Boosting Gift Card Conversions by 12.5% with an AI Search Assistant

Solution

We wanted to quickly make sure that our idea worked, so we started with AI MVP development. We built a working prototype in three days using Replit, a platform where we could describe requirements in plain language and have AI generate code. This let us skip setup and infrastructure, so we could launch a live version for real users to try – all that at a cost of just $25. Since the AI assistant prototype ran in the browser, we didn’t need to set up servers or cloud environments. We could share a demo link right away, get feedback, make changes, and test again within minutes.

Next, we worked on validating the conversational AI in a way that could grow beyond a simple prototype. Instead of building our own model infrastructure, we used AWS Bedrock, which gives access to several large language models through one interface and is built for understanding intent and processing text efficiently. This choice saved setup time, avoided early vendor lock-in, and made it easier to move from testing to controlled rollout. In about six weeks, we went from the first demo to A/B testing, collecting real data on how well the assistant helped users who didn’t know what to search for and how it guided them to the right gift options.

We built the AI search assistant with several connected features. At its core is a semantic search layer that uses embeddings, so both user queries and products are mapped as vectors in the same space. This lets the system match by meaning, not just exact words, and understand full sentences, vague descriptions, and unusual phrasing. On top of this is a comprehension block that focuses on intent and can handle typos, slang, and incomplete queries. Finally, the Gift Advisor layer turns search results into helpful guidance, suggesting how each gift card can be used and presenting them as thoughtful ideas instead of just products.

We added small but important features to make the experience feel more natural. The system can read emojis as part of intent, which is especially useful on mobile. For instance, emojis of a heart and chocolate sent together are interpreted as a romantic vibe, and the search results adjust accordingly. At the same time, Gift Advisor adds context to brands like HelloFresh, presented as a meal kit option, and Sephora, as a beauty and self-care brand, so users quickly see the idea behind the possible present. Plus, the assistant suggests related categories, letting shoppers explore and combine cards into bundles. All these features turn search into an intent-driven assistant that helps customers go from a vague idea to a confident choice.