Systems · AI inside the business
AI works wherethere is a system.
A language model answers customers at night, reads work chats and writes reports. But it only becomes accurate when the company has its own data and rules: a CRM, booking, a rota, money accounting. We build both.
Order in the data first. Then AI.
It rarely starts with technology. It starts with something everyone is fed up with: night-time enquiries waiting until morning, a manager matching payments by hand, a rota living in a chat, payroll taking half a day.
We find where money and time leak, build a system that knows the company’s rules, and only then put AI where it saves people hours. Below are systems already running for our clients: escape rooms, a horror cinema and a theatre in Omsk, an agency and a business consultant.
AI inside the product
The model understands people and writes the text. Numbers, prices and decisions stay with the system and a person.
AI in sales · three brands, six channels
AI sales operator
At night and in the bots it answers customers of three brands with the exact price and free times. Managers get a digest in the morning.
AI for managers · work chats
AI work-chat assistant
Reads an agency’s 10 work groups, transcribes voice notes and three times a day reports what’s on fire and which client is waiting for a reply.
Client project · AI business diagnostics
AI business diagnostics for a consultant
A quiz and an interview become a review: a web page, PDF and slide deck. The code does the maths, a language model writes the text, and every number it uses is checked against the client’s answers.
The systems it stands on
The company’s own data and rules. Without them AI guesses; with them it answers exactly.
Sales · own CRM instead of amoCRM
Our own CRM instead of amoCRM
4,148 deals and the full history moved with zero discrepancies. Deals, clients, shifts and the till share one database.
Operations · shifts, till, staffing
Staff portal
Shifts, cover and the till for two branches on staff phones. The manager sees the whole week and gets only deviations.
Booking and payments · own system
Horror Cinema: own booking with online payment
Time and prepayment first, the film later. Twice as many visitors start a booking; more than half of deals need no manager.
Booking · one core for three venues
Party booking and theatre booking on one core
A kids’ party splits across five venue resources, and of 12 simultaneous requests for one time, one gets through. The theatre moved from YClients onto the same core in one day.
Bots · bonuses · pricing
Bonuses instead of promo codes, and hot sessions in the bots
A personal bonus wallet, discounts only on empty time and one price on the website, Telegram and VK. 10 of 11 discounted bookings landed on time that usually sits empty.
Finance · cash, payroll, receipts
Cash Flow and payroll
A daily answer to how much money can be spent. Payroll from events in 10–20 minutes instead of 4 hours; self-employed receipts checked with the tax service automatically.
Infrastructure · disaster recovery
A new server within a day of a cloud outage
A cloud outage took down the websites, booking, bots and accounts. Everything came back on a new server and the data was rebuilt without the old disk. Final check: 45 of 45.
Gift certificates · payment and fulfilment
Gift certificates with online payment and a personal link
Buy a cinema or theatre gift without choosing a date. Branded artwork, online payment and protection against repeated issuing; automatic fulfilment is in pilot.
Reports · self-contained documents
A report with its evidence inside
Self-contained HTML with a clear conclusion, full sources and an action plan ready to hand over.
Real report interface with fictional data; text adapted into English for this case. The figures are illustrative. Data tools
Websites, PDFs and job boards into usable data
Records with provenance, reusable snapshots and safeguards against false removals when a read fails.
Also built
Smaller systems, each removing manual work or lost money. The status is stated plainly.
Own analytics and a weekly report nobody assembles
A scheduled collector checks 20 sources — ads, websites, the CRM and bookings — and builds a verified package once a week. ChatGPT writes the findings; the code does the numbers. A full write-up follows the dashboard redesign.
Live
How we work
Results in money and hours, not in technology.
Every system goes the same way: from business rules to a launch that can be rolled back.
Map the process
Where money and time leak, which rules actually apply and which live only in people’s heads.
Data and rules first
One database for clients, deals, bookings and money. Every number is calculated in one place.
AI where it helps
The model understands people and writes; the system supplies the facts. A reply with an invented number never goes out.
Launch in parts
Each process is switched on separately and rolled back with one command. Sales never stop.
Watch the numbers
Conversion, staff hours, errors. If something didn’t work, we say so and fix it.