Allie from Skaimart

By 2030, AI is projected to handle 80% of all customer interactions. -sellerscommerce.com

E‑commerce is rapidly embracing AI assistants, yet current shopping bots often suffer from poor user experience and limited user control. I led the design of “Allie”, an AI shopping assistant for the fictional Skaimart app, to tackle these gaps.

We introduced

• Customizable AI profile for personalized recommendations.
• Vision to extent this profile to the greater Conversational Commerce ecosystem.
• Enhanced chat interactions (multimodal voice/text, smarter product comparisons, seamless integration)

A/B test against Amazon’s own shopping assistant proved Allie’s impact:

• 18 of 20 of participants preferred Allie’s experience.
• 100% said the new AI Profile was the feature they most wanted in real shopping apps.

With retail giants like Amazon, Walmart, and others rolling out AI chatbots in their apps, I set out to investigate how effective and user-friendly these chat experiences really are.

Are these bots truly helpful “virtual shopping assistants,” or just gimmicks?

Do they mimic the feel of an in-store sales associate, or do they frustrate users?

Could it narrow choices or steer shoppers in ways that undermine their control?

Our challenge was clear:

How might we design an e-commerce chatbot that augments the shopping experience without replacing it, providing guidance while preserving the user’s autonomy?

This project was executed under intense constraints. I led the process end-to-end (UX research, interaction design, and testing) with one collaborator supporting the study throughout.

We completed the entire case study in just a week, from research through hi-fi prototype to testing.

A fast-paced sprint that demanded sharp focus on the most impactful problems and solutions.

Evaluating Current Chatbots

We first conducted a structured evaluation of four leading retail apps’ AI chatbots (Amazon’s “Rufus,” Walmart’s “Ask Sparky,” Safeway’s in-app assistant, and Home Depot’s “Magic Apron”).

We devised a sandardized test process with 6 key tasks for each chatbot, emulating common customer queries:

• Search by Asking
• Product Q&A
• Compare Products
• Store Navigation
• Policy/Customer Service
• Recommendations & Justification

And, we uncovered a wide range of maturity from Amazon's highly sophisticated to Safeway's beta model.

Critically, common pain points emerged:

• Limited memory of context
• No way to refine what the bot shows you
• Awkward or verbose answers for comparisons
• Bots appear in only parts of the app rather than throughout the journey

These findings confirmed that today’s chatbots often deliver a fragmented experience: they assist with Q&A, but they don’t truly adapt to user needs or integrate seamlessly with browsing.

Users have to cede control to the bot, rather than collaborating with it.

We went through 30+ published research and articles to understand, validate and better ideate solutions.

Thematic findings

• Adoption is real, not hype: broad rollout + high expectations for personalization.
• Company bet: purpose-built, vertically integrated assistants (Rufus, Magic Apron) to reduce effort and increase confidence.
• Trust grows when the bot feels human-attuned and interactive – not just accurate.
• But autonomy has a ceiling: users still want the wheel; over-automation risks narrowing choices.

Net-net:

Assistants must be helpful and steerable, not black boxes.

Problem statement (refined)

Current e-commerce chatbots need to evolve into truly assistant-like experiences, ones that integrate into the entire shopping journey and offer hyper-personalized help, while empowering the user to steer recommendations and decisions.

What guides Allie

• Visible agency controls – so users can steer the bot mid-conversation.
• Human-attuned interaction: fast, concise answers + follow-ups; multimodal in/out; comparative cards over walls of text.
• Transparency against constraint: explain why each rec fits the current profile and show alternatives.

Building on our directive of hyper-personalization, we extended the idea beyond a single chat into the wider conversational commerce ecosystem. This led us to define the concept of a shareable AI Profile that carries user preferences across sessions, products, and even apps.

The AI Profile

A persistent, customizable profile that users can edit mid-chat and even export in the future to other shopping apps. It gives shoppers direct control over how the assistant behaves:

• Focus: Deals, Fast Delivery, Sustainable, Budget Alternatives, etc. (multi-select)
• Price band: Budget · Value · Neutral · Premium · Luxury
• Quality: Basic ↔ High-End
• Brand familiarity: Explore · Varied · Neutral · Familiar Only

A quick access panel shows current preferences and allows one-off overrides (e.g., “show premium just this once”).

MVP and Enhancements

For our MVP, we prioritized creating an interface that is human-attuned and steerable. The AI Profile anchors the experience, while design enhancements focus on:

• Fast, concise answers and natural multimodal input (text/voice)
• Clear follow-ups instead of walls of text
• Comparative cards over narrative blocks

Future enhancements explore a portable profile, allowing shoppers to carry preferences across multiple shopping platforms.

Additional Features

• Better comparisons: Side-by-side product cards inside chat with specs, pros/cons, price, and ratings. Scroll to scan; tap to dive deeper.
• “Ask Allie” everywhere: An icon on listings and product pages opens a context-aware chat that already knows the item you’re viewing.

In short: from a directive of hyper-personalization, we shaped Allie into an assistant that is integrated, steerable, and scalable across the shopping journey.

On a personal note, this project was an intensive learning experience. Leading a rapid one-week case study taught me how to quickly synthesize research into actionable design decisions and confirm those decisions with just-enough user feedback. I also deepened my understanding of the balance between AI automation and human-centered design, and concepts like maintaining user agency, building trust through transparency, and the ethics of personalization were not just theoretical, but things I had to address in the design.

Finally, this project was a breakthrough for me in mobile UX and emerging tech: it pushed me to design beyond web interfaces, adapt to mobile constraints (like small screen comparisons), and explore the cutting edge of AI-driven UX. I’m excited to carry these lessons forward as AI becomes an ever bigger part of product design.