Magic Apron AI Q&A Bot
is an AI-powered Q&A tool that delivers associate-like product guidance online, helping customers quickly and easily find accurate, relevant information and make confident purchase decisions.
Magic Apron AI Q&A Bot
is an AI-powered Q&A tool that delivers associate-like product guidance online, helping customers quickly and easily find accurate, relevant information and make confident purchase decisions.
Magic Apron AI Q&A Bot
is an AI-powered Q&A tool that delivers associate-like product guidance online, helping customers quickly and easily find accurate, relevant information and make confident purchase decisions.
Outcome
Magic Apron AI Q&A Bot drove a clear positive shift in customer behaviour.
Users who engaged with the feature were significantly more likely to explore, add to cart, and convert, validating it as a high-impact decision-support tool.
Magic Apron AI Q&A Bot drove a clear positive shift in customer behaviour.
Users who engaged with the feature were significantly more likely to explore, add to cart, and convert, validating it as a high-impact decision-support tool.
86000+
86000+
Questions asked
(Nov 6 – Dec 1)
Questions asked
(Nov 6 – Dec 1)
207%
207%
Revenue per visit
($12.40 lift)
Revenue per visit ($12.40 lift)
Revenue per visit
($12.40 lift)
177 bps
177 bps
Conversion rate
Conversion rate
808 bps
808 bps
Add-to-Cart rate
Add-to-Cart rate
Overview
Home Depot customers rely heavily on Product Page to make purchase decisions, yet key details such as specifications, installation info., warranties, availability, and product nuances were buried deep on the page or split across multiple accordion sections.
The AI Q&A Bot introduces a generative-AI layer that lets shoppers ask any question about the product and receive instant, accurate, natural-language answers grounded in verified Product Page data sources.
Home Depot customers rely heavily on Product Page to make purchase decisions, yet key details such as specifications, installation info., warranties, availability, and product nuances were buried deep on the page or split across multiple accordion sections.
The AI Q&A Bot introduces a generative-AI layer that lets shoppers ask any question about the product and receive instant, accurate, natural-language answers grounded in verified Product Page data sources.
Customers frequently asked questions that were already answered somewhere on a Product Page, but not easily discoverable.
Customers frequently asked questions that were already answered somewhere on a Product Page, but not easily discoverable.
Goal
Help users get the information they need faster, make confident purchase decisions, and increase Add-to-Cart performance.
Help users get the information they need faster, make confident purchase decisions, and increase Add-to-Cart performance.
The Challenge
How might we make it easier for users to find product information at the exact moment they need clarity - without forcing them to scroll, guess, or leave the page?
How might we make it easier for users to find product information at the exact moment they need clarity - without forcing them to scroll, guess, or leave the page?
User Pain Points
Product details are located far below the fold; customers often don’t find critical info.
On mobile, accordion-heavy layouts hide essential content.
Dense and unstructured information creates high cognitive load (users prefer “quick answers”).
Legacy Q&A sections have low engagement (<2% views).
AI placement in the Home Depot US suffered from extremely low discoverability, resulting in low engagement and no measurable revenue lift.
Product details are located far below the fold; customers often don’t find critical info.
On mobile, accordion-heavy layouts hide essential content.
Dense and unstructured information creates high cognitive load (users prefer “quick answers”).
Legacy Q&A sections have low engagement (<2% views).
AI placement in the Home Depot US suffered from extremely low discoverability, resulting in low engagement and no measurable revenue lift.
Business Problems
Customer support receives large volumes of repetitive product questions that Product Information Page already answers.
Poor discoverability leads to under-utilization of features that otherwise perform well.
Users who leave Product Page to “Google search” often don’t return.
Customer support receives large volumes of repetitive product questions that Product Information Page already answers.
Poor discoverability leads to under-utilization of features that otherwise perform well.
Users who leave Product Page to “Google search” often don’t return.
My Role
As a Senior Product Designer, I led the full UX workstream:
As a Senior Product Designer, I led the full UX workstream:
Defined experience strategy, requirements with PM and engineering feasibility
Designed conversational UI component (new to ACL), interaction patterns, and fallback logic
Partnered with Data Science on prompt engineering & guardrails
Led moderated usability testing (Canada) and synthesis
Worked with Legal on safety, disclaimers, and pronoun strategy
Defined experience strategy, requirements with PM and engineering feasibility
Designed conversational UI component (new to ACL), interaction patterns, and fallback logic
Partnered with Data Science on prompt engineering & guardrails
Led moderated usability testing (Canada) and synthesis
Worked with Legal on safety, disclaimers, and pronoun strategy
This project required deep collaboration across UXR, Product, Data Science, Engineering, UXW, Legal, and Design Ops teams, aligning with Home Depot’s broader AI strategy.
This project required deep collaboration across UXR, Product, Data Science, Engineering, UXW, Legal, and Design Ops teams, aligning with Home Depot’s broader AI strategy.
Research Insights
I leveraged US concept tests, US A/B experiments, and ran a newly conducted Canadian moderated study. Together, the insights were highly consistent: discoverability being the main barrier to feature adoption, and the "AI" label created immediate expectation for assistance-like capabilities.
I leveraged US concept tests, US A/B experiments, and ran a newly conducted Canadian moderated study. Together, the insights were highly consistent: discoverability being the main barrier to feature adoption, and the "AI" label created immediate expectation for assistance-like capabilities.
Discoverability was the no.1 issue
From US UXR and Canadian UXR:
Users did not expect an AI experience on Product Page.
They scroll directly to images → details → reviews, skipping cross-sell rails and legacy Q&A.
When the bot appeared below those sections, users never saw it.
From US UXR and Canadian UXR:
Users did not expect an AI experience on Product Page.
They scroll directly to images → details → reviews, skipping cross-sell rails and legacy Q&A.
When the bot appeared below those sections, users never saw it.
When users did find it, the experience was loved
Across both countries:
Users described answers as fast, clear, and helpful
They felt it saved them time digging through specs
They trusted factual answers (dimensions, materials, warranty)
They appreciated jargon-free explanations (ex: “What is veneer?”)
Many users explicitly said:
“I didn’t know it was here, but this is really useful.”
Across both countries:
Users described answers as fast, clear, and helpful
They felt it saved them time digging through specs
They trusted factual answers (dimensions, materials, warranty)
They appreciated jargon-free explanations (ex: “What is veneer?”)
Many users explicitly said:
“I didn’t know it was here, but this is really useful.”
Users expected ChatGPT-level capabilities
From Canadian UXR, users have shown these expectations:
Follow-up questions
Context memory
Product comparisons
Broader knowledge beyond Product Page
Personalized recommendations
Missing these capabilities caused dissatisfaction when the bot couldn’t answer beyond SKU data scope.
From Canadian UXR, users have shown these expectations:
Follow-up questions
Context memory
Product comparisons
Broader knowledge beyond Product Page
Personalized recommendations
Missing these capabilities caused dissatisfaction when the bot couldn’t answer beyond SKU data scope.
Trust depended on transparency
Users wanted to know where the information came from
Manufacturer-only answers felt biased (ties back to comparison)
Users validated AI answers by “Googling to double-check”
Trust increased when disclaimers made source limitations explicit
Users wanted to know where the information came from
Manufacturer-only answers felt biased (ties back to comparison)
Users validated AI answers by “Googling to double-check”
Trust increased when disclaimers made source limitations explicit
Technical & Organizational Constraints
Data Limitations
AI can only reference Product Page data sources (specs, manuals, descriptions, etc.)
Cannot access customer reviews or external sources (future state)
Cannot support conversational memory (1 question → 1 answer)
Layout Constraints
Full Product Page overhaul planned separately
AI Q&A Bot must fit into existing architecture
Legal Constraints
Avoid legal liability in safety & subjective scenarios
Must gracefully decline unsupported questions
Data Science / Engineering Constraints
Latency increases with complex instructions
Conflicting data sources must follow prioritization rules
Safety testing required across EN/FR (jailbreak + harmful prompts)
Experience Strategy
Introducing AI into Product Page required more than interface design. It meant aligning customer expectations with data constraints and risk boundaries. This strategy defined where the experience should live, what it can responsibly answer, and the interaction, guardrails, and language needed to deliver fast, grounded answers at scale.
Introducing AI into Product Page required more than interface design. It meant aligning customer expectations with data constraints and risk boundaries. This strategy defined where the experience should live, what it can responsibly answer, and the interaction, guardrails, and language needed to deliver fast, grounded answers at scale.
Solve Discoverability First
Set Honest Expectations
Build a Conversational UI Component
Establish AI Safety & Guardrails
Improve Microcopy & Trust Signals
Accessible, System-Aligned UI
Final Design
Key Features
Higher placement near product details
Higher placement near product details
AI iconography that communicates capability
Clear section header: “Ask Our AI About This Product”
Fast, natural-language answers with readable structure
AI iconography that communicates capability
Clear section header: “Ask Our AI About This Product”
Fast, natural-language answers with readable structure
Precise failure messaging
Inline feedback collection
Precise failure messaging
Inline feedback collection
Prompt Engineering
Magic Apron AI Q&A Bot required continuous tuning to ensure responses were accurate, safe, and consistent across thousands of products. Prompt engineering was not a one-time activity, but an iterative, cross-functional quality process.
Magic Apron AI Q&A Bot required continuous tuning to ensure responses were accurate, safe, and consistent across thousands of products. Prompt engineering was not a one-time activity, but an iterative, cross-functional quality process.
How We Worked
I partnered closely with a Product Manager, Backend Developer, Data Scientists, and UX Writer, meeting bi-weekly to review AI performance and refine behavior.
These sessions focused on improving:
Answer accuracy
Tone and clarity
Safety and compliance
Consistency across product categories
I partnered closely with a Product Manager, Backend Developer, Data Scientists, and UX Writer, meeting bi-weekly to review AI performance and refine behavior.
These sessions focused on improving:
Answer accuracy
Tone and clarity
Safety and compliance
Consistency across product categories
Prompt Engineering Process
Examples
Post-Launch Performance
After a one-week internal pilot, we rolled Magic Apron AI Q&A Bot into a month-long 50/50 A/B test to validate real customer impact. The results showed that when customers discovered and used the bot, it positively influenced the shopping journey: increasing confidence, reducing friction in finding product details, and supporting more decisive actions on product pages.
After a one-week internal pilot, we rolled Magic Apron AI Q&A Bot into a month-long 50/50 A/B test to validate real customer impact. The results showed that when customers discovered and used the bot, it positively influenced the shopping journey: increasing confidence, reducing friction in finding product details, and supporting more decisive actions on product pages.
Impact Summary
1. Customers who engage with AI convert better
Massive increases in confidence & ATC behavior.
2. AI reduces cognitive load
Customers said it saved time and helped clarify what they were already looking for.
3. AI improved product understanding
Particularly around materials, installation, and dimension questions.
4. Major organizational milestone
Magic Apron Q&A became:
One of the first customer-facing AI deployments
A foundation for future AI-enhanced shopping experiences (review summaries, comparison assistants, personal shopping assistants)
1. Customers who engage with AI convert better
Massive increases in confidence & ATC behaviour.
2. AI reduces cognitive load
Customers said it saved time and helped clarify what they were already looking for.
3. AI improved product understanding
Particularly around materials, installation, and dimension questions.
4. Major organizational milestone
Magic Apron Q&A became:
One of the first customer-facing AI deployments
A foundation for future AI-enhanced shopping experiences (review summaries, comparison assistants, personal shopping assistants)