Context.

The Amazon Visa, a co-branded credit card with Chase, is Amazon's flagship payment product. Until 2023, customers could only manage their cards through Chase, resulting in the following pain-points:

1. Rising late fee incidence leading to attrition: After approval, the Amazon Visa becomes customers’ default payment method. Those without a Chase relationship (66% of new applicants in 2020 ~1.2 million) must wait 7-10 days for the physical card to setup their card account during which time they may accrue a balance. This delay in account setup led to one in five new customers paying late fees, and 6% closed their accounts within six months—three times the Chase benchmark.
2. Low customer ratings: 50% of 1-2 star reviews for the Amazon Visa cited poor account management experience. Notably, 90% of potential and 72% of existing customers wanted to manage their card directly on Amazon. 

As part of Amazon's S-team goal, the card Product leadership to launch 6 projects to bring card account management within Amazon. I led the UX efforts for 3: Pay Bill, View Statements, and View Transactions. Before starting, I collaborated with an external research agency to identify customers' key needs and the top seven Jobs-To-Be-Done in context of credit card account management. I used these insights to shape my UX strategy and aligned with the Product and Engineering teams on customer goals.

Background.

"I was disappointed when I got my Xbox. I realized I needed the HDMI cord to make it HD and I wasn't aware of that when I bought it."
Research participant, Basket Building customer study

Buying one product is rarely enough to actually use it. A TV needs a mount, a cable, maybe a soundbar. Amazon's Basket Building org exists to close that gap, so customers aren't discovering the missing piece after the box is open. It runs two programs: bundles (a set of products a seller packages and prices together, vouching that everything works together) and complementary recommendations (individual add-ons the customer judges for themselves).

Today, every bundle and comp rec offer either appears on a product page or is initiated from one. But that's not where customers are committing. Our clickstream analysis showed a large share of purchases get added to cart directly from Search, where neither program exists. No bundle, no comp recs. Those customers are never offered the items that complete their setup, so the opportunity is missed entirely.

In 2026, I took on this project through an experimental role called the AI Builder. The idea: can a pod of three (PM, designer, engineer) use AI to take a feature from analysis all the way to production? That's work that normally crosses several specialist teams with a handoff at every step. I owned the UX and the front-end code.

Project goals.

The goal was to bring basket building upstream to Search, meeting customers at the moment they commit instead of hoping they visit a page they've already skipped. Our team's success is measured by how complete each customer's basket is (units per purchase), and purchases from Search were the ones most often missing the items that complete the setup. I proposed two experiment categories: surfacing basket building after add-to-cart (finish the setup) and before add-to-cart (start with the setup).

Project was initiated in May 2026. The scope entailed sizing the opportunity with clickstream analysis, synthesizing 11 Basket Building UX studies into a design framework, and designing plus building the front-end for:

• Bundle Swap: a seller-packaged bundle offered the instant a customer adds to cart, one tap to switch to the complete set (3 CX treatments)
• Build Your Own Bundle: the customer assembles their own set of complements in that same moment (2 treatments)

Team.

AI Builder pod: Product Manager, Software Engineer, UX Designer (me).

I owned the UX and the front-end build. That covered interaction design flows, high-fidelity mock ups, the interactive prototype, organizing leadership reviews, accessibility annotations, and writing the production front-end code with AI agents, submitted through Amazon's engineering review process.

I also absorbed a chunk of the product analysis. The pod's PM had limited bandwidth, so sizing the opportunity became my job: I used AI agents to trace 1.6 million purchase journeys through Amazon's clickstream data and built the case for investing in Search. In practice I was wearing the analyst hat, the designer hat, and the front-end engineer hat, with the PM and SDE as partners on hypothesis and backend.

Design & outcomes.

Our clickstream analysis showed that 4 in 10 consumer electronics purchases happen without the customer ever seeing a bundle, and about half of those happen entirely on Search. This validated our decision to bring basket building to the search page first.

Using insights from 11 prior Basket Building research studies, I designed 3 Bundle Swap treatments and 2 Build Your Own Bundle concepts. Seller-provided bundle names were all over the place ("TV + S95TR SB Bundle"), so I also proposed a new naming framework that leads with what's new: "with [Brand] [Product] [Key Feature]". In Bundle Swap the customer just added the first item to cart, so the card shows only what they don't already know.

I then used AI agents to write the front-end code and submitted my first production code review within the first week. When we learned a designer's verification isn't enough for production code, we set up a two-step review where I verify the customer experience and my engineering partner verifies the code. That process became part of the playbook my org adopted as the pilot's primary deliverable. We launched Bundle Swap first with the three treatments, with Build Your Own Bundle planned for December 2026. 

11%

Bundle swap rate

+0.11%

UPPu on treated purchases

Bundle swap treatments.

Build your bundle (December 2026).

📝 Note: This page shows only parts of the project. In-depth/sensitive details and end-to-end UX can be covered in person or during the interview process.