~18 months ahead of the category and 30% more sales.

Overview

When a traveller searches "Where should I stay in Lisbon with good transport links and great food, under €100/day?", the real work starts after the results load — sifting through reviews, cross-referencing blogs, figuring out which advice is current. A consumer travel planning company wanted to skip all of that: ask one question, get a single researched recommendation good enough to be trusted.

When the feature shipped, it sold. Sales rose 30 percent, and the product sat ahead of the category for roughly eighteen months before competitors caught up.

Challenge  | Three problems that needed solving before the product could exist.

The requirement wasn't a chatbot answer. The recommendation had to read like advice from a well-travelled friend, informed by current information about places, businesses, and what real visitors had said about them. Nothing off the shelf met that bar.

• Multi-source research, not a single API call. A useful destination answer draws on many sources and must reconcile conflicting information between them. That meant designing a multi-step research pipeline, not wrapping a single API call.
• The wait had to feel earned. Generating a thorough answer takes time. If that time feels like waiting, users leave. The experience needed to make the research process visible so the wait felt earned.
• The company had to own it. Whatever got built had to be owned and operated by the client. A capability they couldn't evolve independently would become a liability the moment the market moved.

The work  | A research experience built with the founders, not handed to them.

We worked directly with the founding team — hands on the keyboard, on the same Slack, shipping together. The system uses agentic research rather than a fixed script: it plans its own steps, decomposes the traveller's question, decides what to investigate, and researches each thread against live web data through Ares, our own agentic search engine. It pulls from travel sources, local business listings, and reviews, then synthesises the findings into structured destination intelligence — not a list of links.

Outcomes  | Live with users, ahead of the category.

The system runs on infrastructure the client controls. The team got working code, documentation, and a roadmap for evolving the experience as the product grew.

About the client

Ask one question, get a recommendation that reads like a local wrote it. We partnered with a consumer travel company to build a Travel recommendation agent that researches like a human and answers like a friend, with guardrails so it never goes off-script.