Delta says its AI assistant helped lift satisfaction 25 points when flights break
A 25-point NPS swing during delays and cancellations, attested by the COO on an earnings call. It is the most concrete consumer-scale agent metric any company has put on the record.
- On its July 10 earnings call, Delta said its simplified rebooking, self-service tools, and Delta Concierge AI assistant drove a more than 25-point Net Promoter Score improvement during irregular operations.
- Concierge is live for more than half of Fly Delta app users, with rollout to all users completing by end of July; the beta began in October 2025.
- The digital push rode a strong quarter: revenue up 14% year over year.
Delta Air Lines put a number on what an AI assistant does for customers at the worst possible moment. On the airline's July 10 earnings call, COO Dan Janki said the company's digital tools drove a more than 25-point Net Promoter Score improvement during periods of irregular operation, the industry's term for delays and cancellations. "Our customers are noticing," Janki said, per PYMNTS.
What is actually driving the number
Precision matters here: the 25 points belong to the combined push of simplified rebooking, self-service tools, and the Delta Concierge assistant, not to the AI alone. But Concierge is the anchor of the strategy. It is live for more than half of Fly Delta app users, with rollout to all users completing by the end of July, after a beta that began in October 2025, per CX Dive. The backdrop was a strong quarter: revenue up 14% year over year and loyalty revenue up 19%.
Why this number travels
Executive-attested, consumer-scale agent metrics are still rare. Most enterprise proof points are back-office, like the deflection rates ServiceNow's customers put up. What Delta offered is different: a customer-experience number, spoken on an earnings call where misstatements have consequences, measured in the moment airlines usually lose customers. Anyone selling customer-facing agents now has a benchmark to be measured against, and anyone buying one has a question for the vendor: what is your 25 points, and does it show up when things go wrong, not just when they go right? That is also the honest test outcome-based pricing keeps promising: pay for the save, not the chat.