Architecture review questions I use before a build vs buy call
Feature checklists hide the decision that matters: can you own this system in year two?
Teams bring fifty-row feature matrices. The decision lives in three questions: who is on call at 2 a.m., what exit costs in data and workflows, and did CX define how agents will use the output.
Year-two ownership
| Question | Pass | Fail |
|---|---|---|
| Named on-call after vendor leaves | Internal owner in org chart + budget | “Vendor will stay” |
| Index / model refresh owned | Runbook + schedule | Ad hoc tickets |
| Exit export in contract | Formats specified | Discovered at renewal |
| CX workflow walk done | Agents in the room | Feature score only |
Open-ended weeks usually mean missing criteria, not missing data.
Memos should cite pilot adoption signals: bypass, time-to-action, exception load. Pair with RAG ownership when retrieval is in scope. Schema.org metadata helps document ownership boundaries for internal platforms.
Frequently asked questions
How long should an architecture review take?
Long enough to answer year-two ownership, exit cost, and CX workflow fit, not to read a feature matrix. Deciding build vs buy before those three are clear is premature.
When is build clearly the right call?
When workflow integration is your moat and you have planned internal staff for on-call. A build CX cannot run costs more than a bad buy.
When is buy clearly the right call?
When the component is commodity, boundaries are clear, and handover and exit are contractual. The goal is not “vendor does everything”; it is “vendor does this piece, we operate the rest.”
For, CX and product leaders
Bring real workflow constraints. Technology can only buy or build what operations will run.
Paired perspective: Pilot design: what business teams must own in the first 90 daysFurther reading
- Gartner: AI production and value barriers, 48% production; 49% value barrier
- McKinsey: capturing AI value, Rewiring vs pilots
- NIST AI RMF, Risk register inputs
- Schema.org SoftwareApplication, Platform documentation
Running a program on this topic? Describe your context, technology, experience, or both.
Discuss an engagement