Search is the front door of retailing. If you rank poorly, prune aggressively, or reprice too late, the traveler never sees the right option. This breakdown covers the three major levers: ranking, pruning, and repricing.
Ranking that reflects traveler intent
Ranking should balance commercial goals with traveler signals. Start with features you can explain: price, duration, number of stops, fare conditions, loyalty value. Layer machine learning models only after you have solid baselines. Keep an explanation object alongside each offer so you can audit why it appeared first.
Prune with rules, not guesswork
- Set hard limits per search (e.g., maximum itineraries per cabin) to protect latency.
- Apply policy filters early: remove ineligible fare brands for corporate travelers, block combinations that violate regulatory rules.
- Monitor pruning ratios-if you throw away too many offers, you might be overfitting to historical behavior.
Repricing discipline
Repricing is mandatory before you present the final selection or commit an order. Define when it happens:
- Pre-display validation: Confirm fare, tax, and availability for the top-ranked offers.
- Pre-commit repricing: Revalidate the chosen offer using fresh availability and payment rules.
- Post-commit audit: Store the repriced snapshot in the order so you can defend the amount later.
- Average search latency split by component (cache, host, pricing).
- Drop-off rate after repricing errors.
- Offer diversity (number of unique fare brands, cabins) in the top results.
Great search architecture makes the rest of retailing easier. Travelers see relevant options, operations trust the prices, and analytics receives consistent data to refine ranking further.