Galactic Air priced add-ons per shopper, but only below the human price

Airlines increasingly earn money from ancillaries, but many still price bags, seats, and extras from static rules aimed at average customer groups. That misses the context of each shopping session.

Deepair and a major European airline, anonymized as Galactic Air, tested a different approach: dynamic personalized prices based on flight and customer-session features. The model did not have full freedom; its offers had to stay below the human-curated price for the same customer group.

That constraint is the elegant part. It turned machine learning into a controlled discounting engine rather than an opaque price-raising machine, making online A/B testing easier to accept.

Over a six-month live implementation, the system increased ancillary revenue per offer by 25% and conversion by 15% versus human-curated rule-based prices.

Personalize prices with a fairness guardrail

It changed ancillary pricing from segment averages to session-specific offers.

The lower-than-manual cap limited reputational and customer-service risk.

Live A/B testing measured the effect against the airline's existing pricing process.

It improved conversion and revenue at the same time, suggesting the old prices were often too blunt.

Dynamic pricing lands more safely when it includes a business guardrail. A capped optimizer can learn where the old rule is too high without creating new price spikes.

The published 2022 paper reports that the best-performing models went live in the airline booking system for six months, raising revenue per offer by 25% and conversion by 15% against the incumbent rule-based process.