A travel business · Travel · AI search and discovery
Simplifying Search with AI
A travel business with thousands of product categories needed to replace a slow browse-and-filter journey. We built an Azure AI Search solution with semantic embeddings that lets customers describe what they want and returns a ranked shortlist by intent, not just keyword.
What happened?
How it went.
The problem
Our travel client had thousands of product categories and SKUs, and getting the right customer to the right product was slow and cumbersome with a traditional navigation, search and filtering journey.
We knew that each customer had a range of purchase criteria: category, price, location, number of people, and more nuanced ones such as "vibe", food and drink, live music, or family-friendly.
We wanted to replicate the offline sales experience by providing an entry point that allowed the customer to simply describe what they were looking for, and then return with a shortlist of options that made their buying decision easier.
What we built
We chose to implement Azure AI Search as the underlying technology to provide keyword and semantic capabilities to capture a customer's request and to search and rank options based on the closest fit, not just specific criteria but also their intent.
We began by creating a search index schema with standard fields for keyword search and vector embeddings. We then created APIs to pull product, pricing and availability into a single search index, then deployed an embedding model via Azure OpenAI.
Following the initial evaluation, we built custom logic to ensure that selected key criteria were always assessed (for example day and month) and identified content gaps to improve search accuracy. Finally, we built the front-end interface, which combined Suggestive Search with Azure AI Search.
The new search now searches a vast catalogue in one to two seconds and ranks results not just by keyword but also by meaning, taking customers to the right result quickly. Even ambiguous terms like "a family day out", which a traditional search would fail to match, now bring back results suitable for kids.
The next phase will introduce LLM capabilities built on the existing search index to deepen understanding of a user's search, pass through filters to product selection pages, and enable full conversational guidance, providing customers with a full online concierge experience.
Where do I start?
Close to your problem?
Tell us what the equivalent would be in your business, and we will give you an honest first read.
