An AI consultant for the Dim Optiki chain that picks in-stock frames and books try-ons
A widget on dom-optiki.ua talks to customers in plain language, shows models that are actually in the shops, and hands managers ready-to-call leads. The owner sees everything in a dashboard.
Customers write the way they'd talk to an optician in the shop: "I want a light frame under 3,000". The bot clarifies the need, shows specific models that are in stock today and books a try-on in one step. The manager gets a lead with the model, shop and time, and the owner sees in the dashboard what is happening on the site right now.
The problem
Dim Optiki is a chain of optical shops: eyewear selection, optician consultations, ophthalmologist appointments, contact lenses. The website lists hundreds of frames, and it's hard for a visitor to know what suits them and what the nearby shop has. Customers often left just a phone number, and the manager had to find out what they needed during the call.
They needed a consultant that: — talks to customers in plain language, without button questionnaires; — never makes things up: shows only models that are really on sale and never diagnoses; — hands the manager a complete lead, not just a phone number; — gives the owner clear analytics instead of "the bot is doing something".
What the consultant does
Free conversation
No buttons or forms. The first replies clarify the need, then the bot recommends frames, sunglasses or lenses and adds a relevant offer or article.
Only what's in stock
The bot takes candidates from the site's catalogue and checks each model's page live. It shows up to three models marked in stock, with photo, price and stickers.
Try-on booking
"Try on in a shop" on a model card: shop, time, name, phone. The lead lands in the Try-on tab with the model, shop and time.
Knows the product page
When someone is looking at a specific model, the widget names it and offers to check stock or book a try-on right away.
Remembers the visitor
Someone who comes back a week later sees "Welcome back!" and continues where they left off. Only the topic is kept, never contact details.
Safe answers
No diagnoses: symptoms get a recommendation to see a doctor. Stock, prices and links are added by code, not by the model, so they can't be invented.
Leads and analytics
Every lead is saved to the database first and then copied to the managers' Google Sheet, into the tab for its type: AI consultant, try-on, eyewear selection, optician consultation, doctor appointment, availability, callback. Each row has an AI summary of the request, the preferred messenger, the model and product link, the shop, a convenient time, the traffic source and a link to the whole conversation. If the sheet is unavailable, the lead is never lost.
The owner gets a separate password-protected dashboard: who is on the site right now and what they're doing, every conversation with filters, the funnel from opening the widget to a lead, lost leads, traffic sources and a report for any period, ready to forward.
How we worked
June · Telegram qualifier
First version: eyewear selection, consultation, doctor appointment and callback scenarios, with leads passed to managers.
July · Website widget
The chat moved to dom-optiki.ua with a single script tag. Leads go to a Google Sheet with a tab per scenario.
September · AI assistant and catalogue
Our own catalogue and promo crawler, medical safety, brand priority, and a code and security audit followed by fixes.
October · Free AI chat and dashboard
Button-free conversation, models with live stock, try-on booking, streamed replies, a 10 KB launcher and analytics for the owner.
The product grew in iterations alongside real use: about three and a half months from the first bot to the AI consultant. Over 1,700 automated tests guard the logic, and the key decisions are recorded in 12 architecture decision records.
Can it be added to an existing website?+
Yes. The widget is embedded with one script tag, with no changes to the site. Pages load only a 10 KB button; the chat itself loads when the browser is idle.
Can the AI make up stock or prices?+
No. Stock is checked live on the product page, and prices and links come from the catalogue in code. The model only writes the advice text, and replies are checked before they are shown.
What if the shop's site or the sheet is down?+
The conversation continues: model cards disappear for a while, and the lead is still saved in the database and reaches the sheet later.
How is customer data protected?+
Phone numbers and nicknames are masked in analytics, the dashboard is password-protected and links lead only to the shop's site. A code and security audit was done before launch.
Want an AI consultant like this on your website?
Let's discuss your challenges and find the best scaling solution.