Pandora Hands 60% of Customer Support to an AI Agent Named Gemma

Gemma, Pandora's AI concierge built on Salesforce Agentforce, now resolves 60% of routine support requests during peak traffic without a human touching them, and the jewelry brand says customer satisfaction has climbed 10% since launch.

The agent was built to absorb the seasonal spike in support volume that normally forces retailers to staff up temporarily. Instead of routing that overflow to a larger seasonal team, Pandora points it at Gemma, which handles routine requests end to end and hands off only what genuinely needs a person.

The number worth sitting with here is not the deflection rate on its own, it is that it came with a satisfaction gain rather than the usual trade-off. Most automation stories are framed as a cost play: fewer humans, cheaper support. Pandora's version pairs the cost story with a quality one, which is the harder combination to pull off and the one that actually changes minds in a boardroom.

“Most automation stories are framed as a cost play: fewer humans, cheaper support.”

It also points to something the market has been slower to notice: consumer brands are quietly ahead of B2B on agent-led customer experience. Retail has the volume, the seasonality and the tolerance for experimentation that pushes teams to ship agents into production rather than pilot them indefinitely. Enterprise software and B2B services, with their longer sales cycles and more conservative risk appetite, are still largely watching from the sidelines.

For any CX or GTM leader weighing whether agentic support is ready for a real production environment, Pandora's peak-season numbers are a useful data point: not a lab result, a live one, during the exact period when things are most likely to break.