Build, buy, rent, or open — sourcing is strategy, not procurement
APIs, platforms, and enterprise integration
The question: How should a company decide between an AI SaaS tool, a model API, an internal platform, or an open-source model?
AI strategy becomes procurement and architecture very quickly. You need a practical decision framework for vendors, APIs, data sharing, and lock-in — because the wrong sourcing choice compounds monthly.
What the lesson covers
APIs turn model capability into a service: your workflow sends a request (prompt + data), the provider returns a result, you pay per token or per call. This mental model unlocks everything: copilots, embedded AI features, automations, and agents are all API calls arranged around a workflow. Connectors and webhooks glue AI into the systems you already run (CRM, ERP, e-mail, docs).
The sourcing ladder: SaaS with AI baked in (fastest, least control), model APIs orchestrated by you (flexible, you own the workflow), enterprise platforms (governance and integration at a price), open-source/open-weight models (maximum control and data privacy, but you own operations, security, evaluation, and talent). None is "best" — the fit depends on strategic differentiation, data sensitivity, integration depth, speed, cost, and internal capability.
Cost management is model routing: send routine tasks to small/cheap models, hard steps to frontier models; cache what repeats; watch token economics (context length is a cost driver). Latency, privacy, and auditability constraints often decide the routing as much as price does.
Vendor claims must be tested against your workflow metrics, not their benchmarks. The procurement essentials: what happens to your data (training use? retention?), model-change policy (silent upgrades can break your prompts), logging and audit rights, residency, SLAs, liability, and exit (can you export your prompts/data/evals and switch?). Lock-in is a compounding tax.
Integration patterns to recognise: copilot (human-triggered assistance in the flow of work), embedded feature (AI inside an existing product), workflow automation (AI step inside a process), internal platform (shared gateway with governance). Start with the workflow, then choose the pattern, then the vendor — never the reverse.
Key points
- Everything is an API call around a workflow — copilots, features, automations, agents.
- Sourcing ladder: SaaS → API orchestration → enterprise platform → open models; fit follows differentiation, sensitivity, and capability.
- Model routing (small for routine, frontier for hard) + caching is the cost strategy.
- Procurement essentials: data use, model-change policy, logging, residency, SLA, liability, exit.
- Choose workflow → pattern → vendor. Never vendor first.
Framework — AI Sourcing Matrix
Score the use case on: strategic uniqueness (differentiator or commodity?), operational risk, data sensitivity, and integration depth. Commodity + low sensitivity → SaaS/rent. Differentiator + high sensitivity → build on APIs or open models with your own governance.
The lab
Evaluate vendors like an architect: scorecard, sourcing comparison, red flags.
Open this lesson, its lab and its quiz