What the work actually involves

Mercor is staffing this funnel on behalf of an AI lab that wants insurance reasoning to hold up under professional scrutiny. Day to day you will be doing two things in rotation: authoring tasks and judging outputs. Authoring means taking a decision you genuinely make at work — a submission you declined and why, a coverage question that turned on one endorsement, a reserve you strengthened mid-litigation, a rate filing objection you answered — and rewriting it as a self-contained scenario with a defensible answer key. Judging means reading two or more model responses to such a scenario and explaining, in writing, which reasoning is sound and where the other went wrong: a misread exclusion, a duty-to-defend conclusion that skips the pleadings, a trend selection nobody could support, a compliance claim that cites the wrong state's standard.

The written feedback is the deliverable that matters most. Researchers are not insurance people; a rubric line that says "incorrect coverage analysis" is unusable, while one that says "applies the professional services exclusion to a bodily injury claim arising from premises conditions, which the exclusion does not reach" is training signal. Expect to defend your rating when a reviewer disagrees.

What the platform screens for

  • Verifiable specifics. Which line of business, which side of the desk (carrier, broker, reinsurer, advisory), which states or jurisdictions, which systems — Guidewire, Duck Creek, Sapiens, Applied Epic, ISO forms, in-house rating engines.
  • Depth under follow-up. The screen asks a follow-up to whatever you claim. Claimed reserving experience gets a question about case versus bulk; claimed E&S underwriting gets a question about diligence and filing exemption.
  • Evaluation judgment. Whether you can separate "answer I disagree with" from "answer that is wrong," and whether you can grade a response that reaches the right outcome by bad reasoning.
  • Credentials as stated. CPCU, ARM, AINS, CLU, AIC, ACAS/FCAS, ASA/FSA, or a state adjuster or broker license are preferred, not required. Three or more years of insurance experience is the stated bar.

Logistics

Fully remote and asynchronous — no scheduled shifts, no client calls. Contributors typically commit a stated weekly band and draw tasks from a queue; volume fluctuates with the lab's project cycles, so treat it as variable contract work rather than a guaranteed load. Hourly pay at $75 reflects what has been observed on this funnel and is not a guarantee across cohorts or specialties. The application path is resume or experience summary, a practice-area and certification form, a possible short sample task, and follow-up usually within a few days. Strong contributors are moved into reviewer, lead, and domain-expert tracks.