can switch models is cheap insurance.
4.4 Evaluating vendors
| Area | Questions to ask |
|---|---|
| Fit | Does it solve our use case on our data? Can we trial it with real examples? |
| Data | Is our data used for training? Where is it stored and for how long? Can we delete it? |
| Security & compliance | Certifications, access controls, audit logs, regulatory support |
| Quality & safety | How is accuracy measured? What guardrails exist? How are incidents handled? |
| Cost & exit | Pricing at our volume, price-change terms, data export, switching costs |
4.5 Skills and talent
You'll need a blend: product owners who understand the business, engineers who can integrate AI safely, data specialists, risk and legal partners, and — most of all — AI-literate staff across the business. Upskilling existing people who know the work is usually faster than hiring. See Leading AI Adoption & Change.
For three use cases, decide build, buy, build-on-platform or partner. Justify each in one sentence using differentiation, data, speed and skills.
See a worked example
Example: Meeting summaries → buy (common need, built into existing tools). Support drafting on our knowledge base → build on a platform (needs our data and CRM integration). Pricing optimisation using 10 years of sales data → build custom with a partner (differentiating, unique data, skills gap).
Key takeaways
- Buy for parity, build for differentiation.
- Your data — findable, accessible, good, governed — is the foundation.
- Design to switch models as the market changes.
- Ask vendors about data use, security, quality and exit.
- Upskilling people who know the work is often fastest.