Build vs Buy vs Blend: Deciding Where AI Belongs in Your Business

"Should we build this ourselves or buy it?" — for AI, the old binary is obsolete. The answer is almost always blend, and the companies that get the blend right save multiples of what they spend on the decision. Here's the framework I use when clients ask me this, refined over years of building both sides.

Why the old binary fails for AI

Traditional build-vs-buy assumed the thing you're buying is stable — a CRM is a CRM. AI products are not stable. Models change under you monthly, pricing swings wildly, and today's "impossible" feature becomes next quarter's API endpoint. This creates a third option that didn't use to exist: rent the capability, own the integration.

The three-layer map

When I assess an AI opportunity, I draw three layers:

Layer 1 — The model. The raw intelligence. Never build this. Even companies with billion-dollar budgets rent from model providers, and the gap between "frontier model" and "your fine-tune" keeps shrinking in favor of renting. The only exceptions are extreme data-sensitivity or extreme scale, and both are rarer than vendors claim.

Layer 2 — The capability. "Answer questions from our contracts." "Qualify inbound leads." "Draft support replies." This is where vendors sell — and where the build/buy question actually lives. The honest answer: buy the boring ones, build the differentiating ones. If everyone in your industry needs the same capability, a product exists or will exist; your advantage isn't there. If the capability encodes your specific process, data, or edge — that's where building pays.

Layer 3 — The integration. How the capability connects to your actual business: your data, your workflows, your team's habits. Always own this. This is also where every successful AI project actually lives. I've watched companies buy a capable AI product and fail anyway — because nobody owned the integration layer: the data was a mess, the workflow contradicted the tool, the team was never brought along.

The decision questions that matter

For any specific capability, I run through five questions:

1. Is this a competitive edge for us, or table stakes? Table stakes → buy. Edge → build.
2. How often does the requirement change? Requirements that shift weekly favor building on flexible primitives; stable, well-understood problems favor products.
3. What's our data situation? Proprietary data is the strongest argument for building — no vendor can replicate your context. No clean data is an argument for buying later: clean up first.
4. Who maintains it? Building means owning maintenance forever. Budget for it honestly — AI systems need monitoring and re-evaluation in a way traditional software doesn't (shipped is not running).
5. What's the exit cost? For each option: how expensive is it to walk away in a year? Cheap exits reduce the cost of being wrong.

A realistic example of the blend

A services company I advised wanted "AI that handles client inquiries." The naive path: buy an AI chatbot, done. The blend that actually worked:

• Bought the model access and the voice/telephony infrastructure (commodity)
• Built the intake logic that classified inquiries against their service catalog (their edge, their data)
• Owned the escalation rules deciding when a human takes over (their risk policy)

Total build effort: a few weeks. The bought pieces did the heavy lifting; the built pieces carried the differentiation; the owned integration made it actually theirs.

The trap to avoid

The most expensive mistake I see: buying a product to solve a problem the company doesn't understand yet. If you can't specify what success looks like numerically, you're not ready to buy or build — you're ready to spend a cheap week defining the problem (discovery is the project). That definition week is the highest-ROI AI investment that exists.

---

If you're staring at this decision right now, a consulting day compresses it: we map your opportunity across the three layers, run the five questions on your real candidates, and you leave with a written recommendation and cost model. Book one, or describe your situation first if you prefer.