⚖️AI Build vs Buy Decision Guide
ai
A structured way to decide whether to build internal AI capabilities, buy a vendor product, or partner—without hype.
Teams waste quarters debating “build vs buy” for AI while competitors ship narrow workflows. This guide is a decision tool you run in a working session—not a calculator that pretends certainty.
## When to use this
- You are evaluating an AI feature for support, ops, research, or product
- Budget and eng capacity are both constrained
- Vendors promise autonomy; you need control and cost clarity
## Step 1 — Define the job, not the model
Write one sentence: “When [user/system] needs [outcome], the system should [action] with [constraint].”
If you cannot write the job, you are not ready to build or buy.
## Step 2 — Score five factors (1–5 each)
1. **Differentiated IP** — Is this core to how you win?
2. **Data sensitivity** — Can prompts/data leave your boundary?
3. **Change rate** — Will requirements shift weekly?
4. **Ops burden** — Who owns evals, incidents, cost spikes?
5. **Time-to-value** — Need results in weeks or quarters?
## Step 3 — Interpret
- High IP + high sensitivity → bias to **build or tightly controlled buy**
- Low IP + fast time-to-value → bias to **buy**
- High change rate → avoid rigid multi-year platforms without exit ramps
## Step 4 — Pilot rules
Any pilot needs: success metric, max cost, kill date, and a human owner—not “the model.”
## DigiSone note
Buying is not failure. Unowned AI in production is.