Customer-Facing AI Chat — Smashley's Portfolio

🎯

Context & Challenge

This project began without Product or UX involvement. By the time our team was brought in, leadership had already approved adding an internal AI tool—originally built for store associates—into the customer-facing Publix app.

Our “kickoff” was a two-minute screen recording of the associate tool, with no access to a working build.

What initially seemed like a straightforward adaptation quickly revealed a larger issue: the tool wasn’t designed for customers. It lacked a clear purpose in the shopping journey and introduced meaningful usability and trust concerns.

As a result, our role shifted from execution to definition—clarifying the product’s purpose, identifying critical gaps, and establishing a direction aligned with customer needs and industry expectations.

💡

Research

Given limited access to the product, we grounded our recommendations in POC analysis + competitive research across leading retail apps (Amazon, Walmart, Lowe’s, Home Depot).

Key Findings

➊

No clear customer value or defined use case The tool inherited capabilities from an internal associate experience, but key features (e.g., stock visibility, pricing) were not permitted in the customer app. This left the feature without a clear role—until we aligned on a focused goal:
helping customers navigate weekly savings through natural language.

➋

Experience designed for associates, not customers The tool relied on internal concepts like planograms and inventory terminology, which are unfamiliar and irrelevant to customers. This created a disconnect between how the system worked and how customers shop.

➌

UX patterns introduced friction and reduced usability Results were displayed as numbered text lists, with limited product card support and missing quick actions (e.g., Add to Cart, Add to List). This forced unnecessary navigation and made the experience slower than traditional browse or search.

➍

Missing context created trust and accuracy risks The AI did not clarify which store data it referenced, despite variation in availability and promotions. Additionally, incomplete product data (e.g., only ~24% with nutrition/allergen info) increased the risk of incorrect or misleading responses.

These findings highlighted that the challenge was not just interface design—it was defining a clear, trustworthy, and valuable role for AI within the customer experience.

✨

Solution

Still under construction…

Business GoalIncrease conversion, reduce time and abandonment

Customer GoalSimpler, faster checkout

🔑

Key Design Decisions

←

1 of 8

→

🚀

Projected Impact

Still under construction…

💭

Reflections

This project reinforced the importance of involving Product and UX early—especially when introducing emerging technologies like AI.

By the time we engaged, many foundational decisions had already been made, requiring us to shift from execution to reframing the problem and redefining the product direction.

It also highlighted that AI features must be grounded in clear user value, accurate data, and trust—not just technical capability.

What I’d Do Differently

Involve Product & UX earlier to define clear use cases before development
Validate AI capabilities against real customer needs before adapting internal tools
Establish guardrails for accuracy, context, and trust from the start