Whatnot — Customer Experience — Product Case Questions

Role context: Data Scientist, Customer Experience (live commerce marketplace) · Est. study time: 60 min · 5 questions

How to approach product cases here

Every case is one move in the same chain: understand the goal, break it into a data problem, choose the metric or method, name the bias and the trade-off, land on a decision and an action.

Five things make Whatnot CX cases distinctive.

1. Contacts are a selected sample, and this is the defining bias. Roughly 87% of bad experiences never generate a contact. The people who complain are not the people who were harmed, and they skew toward high-value transactions and articulate users who believe complaining works. Ranking work by contact volume serves the loud and ignores the churning.

2. Support is a cost centre delivering value it cannot claim. Agent hours are in a system. Retention saved by a good resolution is a counterfactual. That gap is the role: measuring the tradeoffs between user experience and business value.

3. Both sides are customers. A dispute is between a buyer and a seller, and 1 in 8 sellers have Whatnot as their primary income. Every resolution is a transfer. There is no outcome where everyone is happy.

4. There is no ground truth on disputes. Nobody photographed the card at packing. Any model here is trained on prior decisions, not on truth, so it learns to reproduce the org's existing bias.

5. Support is being pushed onto sellers. SPS expanded through 2026 with all US sellers enrolled from September 1, 2026. The accused is increasingly the judge, and SPS-closed cases have no independent label.

The traps: reporting CSAT, celebrating deflection, ranking by contact volume, and claiming a retention number CX cannot defend.

The five below weight diagnosis, measurement design, estimation under selection bias, impact measurement where the counterfactual is unobservable, and capacity forecasting. If you only have time for two, do bc1 and bc2.