What the work actually is

You are handed a snapshot of a SaaS company's revenue stack in a state of drift. A validation rule blocks what an approval flow requires. An automation still applies a discount policy that was retired in 2023. The same customer exists under two legal names, one with the billing integration attached and one with the open tickets. Permission sets grant access the documentation swears is restricted. Your job is not to click through a cleanup — it is to decide what the intended state should be, act on it, and write out why, including the cases where the right answer is to change nothing and escalate.

That written reasoning is the deliverable. Model trainers need to see the chain: which rule takes precedence, what evidence in the data layer supports the read, what breaks downstream in billing, CPQ, ticketing, or middleware if you act. A confident conclusion with no visible reasoning is worth very little here. So is a tidy summary that hides a judgment call you actually guessed at.

What the screen looks for

  • Build history, not certification history. The listing says it outright: practical build experience is prioritized. Expect probing on implementations, migrations, and consolidations you led yourself.
  • Breadth across the revenue stack. Working knowledge of at least two adjacent systems — subscription billing, CPQ, ticketing, project tracking, integration middleware — and a real sense of how changes ripple through them.
  • Comfort in the data layer. SQL, APIs, object models, cross-system reconciliation. Enough to verify a claim rather than accept the UI's version of it.
  • Written English under technical load. You will be asked to justify a precedence decision in prose. Ambiguity reads as weak judgment.

Logistics

Remote, contractor, asynchronous. Work arrives as discrete tasks rather than a standing schedule; contributors commonly report flexible part-time blocks alongside other consulting work. No prior AI experience is expected — the domain knowledge is the qualification. Screening is AI-led, conversational, and follow-up heavy: a thin first answer gets pressed, not politely accepted.