Barun Kwak
Barun Kwak

Rufus Beta Launch, Amazon's Generative Shopping AI at Amazon

*This project is confidential, and certain details cannot be published.

Background

Design features with new generative AI-powered capabilities in Amazon’s store to make shopping even easier and more convenient for customers.

Rufus is a generative AI-powered expert shopping assistant trained on Amazon’s extensive product catalog, customer reviews, community Q&As, and information from across the web to answer customer questions on a variety of shopping needs and products, provide comparisons, and make recommendations based on conversational context.

Fun fact: Rufus is named after a beloved Welsh corgi who roamed the first Amazon warehouses and apparently loved meetings. :)

My Role in Rufus

I led design end-to-end for the following Rufus generative AI/LLM features, working with product, engineering, research, and science partners to define an MLP and long term vision to work backwards from:

  1. The AI default response when a user swipes up on the bottom sheet after they type in a keyword like "coffee maker" and taken to a search results page with product results.

  2. The AI response for when a user types a simple keyword query. i.e. "coffee maker" or "air fryer"

  3. The AI response for when a user types a broad research query, i.e. "what to look for in a coffee machine" or "what are different types of air fryers"

Design Process

Project Kickoff

I work closely with my project manager to define scope of work and working backwards plans from launch timelines & prioritization from our C-suite. At this phase I determine the action items for design & eventually what are our ideal outcomes, plan upcoming research, and schedule leadership reviews.

Design Iteration

After defining scope, I work on an initial design proposal that I vet with science partners to determine LLM feasibility, otherwise define iterative steps to get to the long term vision so that the model can start learning new patterns and behaviors for answering shopping questions depending on customer queries.

For example, a recent Rufus project I’ve worked on is creating a taxonomy of how the AI should produce actionable responses depending on query specificity. I defined logic for how the AI should interpret different shopping queries and respond accordingly depending on where the customer is in their shopping journey. For broader queries customers may need more of a high-level guided experience, whereas for narrower queries customers may be ready to see direct product recommendations. In either situation, each experience brings customers closer to buying their desired product.

In order to have these types of forked experiences, I work with science partners to define what expected responses are, and how we can start to route the right ones to customers.

Usability Research

I strengthen design rationale with usability research studies working with research partners and external Amazon customers to get insights on features.

Leadership Review

Once designs are finalized, they are presented to leadership, from VP to CEO level depending on priority of the feature and once aligned on direction, I work with engineering and product to execute designs in the beta app that are later dialed up to prod.

Initial Beta Launch Metrics & Press

On Thursday, 2/1, we announced Rufus, a generative AI-powered expert shopping assistant becoming available in Amazon’s mobile app. We broke the news on About Amazon with an exclusive article authored by Rajiv Mehta and Trishul Chilimbi, and conducted broad outreach to reporters across business, retail, tech, and local publications. We reinforced Rufus during earnings later in the day, and amplified the news through Amazon and executive social channels.

Key Metrics: Our About Amazon post garnered 33k page views (+810% above benchmark) in the first 26 hours, outperforming other media publications (CNBC had 5k page views and TechCrunch had 3k page views). Organic Google search and Google News drove a strong influx of readers (8k page views), where we ranked #1 in the newsfeed and for relevant keywords like “Amazon AI,” “Amazon Rufus,” and “Amazon.” The majority of other readers came from executive social content on LinkedIn (6.7k page views, including posts from Andy Jassy and Doug Herrington), direct sources (14k page views, including third-party AI newsletters and news-generator apps like Apple News), and backlinks from media outlets (1.1k page views, including CNBC, Engadget, TechCrunch, and The Verge).

External Media Highlights:

Internal Media Highlights:

Social Media Highlights:

Executive Social Highlights

Top Quotes:

  • “And the company is now looking to give customers a new way to shop, while signaling to investors that it is doubling down on artificial intelligence. Ahead of the earnings report on Thursday, Amazon announced it was adding an AI-powered shopping assistant dubbed “Rufus” to its e-commerce store.” – Catherine Thorbecke, CNN

  • “Last year, Amazon CEO Andy Jassy said that every business division at the company was experimenting with AI. Today, Amazon has announced its most ambitious AI product yet: a chatbot named Rufus to assist with your online shopping.” – Nadeem Sarwar, Digital Trends

  • “The launch of the AI chatbot comes on the heels of other AI-powered additions across Amazon.com aimed at improving the shopping experience for consumers, ranging from tools that help customers find clothes that fit to those that enhance product reviews with summaries of product highlights and customer sentiment, as well as others aimed at advertisers and sellers.” – Sarah Perez*, TechCrunch*

  • “Consumers are more than twice as likely to search first on Amazon versus other search engines when they are looking for a specific product to buy. But the e-commerce giant has long wanted to attract customers when they are still brainstorming and researching their options, when they typically turn to other sources, from TikTok to Google. Rufus is an attempt to bring customers into Amazon before they know precisely what they want.” – Karen Weise, The New York Times

Send Barun Kwak a reply about this page
More from Barun Kwak