What the work involves

This is contract evaluation work, not advisory work. On a typical task you'll either author an accounting scenario drawn from your own practice — a month-end close with an accrual cutoff issue, a bank reconciliation with unrecorded items, a Schedule C with mixed-use asset questions, a substantive analytical procedure on revenue — or you'll review what a model produced against it. Authoring means writing the fact pattern, the correct treatment, and the reasoning an accountant would actually follow, including the judgment calls where a defensible answer differs from a technically literal one.

Evaluation tasks usually mean comparing two or more model responses and ranking them, then explaining the ranking in writing. The written justification is the product. Reviewers want to know why an entry is wrong — wrong period, wrong account, wrong standard, right answer with unsupported reasoning — not just that it's wrong. Contributors who write "incorrect, see ASC 606" get less work than those who cite the specific criterion the model skipped.

What the platform screens for

Mercor's screening runs through an AI-led interview plus a résumé parse, and often a short sample task. It's checking that your credential is real and current, that your stated specialty holds up under follow-up questions, and that you can explain accounting reasoning in clear prose. Expect to be pushed on specifics: which standard, which threshold, what you'd do when the client's records don't support the position. Generic answers about "ensuring compliance" read poorly. Prior AI or data-labeling experience is not required and is not what's being measured.

Logistics

  • Fully remote, asynchronous, 1099 contract — no fixed shifts
  • Observed rates for this listing run $80–120/hr, varying by specialty depth and credential; rates are as reported, not guaranteed
  • Most contributors work 10–20 hrs/week; some projects offer more volume in bursts
  • Work is compatible with an active practice, though conflict-of-interest and confidentiality rules mean scenarios must be constructed, not lifted from client files
  • Strong contributors are moved into reviewer and domain-lead tiers at higher rates