At the time, most Amazon customers arrived on our site with a very specific product in mind, searching for it and purchasing it — a behavior we referred to as "spearfishing."
The customers I designed for, however, were the opposite of "spearfishers." They were in the early stages of their shopping journey, and the broad query experiences aimed to provide them with relevant information and navigational actions to help them quickly and efficiently narrow down their options.
Broad queries accounted for 23% of all Rufus queries — while not the primary use case on Amazon, they represented a significant unmet need, largely because Amazon had not historically supported early-stage product research. Some common examples included:
A customer who came to Amazon to buy a TV but didn’t know what to look for to ensure it was a good option.
A customer who wanted to browse available TVs but felt overwhelmed by the vast selection.
A customer who had a specific type of TV in mind but wanted to explore other options.
In all of these cases, conversations with Rufus were designed to be insightful, succinct, visually rich, interactive, and to help customers discover and evaluate products more easily while shopping on Amazon.
Broad query - An upper funnel search where the customer is still in the research phase of shopping and has not narrowed down to a few options.
i.e. “Sneakers”
Narrow query - A lower funnel search where the customer has narrowed down to a few options or a single product.
i.e. “Nike Alphafly 3 Women's Size 7”
I worked closely with my product partner to influence the product direction, and identified these 4 themes as the key areas to tackle in broad query search responses:
Drive adoption by surfacing high quality broad query experiences at the right stage of the customer journey
Helpfulness of suggestions
Keep responses concise and actionable
Keep defect rate low
Breaking down the response
Answer the question
Start with a concise LLM-generated text response directly addressing the customer's query.
Keep it shoppable
Insert shoppable components such as category cards or product recommendations.
Follow up questions
End with related questions that keep the conversation going, and teach the customer how to talk to Rufus in the future.
After testing in multiple user research studies, and working with product, engineering, and science partners, we launched our initial set of broad query experiences in March 2024.