How a Mattress Store Handed Off Its Nighttime Sales to AI — and Didn't Lose a Single Customer at 2:47 AM
EXPERT-MATRAS connected AI chat and smart search. In six months, AI handles 8 out of 10 inquiries on its own, processed 24,000+ searches, and collected dozens of leads.
A Ukrainian online store for mattresses and sleep products.
Over 15,000 modelsDozens of factoriesCustom sizesShips across Ukraine
The catalog is vast: spring and spring-free mattresses, sofa toppers, children's and teen models, premium orthopedic mattresses, pillows and accessories. And within each category — dozens of specs: height, firmness, fill, weight limits, warranty. For shoppers, it's a joy of choice. For the store, it's a daily burden on consultants.
The Challenge
A mattress isn't an impulse buy. Before ordering, a customer wants to understand: what firmness for my weight, what size, what fill, can I get a custom size, and does it fit my budget? They don't know the terminology — they ask for "something not too firm, for a 7-year-old, under 6,000."
For a store with a catalog this size, this creates three constant pain points:
10–20 Minutes per Consultation
A manager manually filters thousands of SKUs for each request — dozens of times a day.
Night and Weekend Inquiries Go Unanswered
When managers are offline, customers don't wait until morning — they go to a competitor.
Complex Bilingual Conversations
Customers compare 5–7 SKUs, ask about foam density in kg/m³, weight limits, delivery — half in Russian, half in Ukrainian.
The Site's Search "Doesn't Understand" the Customer
People type model names phonetically and with typos — and a standard search returns "nothing found."
The Solution: AI Configured to Fit the Store
Looma connected as an AI sales assistant in chat and as smart search. But the key: the EXPERT-MATRAS team configured everything for their specific needs on their own, without developers. Here's exactly what they did.
What the store configured
AI Behavior Rules — in Plain Language
The store set the rules in plain text, in any language — and the AI follows them in every conversation:
"Always respond in Ukrainian, even if the customer writes in Russian," "For back pain — recommend memory foam," "Stock availability — only an operator handles that." Rules are written in plain language. Guide →
Product Selection Rules
"When a customer asks for a bed — don't show frames," "No available size — suggest the closest or connect an operator." The store controls the selection logic itself.
Knowledge Base: Delivery, Warranty, Store Locations
For the AI to respond as well as an experienced manager, the store built out a knowledge base — and added a live source from Google Sheets that updates daily.
Knowledge base topics: delivery and payment, store addresses, furniture care, warranty service. Guide →A Google Sheet with delivery terms is connected as a source — Looma checks it daily and always answers with up-to-date information. Guide →
Precise Category Mapping
The store mapped dozens of attribute categories — load from 80 to 200 kg, memory-foam effect, latex, removable covers, winter-summer — each linked with a toggle to a real catalog page.
And All of This — Across 3 Channels at Once
One AI serves the website, Telegram, Facebook Messenger, and Instagram Direct — customers write wherever they prefer, the response is the same. Guide →
In effect, the store assembled an AI sales rep in a matter of hours — one that knows the entire catalog, follows company rules, and works around the clock.
Seeing it in action
The AI Consultant in Action
These are real chat conversations (genuine dialogues, personal data redacted). Notice the range of requests — and how the AI guides each one to a result.
Safely recommends by weight — where an inexperienced salesperson would get it wrong
I weigh 110 kg. Which one is better?
LATEX is the better choice — rated up to 200 kg. ВТІХА is limited to 120 kg; exceeding that is not recommended.
Answers niche technical questions — and bilingually
What's the density of HiDens Foam in kg/m³ and why is the warranty only 18 months?
HiDens Foam® — approximately 30–35 kg/m³. 18-month warranty — it's a budget series, 15 cm height, designed for lighter loads.
Handles a complex custom-size selection for a child
Looking for a mattress for a 7-year-old, spring-free, size 900×1700, HR foam 30–35 kg/m³
Faino ZORY — HardFoam (HR 33–35 kg/m³ equivalent), medium-firm. The manufacturer makes custom sizes over 1.5 m² — so 90×170 cm can be produced.
Knows about military discounts and drives to a lead
Are there discounts for veterans (UBD)?
Yes — a personal discount on delivery or 5% off the order, and in some cases an extra +3%. Leave your number — an operator will handle the paperwork.
Calculates the final total and hands off to the customer's preferred messenger
Sofa topper 120×190, 10–12 cm thick. Final price including delivery to Chernivtsi?
MaNi ТОППЕР-PRO — 5,366 ₴, 10 cm height. With door-to-door delivery (−30% off Nova Poshta rates) ≈ 5,890 ₴.
Viber is easier for me
Passing your number to an operator — they'll send the description and photos via Viber.
Honest when a product won't fit — no hard sell
Baby cot 58×110, very narrow edges narrowing to 45 cm. Will a flat topper fit?
Unfortunately, a flat topper won't work here, and we don't sell foam separately. Better not to order than to get something that won't fit.
Explains delivery for a specific city
How much does delivery of a mattress to Kyiv cost?
In Kyiv: mattresses from 10 cm — from 750 ₴, under 10 cm — from 300 ₴. −30% off Nova Poshta rates applies.
Sales, not just answers
Leads Collected Right in Chat
Looma doesn't just consult — it drives conversations to a lead: picks the product, confirms size and delivery, collects contact info, and hands the operator a ready-to-close order. Here's what a real lead feed from the chat looks like (contacts redacted):
Chat Leads · EXPERT-MATRAS new
Order for AERO 3D mattress, size 200×150 cm
Collected by AI · contact redacted · passed to operator
Order for a custom-size mattress (non-standard dimensions)
Collected by AI · contact redacted · passed to operator
Order for LATEX / ЛАТЕКС mattress-topper, size 110×200 cm
Collected by AI · contact redacted · passed to operator
Order for REST MEDIUM pillow
Collected by AI · contact redacted · passed to operator
Customer requests a callback for consultation
Collected by AI · contact redacted · passed to operator
The operator picks up not a cold lead, but a ready order with the product already selected — all that's left is to confirm and ship.
More than chat
Smart Search That Understands Typos
Customers rarely type model names perfectly. They enter models phonetically, in Latin instead of Cyrillic, with truncated words and typos. A standard search returns "nothing found" — and the store loses a sale for no reason.
Looma's smart search recognizes intent and finds the product anyway. Here are real search queries from EXPERT-MATRAS customers — all misspelled, all with results:
ап кокос→135 productsmeant "UP COCOS"
фридом покет→388 products"Freedom Pocket" typed in Cyrillic
Smart search shows relevant products with category suggestions — and from there, customers jump to the AI assistant with one click. Guide →
Over six months, smart search handled over 24,000 queries — and even the misspelled ones led customers to products, not dead ends.
Brings customers back
Remarketing email your customers can reply to
A customer browsed mattresses and left without buying? Looma automatically emails them the products they viewed — a gentle nudge back to the cart.
But here's the key — they can reply to that email. The customer writes from their inbox, and the operator sees the message right in the regular chat and answers from there. One conversation: the customer in email, the manager in chat, and nobody switches between systems.
Email screenshot: an automatic reminder with recently viewed products — the customer can reply to it straight from their inbox. Guide →
The foundation of accurate answers
AI Magic Data — attributes for every product
The store ran AI Magic across its entire catalog. For each product, AI extracted structured attributes: height, firmness of each side, maximum load, mattress type, materials, whether it's double-sided.
Dashboard screenshot: AI Magic Data for a product — height, side firmness, load up to 140 kg, materials (coconut coir, thermo-felt), type, double-sided. Guide →
That's exactly why search and chat answer by specifications and synonyms, not just by name. A customer asks for "spring-free, up to 140 kg, medium-firm, with coconut" — and Looma instantly finds it, because that data lives on every product card.
…and it even writes proactive prompts itself
AI Magic Data doesn't stop at attributes. On top of them it generates short proactive messages for each product — little hooks tailored to that exact model. When a customer opens the page, the widget gently surfaces one of them — and it creates a wow effect: the store seems to know just what to ask.
Proactive prompts AI Magic Data generated for the FORTE LITE topper
This is the FORTE LITE topper with a double-sided Winter-Summer effect! Want to hear more?
Looking for a thin, roll-up mattress for your sofa? We'll help you choose!
The FORTE LITE topper has corner straps! Curious?
Here's how the proactive prompt looks on the site: the customer opened the FORTE LITE topper — and the widget instantly offers help with that specific model.
Experience right on the product
A widget that knows which product the customer is on
Looma doesn't only work in an open chat. It's embedded into every product page and knows exactly which model the customer is viewing right now — so the conversation starts not from scratch, but already "next to" the right mattress.
An "ask about this product" button, right at hand
When a customer opens the widget on a product page, Looma immediately shows a button for that product: one tap to ask about it or place an order, without explaining what they mean.
The customer opened the Sono® NOTTE mattress — and the widget instantly shows an "ask about this product" button. The context is already picked up.
The AI remembers what the customer is viewing right now
Best of all — there's no need to name the model. The customer can simply ask "what's the height?" and Looma understands they mean the product they're viewing at that moment. The AI keeps the browsing history in mind — and answers in context instead of asking again.
On a product page the customer asks without a name — the AI knows what they mean
What's the height?
The Sono® NOTTE mattress you're viewing right now is 20 cm tall, medium-firm, and holds up to 140 kg.
The "Products" tab: viewed or found
The widget has a separate "Products" tab with a dual purpose. If the customer hasn't searched anything through chat yet — it shows the products they've viewed. And the moment they make a query — the tab switches to the models found for that query.
The "Products" tab before any search: the customer sees the models they've already viewed — easy to come back and compare.A product card right inside the widget: size, discounted price, "Order", "Open on site" and "Similar products" — without leaving the chat.
Launching the AI straight from the catalog
To avoid waiting for the customer to open the chat themselves, the store embedded its own AI-launch button into its pages. Just add the class looma-inline-button to an element — and Looma turns it into an entry point. EXPERT-MATRAS placed such a button in mattress categories and on the product cards themselves, so AI selection starts with a single click right from there.
The "Mattress selection with our AI expert" button is embedded straight into the page — the class looma-inline-button launches the AI assistant from anywhere.
Beyond sales
AI that analyzes the business, not just answers
Thousands of dialogues and searches aren't just messages — they're data. Every period, Looma turns them into a detailed, multi-page report that reads like a sales-department audit. Here's what it breaks down:
Demand structure — what customers search for most, by category
Smart-search stats — top queries and queries with no results
Prioritized recommendations — what to change first for revenue
PDF report: demand structure — sofa toppers, orthopedic mattresses by weight and back pain, kids' models, pillows — with links to specific dialogues. Guide →A breakdown of an operator's work: activity, tone, strengths and weaknesses, leads collected — based on real conversations.
The store ends up seeing not a "black box" but a clear picture: what to sell, where money leaks, which operators close deals, and exactly what to fix first. This is the kind of analytics you'd normally hire a consultant for — here it's generated automatically from real conversations.
Results from the First ~6 Months
1,800+
live chat conversations with shoppers
8 out of 10
inquiries resolved by AI alone, without an operator
24,000+
search queries processed — even misspelled ones
dozens
leads collected directly in chat and passed to operators
24/7
salesperson-level consultations — nights included
2 languages
customer writes in Russian — Looma responds in Ukrainian
Why It Works
Mattresses are a complex category with thousands of SKUs and a mountain of parameters. Looma turns that complexity into a conversation: the customer asks like they would a person, gets a precise match, comparison, and honest advice. And if they don't know the exact product name — smart search still gets them there.
Meanwhile, the store stays in full control: behavior rules, selection logic, knowledge base, categories, channels. Not a black box — a tool that a sales manager can configure for their catalog in an hour, no code or developers needed.
Want the same for your store?
Looma installs in minutes and starts selling for you 24/7.
