Banana Republic GPT

The Problem: How Might We Improve Customer Loyalty by Keeping Purchase Sessions under 3 Minutes for Online Shoppers?

Shoppers want to find and purchase products on their favorite
e-commerce platforms without sifting through 100+ inventory items each time they return. With the goal of spending the expected 2-3 minutes (peasy.nu) or shorter during a new session, their priority is on meeting with the people with whom and the events where they use their bought items.

78% of Consumers demand Personalization

According to McKinsey, consumers are more likely to repurchase after receiving personalized communications (2021, link). Precisely, personalization entreats loyalty, as the report says, “recurring interactions create more data from which brands can design ever-more relevant experiences—creating a flywheel effect that generates strong, long-term customer lifetime value and loyalty”.

Further on personalization, McKinsey revealed that the #1 shopper expectation is as follows (2023, link):

“Give me relevant recommendations
I wouldn’t have thought of myself”

More Experience Gaps

handshake icon
systems icon

Icons source, respectively: Illustraly on Flaticon.com, and Ricons on thenounproject.com

81% want a Seamless Human Handoff Process

When escalation happens, agents receive no context. Customers them need to re-explain from scratch. As a result, handle times spike. Zendesk reports that 81% of consumers want customer experience representatives to pick up where they left off in their conversation with AI (2026, link). This research involved 11,000+ people across 22 countries.

"I'm transferring you to an agent. Please hold." — and then silence.”

No Connection to Commerce Systems

Legacy bots sit outside your Shopify, Zendesk, and inventory stack. They can read FAQs but can't track packages, initiate returns, or check real stock. IBM speaks to this need by stating that “Advanced chatbots require integration with product catalogs, CRM tools and other logistics. If done incorrectly, it can lead to broken experiences or inaccurate responses” (2025, link).

“Your package is being processed at a logistics facility…”

The Solution

A 4-layer AI stack for e-commerce

Retail GPT connects a chat UI to live commerce systems through GPT function calling - so every response is grounded in real data from the backend.

  1. Chat UI - Web/Mobile

  2. GPT (Open AI) - Reasoning layer

  3. Tool/API Layer - Function calling

  4. Commerce backend - Shopify-like

Why Now

Function-Calling in
GPT 4.1

GPT can reliably decide which backend tool to invoke, pass structured arguments, receive results, and synthesize a grounded response — all within a single sub-4-second round trip. This capability didn't exist at this reliability level late 2024/early.

Icons source, respectively: Oksana, thenounproject.com, Rahul Kaklotar, Flaticon.com, and Alzam, thenounproject.com

Commerce APIs are
real-time

Shopify's 2024 API expansion enables real-time reads of orders, inventory, and customer data at scale. The plumbing now exists to build a single integration layer that connects GPT to the full retail backend stack.

Customer Experience takes first place

Post-pandemic data proved that bad support directly kills Lifetime Value. As Zendesk puts it, “When 85% of CX leaders say one unresolved issue is enough to lose a customer, speed, accuracy, and empathy become non-negotiable (2025, link)”. The same data shares that “86% of consumers say responsiveness and accuracy strongly influence purchasing decisions.”

6 capabilities that define the best retail AI

Using Figma Make and ChatGPT (toolset listed below) primarily, I looked into features of the best conversational AI while comparing UI through live prototyping across Enterprise Support, 
Knowledge Base AI, E-commerce Support and
Proactive Chat interfaces. Claude and Copilot were also employed during my research to source the 6 key qualities listed on the right.

Figma Make icon

Figma Make

ChatGPT

Claude

Copilot

Prototype

Before Testing

Before testing my prototype, the only way to access the GPT was by visiting the purchase history page where the chat with us button was. However, on testing, the feedback from more than 50% of respondents wanted to discover the feature on landing on the home page of the site sot they could expedite their shopping immediately. This led to a second iteration which applied those changes.

After Testing

Following testing with 4-5 users, a key iteration I made was to improve the discoverability of the GPT functionality. This was done by adding 2 elements: a For You button in the menu navigation which is accessible site-wide, and secondly, an in-line strip of recommendations on the My Account page and Purchase History section as shown below.

Clickable prototype

Future Plans

An additional piece of feedback from testers was to consider the possibility of applying this concept to multiple retail stores to scale the experience. So, with more time I want to explore this opportunity, where users can source desired styles across retailers through conversational GPT searches.