Notes from buying this stuff for a living.
Written for the person who has already been sold “AI” and still has nothing in production. No model benchmarks, no vendor cheerleading — the questions worth asking before the invoice.
01 — Blog / 7 articles
Every article, newest first.
21 topics across 7 pieces. Nothing here is gated.
- 5 min read
The hidden costs of tool sprawl in mid-size companies
The 15th SaaS login is rarely the fee. It is questionnaires, champions who leave, and no shared context. How 10–200 person companies should gate the next AI SKU.
- tool-sprawl
- saas
- procurement
- operations
- 5 min read
Data protection questions a buyer should ask any AI product vendor
A usable questionnaire on subprocessors, training, retention, human access, approval, and exit for operators and controllers at 10-200 person firms.
- security
- privacy
- procurement
- compliance
- 5 min read
When the model provider retires the model: dependency risk you can actually plan for
Model deprecation is scheduled maintenance. What breaks, who pays to retest, and which contract terms a buyer should require in writing before you buy.
- vendor-risk
- models
- procurement
- operations
- 6 min read
How to measure automation honestly when time-saved numbers are usually invented
Why vendor hour-savings fail a controller rebuild test, and how volume times minutes times loaded rate stays honest without fake case studies.
- roi
- measurement
- operations
- finance
- 8 min read
What an AI agent actually is — and how it differs from a workflow and a tool
Tool, agent, workflow, and pack in buyer language: who sits at the desk, what runs on a trigger, and what only wires systems together. No fifth type.
- taxonomy
- agents
- workflows
- tools
- packs
- 6 min read
Buy or build: an honest tally of when in-house AI work pays
When two good engineers should build the AI job, and when invoice matching is commodity work you should buy. The 20% API call is not the real cost.
- buy-vs-build
- engineering
- procurement
- operations
- 6 min read
Why most company AI pilots never reach production
Five delivery failures keep AI pilots out of the SOP. How a 10–200 person company can inspect ownership, integration, and trust before another demo.
- operations
- procurement
- ai-pilots
- delivery
