Khrystyna Volianska, home
← All case studies

Case study · Quality · Client products · 2023–2026

The refund reason that didn’t exist: fixing supply and quality, not the policy, cut refunds from 15% to 7% and kept ~$0.7M a year

Head of Product · with Sales and Operations · 2023–2026

Situation

Clients paid upfront for every order. 15% of orders ended in a refund: revenue lost, and a client who didn’t get what they paid for.

Complication

The dashboard named “cancelled by the client” as the top refund reason, so the fix on the table was a stricter refund policy. Before deciding, I tested the dashboard against the source: the chat logs of refunded orders. Genuine client cancellations barely existed. Most of those orders had found no qualified freelancer in time, and the next biggest cause was quality. A stricter policy would have treated the symptom of a supply problem.

Exhibit

The dashboard’s top refund reason barely existed; two operational causes did, and each got its own fix

  • Why did 15% of orders end in a refund? → No qualified freelancer in time (largest cause) → Specialist supply in the worst-hit subjects, routed by expertise
  • Why did 15% of orders end in a refund? → Quality below standard (second cause) → Automated pre-checks and a quality-verification layer before delivery
  • Why did 15% of orders end in a refund? → Genuine client cancellation (the dashboard’s top reason; barely existed) → No fix needed; policy redesigned last, with financial controls (not a real cause)
Causes re-classified from the chat logs of refunded orders.

Decision

Fix each cause first, and the policy last. A stricter policy would have cut refunds on paper and pushed the cost into chargebacks, complaints and lost repeat clients. It would also have charged clients for the platform’s own failure to deliver. So every root cause got its own fix, and the policy was redesigned only once the causes were under control.

What I did

  1. 01

    Fixed the measurement. Refund reasons were re-classified from the chat logs into causes (no freelancer in time, quality, genuine cancellation), so the dashboard reported why orders failed, not how they ended.

  2. 02

    Closed the supply gap. We rebuilt freelancer acquisition around the few subjects with the most “no freelancer” refunds, hired specialists in those fields and routed orders to them by expertise.

  3. 03

    Caught quality problems before the client did. Automated pre-checks ran on every delivery (plagiarism score, word count), with a quality-verification layer before the order reached the client.

  4. 04

    Redesigned the policy last. Store balance became the default refund, backed by financial controls: daily refund caps, per-operator limits on refunds and balance changes, alerting and a transaction-monitoring dashboard.

What didn’t work

The first pre-check thresholds were the same for every subject and flagged too many good orders, which slowed delivery and overloaded Operations. Setting thresholds per subject fixed it.

Impact

  • Refund rate 15% → 7%, worth ~$0.7M a year.
  • Both root causes fixed at the source: specialist supply where orders went unfilled, and quality caught before delivery.
  • Pre-checks, specialist supply and controls run on every order: a capability, not a one-off fix.

Learning

Check the data before you change the policy. The biggest refund reason on our dashboard didn’t exist.

Loading the calendar…