Connect shopper questions to business results.

Conversation reports help your team inspect shopper questions, purchase concerns and conversations associated with recorded sales. Reviewing conversations with and without a recorded purchase can reveal recurring uncertainty about product fit, policies or the next step.

Use these patterns to prioritize changes to product information and the shopping experience. An automated theme or sentiment label is a starting point for review; check the underlying conversation before deciding what it means.

No. A purchase associated with a conversation shows a relationship, not what would have happened without the assistant. Shoppers who choose to chat may already have different purchase intent. Comparing people who chatted with people who did not is therefore not enough to establish uplift.

A conversation without a recorded purchase is not necessarily a lost sale either. Reporting depends on the purchase events captured, the matching rules and the attribution window.

Use random assignment to create an assistant-enabled group and a control group, and keep each shopper in the same group throughout the test. Agree the eligible audience, test duration and primary metric before comparing outcomes. Include all eligible visitors in each group, not only people who chat.

Revenue per eligible visitor helps connect the result to business value. Review uncertainty, traffic allocation and data quality before deciding whether an observed difference supports a rollout decision.

The published anonymized retail example covers July 1–31, 2026. It uses sticky 80/20 user-level assignment and measures product-detail-page visitor-days: 2,631,812 in the Gengage group and 808,779 in control. One person visiting on multiple days contributes multiple visitor-days.

• Revenue per product-page visitor-day: ₺46.88 with Gengage versus ₺38.07 in control, a rounded +23% relative lift.
• Attributed sales-line rows per product-page visitor-day: 0.873% versus 0.705%, a rounded +24% relative lift.
• Sales-line rows are matched on the same SKU/category within 24 hours. Sales-line rows are not a count of unique purchasing customers.

Treat it as a scoped case result, not a forecast or guaranteed uplift. Your catalog, audience and implementation may produce different outcomes. Agree a test for your store and review its methods and uncertainty alongside the result. Use conversation evidence to explain patterns and plan the next improvement.