Home / Use cases / AI shopping assistant and AI search
ECOMMERCE & RETAIL
STARTUP
GENERATIVE AI
AI SEARCH
AI shopping assistant and AI search for D2C brands
Plain-language search, a grounded AI advisor and an AI-first care inbox. An AI shopping assistant and AI search grounded in the catalog, plus a support copilot.
Example engagement: figures show typical results for this kind of project.
3
AI tools: search, advisor, support copilot
10 wks to launch
typical timeline
8
services working together
3
user roles, each with its own workspace
BUILT FOR
D2C and eCommerce brands
INDUSTRY
eCommerce & Retail
STAGE
Series A, 38 people
WHERE WE DELIVER
United States, Canada and Europe
TYPICAL TIMELINE
10 weeks to launch
TEAM
Cross-practice: AI and data, security, software, managed cloud
Services used:
Generative AI & LLM apps
AI agents & automation
Machine learning
Data engineering
AI governance
Web applications
API & integration
Cloud cost optimization
IN SHORT
How can a D2C brand use AI to sell more without risky advice?
By grounding an AI shopping assistant in its own product and ingredient data, testing answers every week and routing anything medical to people.
KEY TAKEAWAYS
→ Ground AI answers in your own catalogue and rules, not the open web
→ Fix zero-result searches first: they are lost revenue
→ Route medical or sensitive questions to people, every time
→ Track answer accuracy weekly on a fixed evaluation set
THE CHALLENGE
Sound familiar?
Strong products and loyal customers, but the website and support team could not keep up with growth.
01
Shoppers could not find the right product
Many searches return nothing, because customers search in plain language like "something for dry patches".
02
Support could not keep up
Routine order and returns questions wait hours for a reply, and support hiring outpaces revenue.
03
AI advice in skincare is risky
Wrong ingredient advice or a medical claim could harm customers and break advertising rules.
TECHNICAL ARCHITECTURE
How the AI fits into the store
Every AI answer is grounded in the brand's own catalogue and ingredient database, checked by guardrails, and measured against a fixed test set.
USERS & CHANNELS
Web store
Headless storefront
Mobile app & web
iOS, Android, mobile web
Care inbox
Web
Merchandising console
Web
↓
EDGE & ACCESS
CDN
API gateway
Customer accounts
↓
COMMERCE
Commerce platform
Storefront API
Cart & checkout
Orders & returns
Subscriptions
↓
AI LAYER
AI skin advisor
RAG over catalogue & ingredients
Hybrid search
Semantic + keyword
Care agent
Order & returns tools
Guardrails
Medical & claims filters
Evaluation set
A set of real customer conversations
↓
DATA
Product & ingredient database
Vector index
Customer events
Analytics warehouse
↓
INTEGRATIONS
Helpdesk
Email & SMS platform
Reviews platform
Shipping carriers
SAFETY & GOVERNANCE
Across every layer
• Answers grounded in approved data only
• Medical questions routed to people
• Claims checker for regulated wording
• PII redacted from AI logs
• No training on customer conversations
• Weekly accuracy evaluation
• GDPR consent for personalisation
• Spend caps and alerts per model
THE SOLUTION IN DETAIL
One AI layer, three types of users
Shoppers get expert advice, merchandisers see what customers really search for, and the care team handles only what needs a person.
Product screens on this page show the design with sample data.
01
Shoppers
Mobile app & web
6 core features
02
Merchandising team
Merchandising console
6 core modules
03
Customer care
Care inbox
6 core modules
01
MOBILE APP & WEB
Shoppers
Shoppers describe their skin in their own words, get a routine with reasons, and sort out orders and swaps without waiting for an email reply.
CORE FEATURES
✓
AI skin advisor
Routines built from the brand's own products.
✓
Why it fits you
Plain-English reasons on every recommendation.
✓
Natural-language search
"Something for dry patches" finds the right products.
✓
Routine builder
Morning and night routines with usage tips.
✓
Order & returns help
Tracking, swaps and refunds in the chat.
✓
Personal reminders
Refill and routine reminders with consent.
9:41
Skin advisor
Powered by AI · answers checked against our ingredient database
My skin gets oily by noon but my cheeks feel tight. I use retinol at night.
Sounds like combination skin with a stressed barrier. I'd pair a gentle gel cleanser with a ceramide moisturiser, and keep retinol to 3 nights a week.
Barrier Gel Cleanser
$24 · Add
Ceramide Cloud Cream
$38 · Add
Shop
Advisor
Routine
Orders
Me
9:41
Ceramide Cloud Cream
$38 · 50 ml · 4.7 ★ (2,914 reviews)
Why it fits you
✓ Repairs a stressed barrier (ceramides NP, AP, EOP)
✓ Fragrance-free, safe with your retinol
✓ Lightweight, won't add midday shine
Add to bag
Shop
Advisor
Routine
Orders
Me
9:41
Order #LS-48213
Arriving Thursday · tracked
The cream isn't right for me. Can I swap it?
Of course. It's within 30 days, so I've set up a free swap. Want the lighter Gel Cream instead? It suits oily T-zones.
Swap for Gel Cream · free
Confirm
Shop
Advisor
Routine
Orders
Me
AI skin advisor
Why this product fits you
Order help and free swaps
02
MERCHANDISING CONSOLE
Merchandising team
Merchandisers see what customers search for, fix gaps in minutes, and publish AI-drafted product content that passes brand and claims checks.
L
Lumea Skin
MERCHANDISING
AI search
Advisor
Content
Experiments
Catalogue
AI search & discovery
Semantic + keyword hybrid search · 31 days
Search products, queries, conversations…
Search conversion
6.8%
+2.9 pts
Zero-result searches
1.2%
−11.4 pts
Revenue from search
34%
of online sales
Avg. order value
$64
+$9
Top queries
QUERY
INTENT (AI)
SEARCHES
CONVERSION
BEFORE
STATUS
something for dry patches
Barrier repair, dry skin
4,812
8.1%
0 results
Fixed by AI
niacinamide
Ingredient
3,904
7.2%
4.4%
Improved
spf that doesn't leave white cast
Sunscreen, sheer
2,611
9.4%
1.9%
Improved
gift set under 50
Gift, price filter
1,920
11.2%
0 results
Fixed by AI
acne safe moisturiser
Non-comedogenic
1,744
6.6%
3.1%
Improved
AI search performance
L
Lumea Skin
MERCHANDISING
AI search
Advisor
Content
Experiments
Catalogue
AI skin advisor
Conversations, conversion and answer quality
Search products, queries, conversations…
Conversations (30 d)
48,210
Advisor-assisted conversion
14.6%
vs 3.1% site avg
Answer accuracy (evals)
97.2%
Escalated to humans
3.8%
Conversations per week
Top topics
Routine building
38%
Ingredient questions
27%
Order & returns
19%
Shade & SPF
16%
AI advisor analytics
L
Lumea Skin
MERCHANDISING
AI search
Advisor
Content
Experiments
Catalogue
Product content
AI drafts descriptions, FAQs and translations · brand and claims rules applied
Search products, queries, conversations…
Draft · Ceramide Cloud Cream
Claims check passed
A weightless cream with three ceramides and squalane that helps repair a stressed skin barrier. Fragrance-free and gentle enough to use with retinol.
Languages
EN, DE, FR, ES · reviewed
Blocked claims
"cures eczema" removed (medical claim)
SEO title
Ceramide moisturiser for dry, sensitive skin
Throughput
Products updated this month
412
Time per product
6 min, was 45
Human review
100% before publish
Product content generator
CORE MODULES
✓
Search analytics
Queries, intent, conversion and revenue.
✓
Zero-result fixes
Gaps found and fixed with AI-suggested matches.
✓
Synonyms & intent
Plain-language queries mapped to products.
✓
Advisor analytics
Topics, conversion and escalations.
✓
Product content generator
Descriptions and translations with claims checks.
✓
Experiments
A/B tests on ranking and advisor prompts.
03
CARE INBOX
Customer care
The AI resolves routine questions in seconds, and the care team gets the conversations that need empathy or judgment, with full context.
CORE MODULES
✓
AI-first inbox
Most routine tickets resolved without a person.
✓
Smart escalation
Reactions, repeat issues and disputes routed to people.
✓
Returns & swaps
Policy-based automation with fraud checks.
✓
Answer quality review
Hundreds of conversations checked every week.
✓
Tone & macros
On-brand responses in every language you sell in.
✓
Customer history
Orders, routine and past chats in one view.
L
Lumea Skin
CUSTOMER CARE
Inbox
Returns
Quality
Macros
Customer care
AI assistant resolves first, people handle what matters
Search products, queries, conversations…
Tickets resolved by AI
71%
First response
9 s
was 7 h
CSAT
4.7 / 5
+0.5
Cost per contact
$0.62
−78%
Escalated conversations
CUSTOMER
TOPIC
WHY ESCALATED
WAITING
PRIORITY
Ana P.
Reaction after using serum
Possible adverse reaction
2 min
Urgent
Mark T.
Wholesale enquiry
Sales opportunity
14 min
Normal
Sofia L.
Damaged parcel, 3rd time
Repeat issue
6 min
High
James R.
Subscription billing
Payment dispute
21 min
Normal
Care inbox with AI handoff
L
Lumea Skin
CUSTOMER CARE
Inbox
Returns
Quality
Macros
Returns & swaps
Policy-based automation with fraud checks
Search products, queries, conversations…
This month
Returns handled automatically
84%
Swaps instead of refunds
46%
Return fraud flagged
1.7%
Refund within 24 h
96%
Returns & swaps automation
L
Lumea Skin
CUSTOMER CARE
Inbox
Returns
Quality
Macros
AI answer quality
Weekly evaluation of 500 sampled conversations + guardrail logs
Search products, queries, conversations…
Factual accuracy
97.2%
On-brand tone
95.8%
Unsafe advice blocked
37 this week
Medical questions routed
100%
Accuracy trend
AI answer quality review
TECHNOLOGY & SECURITY
A lean AI stack a startup can afford to run
Managed services where possible, with cost and quality tracked from day one.
TECHNOLOGY STACK
Storefront
Headless commerce
Next.js
React Native
AI
LLM API
Embeddings
Hybrid search
Agent tools
Data
PostgreSQL
Vector index
Event tracking
LLMOps
Evaluation set
Prompt versioning
Cost dashboards
Cloud
Vercel
AWS
Integrations
Helpdesk
Email & SMS
Reviews
SECURITY & COMPLIANCE
✓ Answers grounded in approved data, no open-web browsing
✓ Medical and adverse-reaction questions escalated
✓ Claims checker for regulated product wording
✓ GDPR consent for personalisation
✓ PII redacted from AI logs
✓ No training on customer data
✓ Model spend caps and alerts
✓ Prompt-injection tests before every release
DELIVERY TIMELINE
From kickoff to launch in about 10 weeks
Search fixes shipped first, then the advisor, then the care agent, with an A/B test before full rollout.
Month 0
Month 1
Month 2
Month 3
Discovery & data audit
Catalogue, ingredients, search logs, tickets.
Wk 0–2
Hybrid search
Semantic + keyword, synonyms, zero-result fixes.
Wk 1–5
AI skin advisor
Grounded answers, routines, guardrails.
Wk 2–8
Care agent & returns
Order tools, swaps, escalation rules.
Wk 4–9
Evaluation & safety review
Conversation test set, claims checks.
Wk 6–9
Launch & optimisation
A/B test, cost tuning, weekly reviews.
Wk 9–10
Typical plan: about 10 weeks to launch · A/B tested before full rollout
Shoppers find the right product faster and the care team focuses on complex cases. An AI shopping assistant and AI search grounded in the catalog, plus a support copilot.
WHAT THIS USE CASE DELIVERS
Use case blueprint. Scope, timeline and team are confirmed in a written proposal for your project.
RELATED USE CASES
More work like this
ECOMMERCE
PCI DSS 4.0
SOC
PCI DSS 4.0 and unified compliance for multi-market retailers
THE CHALLENGE
Stores, a growing online channel and PCI DSS 4.0 requirements that are now mandatory.
OUR APPROACH
Checkout script control plus one unified control framework across every market.
WHAT SUCCESS LOOKS LIKE
One set of controls and evidence for PCI DSS, ISO 27001, SOC 2 and privacy audits.
5
frameworks mapped to one control set
Read the story →
1 / 3
RELATED INDUSTRIES
Built for your industry
The industry behind this use case, and others where the same approach works.
All industries →
BLOG & RESOURCES
Insights for commerce and retail teams
Practical guides on speed, AI and peak-season readiness.
View all resources →
ECOMMERCE AI
AI product search: lift conversion with semantic search
Hybrid search, merchandising rules and A/B testing.
Akash Soni, Founder & CEO · 8 min read
PEAK SEASON
Black Friday readiness: load tests, code freezes and on-call
A 10-week plan for peak traffic, bots and fraud.
Akash Soni, Founder & CEO · 8 min read
GENERATIVE AI
RAG vs fine-tuning: which one, and what each costs
When retrieval is enough and when to fine-tune.
Akash Soni, Founder & CEO · 8 min read
FAQ
AI shopping assistant questions, answered
Short answers to what teams ask us most about this kind of project.
Still have a question?
Book a free consultation
What is an AI shopping assistant?
An AI shopping assistant is a conversational guide on an online store that answers product questions, recommends items based on needs and preferences and helps customers check out, using the store's catalogue and policies.
What is hybrid search?
Hybrid search combines keyword search with semantic, AI-based search, so shoppers find products whether they type exact names or describe what they need, improving relevance and conversion.
How do you stop an AI assistant giving bad advice?
We ground answers in approved product data and policies, block medical or unsafe claims, test against evaluation sets, add escalation to humans and monitor conversations for quality.
How long does it take to launch an AI assistant for eCommerce?
A production AI shopping assistant typically takes 6 to 10 weeks, including catalogue integration, evaluation, safety testing and a controlled rollout to a share of traffic.
Will AI replace our support team?
No. AI handles common questions instantly, while your team focuses on complex and high-value conversations.
SERVICES IN THIS USE CASE
Generative AI & LLM Application Development →
LLM and RAG apps built for production
AI Agent Development & Workflow Automation →
AI agents that finish real work
Machine Learning & Predictive Analytics →
Forecasts and models your team can trust
Data Engineering & Data Pipelines →
Reliable pipelines for analytics and AI
Want AI that sells, not AI for show? Let's talk.
Book a free 30-minute call. You'll leave with your biggest search and support wins and a realistic launch plan, whether or not we work together.
Your biggest AI wins
A realistic launch plan
No obligation