Project: Serverless
Problem Overview
Thomas Cook wanted to re-imagine how people search for travel/holidays. This meant creating a system for people to type "natural language" searches such as:
- "sunny family holiday 2 hours from london"
- "cheap city break next weekend"
- "scuba med 2 people 1 week in june"
Approach
Thomas Cook has the biggest range of destinations, holiday types, and filtering criteria in the industry. In order to provide the user with relevant and interesting results, the first task was to translate queries into distinct criteria on-the-fly and then apply them to the UI so the user could further refine them as desired.
Further engagement then came from “search widening”, with related searches being suggested based on the initial criteria.
Solution
The greatest technical challenge we faced was reducing query latency. It takes time to distil a query into its constituent parts and responses could take upwards of 10 seconds when we started.
To solve this, we created own proprietary algorithm which used machine learning to “learn” from users’ search behaviour. Not only did this improve our ability to iterpret travel-focused natural language queries, but it reduced response times from 10 seconds to 100 milliseconds!