Cost & BoM planning.

What a Bill of Materials is, and why an AI architect produces one

A Bill of Materials (BoM) — a term from manufacturing — is the complete, itemized list of everything needed to build and run a product: parts, quantities, unit costs. For an AI platform, the BoM is every cost-bearing component: model calls, infrastructure, licenses, people, and the operational costs teams tend to forget (evaluation sampling, subject-matter-expert review time, security testing).

An architecture presented without a BoM reads as unfinished at senior level — cost is part of the design, not an afterthought bolted on after the system is built.

Worked example: a Company X-style enterprise assistant

All figures below are illustrative planning numbers with stated assumptions, not a real price quote — in an interview, say "planning numbers, to be validated against current pricing," which is itself the senior move; false precision reads as junior.

Bill of materials (illustrative, per month at moderate scale):

• Model inference (generation + embedding calls)
• Vector index hosting and storage
• Ingestion and chunking compute
• Observability/logging storage and retention
• Human review time (evaluation sampling, SME spot-checks)
• Security testing and periodic red-teaming

Unit economics: express cost per resolved query (not per token) so it's comparable to the value the query delivers, and track how that unit cost moves as usage scales — infrastructure costs subadditively, per-token model costs do not.

Presenting to a CDO/CFO: lead with the business outcome the spend buys, show the unit-economics trend (does cost per query improve with scale), and separate one-time build cost from ongoing run cost — a budget conversation that starts with GPU pricing loses the room.