What the work actually involves

Two things, mostly. The first is authoring: you take an analysis you genuinely shipped — the data, the question a stakeholder asked, the constraint that made it hard — and turn it into a problem a model can attempt, along with the reference answer its attempt gets graded against. The second is grading: reading model-written SQL, Python, notebooks, and pipeline code and judging whether the query answers the question that was asked, whether a join fans out and silently doubles the denominator, whether the metric the model reports means what the model claims it means.

The hardest and most valued category is the third one in the listing: results that are technically correct and still misleading. A cohort defined on a post-treatment variable. A feature that leaks the label. A conversion rate that moves because the denominator changed. An uplift figure computed on survivors. Models are increasingly good at producing runnable code and increasingly bad at noticing that the number they produced would send a decision-maker the wrong way. Your job is to name that failure precisely enough that it becomes a training signal.

What the screen is looking for

Mercor screens for shipped work over credentials — the listing says so explicitly. The AI interview will push on specific analyses you owned: what the question was, what the data looked like, what you had to decide under ambiguity, and what happened downstream. Expect follow-ups that go a level deeper than your first answer, because the interview is testing whether the detail is real. Written clarity matters as much as technical depth: most of the paid work is explaining a judgment in prose, not fixing code.

Logistics

  • Remote and asynchronous; the pool spans BI, data engineering, statistics, and ML.
  • Application is resume plus a location confirmation plus a ~20 minute AI interview.
  • This listing gives no decision. It is a standing pool, not an opening. Matching can happen within a week or take several months, depending on what clients are running.
  • The $70–120/hr band reflects rates observed on data-analysis projects sourced from this pool. Your actual rate, hours, and client are named on the specific project listing you get invited to, and the hiring decision happens there.
  • Project work is typically part-time and flexible, often 5–20 hours a week alongside a full-time role.