The work
You receive AI-produced artifacts that imitate real finance-operations and audit-support output: a month-end close checklist, a bank or intercompany reconciliation in Excel, a PBC schedule, a sample-selection memo, a control walkthrough narrative, a flux analysis, or a slide deck summarizing findings for an audit committee. Your job is to score each against a rubric and explain the score. That means catching the wrong things — tie-outs that don't tie, a reconciliation that balances only because a plug was inserted, materiality thresholds applied inconsistently, accrual reversals double-counted, a control described as operating effectively with no evidence of testing, ASC/IFRS references that sound right but point to the wrong guidance. Formatting counts too: broken formulas, hardcoded numbers where a link belongs, unlabeled units, slides that bury the exception in a footnote.
Most tasks take 20–60 minutes. Written feedback is the deliverable, not the score — reviewers who only mark "inaccurate" get low quality ratings and less work.
What the screen looks for
- Verifiable experience. Five-plus years in accounting operations, internal or external audit, controllership, or shared-services finance. Expect follow-ups on what you actually owned: which close tasks, which cycles you tested, what your sample sizes and thresholds were.
- Depth under pressure. The AI interviewer will push on a claim you make — reconciliation methodology, SOX walkthroughs, revenue cutoff testing — and see whether the detail holds.
- Evaluation judgment. Can you separate a genuine error from a defensible alternative approach, and can you rank severity rather than flagging everything equally?
- Tooling. Real fluency in Excel and in Google Slides / PowerPoint, since a large share of artifacts are decks and workbooks.
An advanced degree helps but does not substitute for operating experience. CPA, CIA, or CA is a strong signal.
Logistics
Remote, hourly, contractor. Observed rates on this listing run $80–120/hr, set per project and per reviewer — not guaranteed, and often revised after a calibration period. Work is asynchronous with task-level deadlines; most reviewers commit 10–20 hours a week and pick up batches when they appear. Volume fluctuates with project cycles, so treat this as supplemental rather than a replacement for full-time income.