Customer support
“Why is this fare first?”
Agents escalate to engineering or guess.
Fare display ranking transparency
Every fare
ranking,
explained.
Your ranking logic sets the order. A local language model explains each placement in one plain sentence.
2
The Iberia fare on CDG-JFK is the second-cheapest of 12 options found due to its single connection and operation by a full-service carrier.
The problem
Few teams can explain why a fare ranks where it does.
Product
“What are we optimising for?”
Rule drift goes unseen until conversion moves.
Compliance
“Can we justify this display order?”
The rationale lives with one subject-matter expert.
How it works
Facts first. Then the model.
The model only rephrases attributes your data already contains. It never ranks.
- 1
Search response in
Fares, carriers, stops and journey times.
- 2
Attribute tagging
Nonstop or connecting, LCC or full-service, longer than median.
- 3
Placement computed
“2nd cheapest of 12” is derived in code, not by the model.
- 4
Local model writes it
One grounded sentence, generated on your own infrastructure.
- 5
Surfaced per fare
A template sentence is the fallback if the model is unavailable.
Three explanation types
Different stakeholder questions, different explanations.
Why was it boosted or demoted?
Solara Airways’ fare ranks first due to its refundability for corporate travelers and being booked in a premium, less-restricted booking class.
Synthetic demo fares · rank 1 of 39
How does it compare?
The Iberia fare is the second-cheapest of 12 options due to its single connection and operation by a full-service carrier.
Real fares · CDG → JFK
What drives price?
On this route, LCC fares tend to be priced higher. A correlation across 12 fares, not proof of cause.
r = 0.25 · seen in 7 of 12 fares
Trust
Built to clear security and risk review.
Your infrastructure
No API key, no third-party inference. The model runs offline.
Never ranks
It explains the order. It cannot change it.
Graceful fallback
If the model is unavailable, a template sentence takes over.
Validated first
Tested against known rules before it was trusted.
Known limitations
- A reference implementation on synthetic data and a cached public fare feed.
- A 1.5B model can misstate details. Keep a human review step for customer-facing copy.
- Price-driver analysis is correlational, not causal. Small samples are flagged.
See it work
The whole product in 56 seconds.
Test it on your own ranking output.
A 20-minute walkthrough on a sample of your ranked results.
- You tell me where fares are ranked and which questions your stakeholders keep asking.
- I reply within 24 hours with a time that suits you.
- We run it together on a sample, and you see every explanation and what it is based on.
For airline, OTA and distribution teams, and for anyone hiring a product lead who can make AI explainable.