AI implementation01 Chaos02 Breakdown03 Order04 Projects05 System06 Your moveUAEN
01 / Chaos

Excel,
contracts,
messages

10–15 files from different partners, each with its own structure. An employee merges them by hand and worries about missing a line in a note.

Free 30-minute review →
Scroll ↓
02 / Breakdown

AI
sorts
it out

Every condition in its place: covered, covered with a limit, not covered, unclear. Where the system isn't sure, it says so.

03 / Order

One
table

Every figure links to a sheet and a cell of the source file. You can check it in one click.

Egor Muratov

We remove routine from business with AI. We started with our own.

StopRoutine is an AI implementation agency. Its founder and director is Egor Muratov, owner of the insurance broker Companion. Here is what already works in his companies.

Who it's for:insurance and financereal estate and rentalsbusinesses that live in Excel and 1C
What you get

Your staff work with clients, and AI does the routine

⏱Less manual work

Summaries, reconciliations and retyping from file to file happen on their own.

💬Client answers 24/7

The assistant answers routine questions at night and on weekends, based on your documents.

✓Figures without errors

Every number can be checked against its source. If there is no data, the system says so.

🔒Everything under control

Nothing goes to clients without your OK. You see what the AI is doing.

04 / Projects

Six
projects
from the founder's companies

Each one started with a routine that was eating the team's time.

05 / System

Works
on its own

Summaries, reconciliations and replies happen without you. Nothing goes to clients without your OK.

What can be automated

If it repeats every day, it can be handed to AI

  • Comparing offers from suppliers and partners in different formats
  • Answers to clients based on contracts, terms and instructions
  • Contract review for departures from your terms
  • Accounting and reconciliations instead of old 1C and Excel
  • Correspondence with contractors: agree, remind, check
  • Website inquiries straight to Telegram with the ad source
Review my routine →
How we start

Three steps, and at each one it is clear what you get

  1. Review30 minutes free · full department audit from $1,500, credited to the pilot

    We look at where the team loses the most time.

    You get a plan: what to automate first and what the benefit will be.
  2. Pilotfrom $4,500 · 3–4 weeks · price and timeline fixed before the start

    We automate one process using your real data.

    You get a working tool at a fixed price and timeline.
  3. Growthfrom $450/month · when you are ready

    We monitor the work, refine it and connect the next processes.

    You get a system that grows with your business.
Book a review →
About us

An agency led by an entrepreneur who automated his own companies first

StopRoutine's founder and director, Egor Muratov, runs the insurance broker Companion, which has been on the market for 12 years. In 2026 he moved its routine to AI: tenders, accounting, contracts, client answers. That is why Egor speaks to you as one owner to another, and we offer only what pays off.

We work under an NDA. We take access only to the data the task needs.

15 000+insured with Companion
70+Companion's client companies
~97%of Companion's clients renew their contract
Frequently asked questions

What people usually ask before starting

How much does it cost?

The first 30-minute call is free. A full department audit starts at $1,500 and is credited to the pilot. A single-process pilot starts at $4,500 and takes 3–4 weeks; support and growth start at $450 per month. We fix the exact pilot price and timeline before any work begins.

What happens to my clients' data?

We sign an NDA before you show us any documents. Nothing goes to clients without your OK, and every number in the reports can be checked against the source file.

Do I need my own programmer?

No. We set up and launch everything ourselves; you only need to explain how the process works today.

What happens after launch?

We monitor the work, fix errors and connect the next processes if you want.

What if the AI makes a mistake?

Where accuracy matters, a person checks the result. If data is missing, the system says so rather than guessing.

06 / Your move

What eats
your time?

Tell us in 30 minutes, and we'll show you for free what can be handed to AI and where to start.

Book a review →Telegram: @EgorMuratov
Health insurance · for broker clients

Health insurance app: an insurance consultant in your pocket

A mobile app for employees of companies insured by Companion. People find out for themselves what their insurance covers, find a clinic and get help if they are refused.

Status: under App Store reviewPlatforms: iOS, AndroidLanguage: Ukrainian
What was optimized

Questions that used to go to a manager

Before

  • The client's employee calls the broker's manager: “is this test covered”, “which clinic can I go to”.
  • The policy and the program sit as a PDF somewhere in HR's mailbox.
  • After a payment refusal, the person cannot tell whether the insurer is right.

After

  • An answer in the chat within seconds, based on the documents of their own program.
  • Policy, coverage, limits and clinics are always on the phone.
  • When a claim is refused, the AI checks the refusal against the program terms and passes the disputed case to the broker.
How the AI chat works

Answers only from your program's documents

A person writes a question in ordinary words, for example naming a diagnosis or a test. The system finds the relevant passages in that employee's own contract and insurance program and answers: covered or not, with what limit, at which clinics, and what to do next.

If the documents have no answer, the AI says so and suggests writing to the broker. It does not diagnose or advise treatment: it is a guide to the insurance, not a doctor.

A separate scenario for refusals. If a service was not paid for, the AI checks what the program says, and if the refusal contradicts the terms, it says so directly and passes the case to the broker's team.

What else is inside

Everything about your insurance in one place

Sign-in in a minute

By phone number, the code arrives in Telegram. No passwords.

My policy

The program, validity period, what is covered, limits and the contract itself.

Clinics

Only those included in the program, with their classes and a route in Google Maps.

Test results from a photo

The AI reads the lab form, flags deviations from that laboratory's norms and suggests which doctor to see. Files are not stored anywhere.

Booking and help

Contact with the insurer's assistance service by phone, Viber or Telegram.

Broker on hand

Coverage refused or something went wrong: the request goes straight to Companion's team.

Relatives

A request to insure family members and a list of what the program does not cover.

Reimbursements

Which documents are needed to get money back for what you paid yourself.

HR dashboard

The client company's HR sees its insured employees and requests.

Screens
FlutterNode.jsSupabaseClaude and other modelsTelegram sign-in
Health insurance · tenders

Health insurance tender: all offers in one table

An internal broker tool. A client company chooses insurance for its employees, 10–15 insurers send offers, and the system merges them for a fair comparison.

Status: working, the first full run on a real tender is under wayUsers: Companion's brokers
What was optimized

Manually comparing dozens of Excel files

Before

  • Each insurer sends its own Excel offer with its own structure and wording.
  • An expert copies the terms into a shared table by hand and looks for what is covered where.
  • It is easy to miss a limit or an exclusion hidden in a note.

After

  • Files are uploaded as they are, with no fitting to a template.
  • Terms are sorted into a single list: covered, covered with a limit, not covered, unclear.
  • Every value links to a sheet and a cell of the source file and can be checked in one click.
How it works

The broker's path, step by step

  1. Client request

    We upload the client's requirements and build a matrix of terms.

  2. Choosing insurers

    We mark the participants from the directory of insurance companies.

  3. Uploading offers

    We drag each insurer's file onto it.

  4. Comparison

    The AI matches the insurers' wording against the client's terms.

  5. Expert decision

    Where the system is unsure, the broker answers right in the table.

  6. Clinics and service

    Clinic coverage by city and a 45-question assistance questionnaire.

Main ruleThe AI can extract and sort data, but only a person marks it as verified. An empty cell does not turn into zero, and an uncertain match is flagged as uncertain.
For the client

The client gets a clear recommendation

In the end the broker shows who has better coverage, exclusions, clinics, service and price, and gives a recommendation with arguments. The decision stays with the client.

The insurer's service questionnaire is compared as it is: if one company writes “83%” and another writes “we don't count” (in Ukrainian), that is visible without any “smoothing over”.

Next.jsNode.jsSupabaseExcel parsingTwo AI agents: developer and checker
Health insurance · legal review

Contract review: AI finds where the insurer departed from the terms

After the tender the insurer sends a contract. The broker has to make sure it contains all the edits the broker usually insists on. Now the system does this.

Status: working inside the tender toolFormats: Word, Excel
Contract on a desk
What was optimized

Before

  • A lawyer or a broker reads the whole contract, relying on memory and past emails.
  • The contract arrives in parts, in several files.
  • The letter to the insurer with the edits is written by hand.

After

  • The system compares the contract with the broker's registry of standard edits.
  • Several files are checked as one contract.
  • Ready-made paragraph wording and a letter to the insurer with the edits.
How it works
  1. Upload

    Up to 8 contract files, in the order of the parts.

  2. Finding departures

    Each finding points to its paragraph and an example from past contracts.

  3. New wording

    The AI rewrites the paragraph the way the broker usually demands.

  4. Letter and file

    A letter to the insurer and a Word file with the edits in track-changes mode.

  5. Learning

    Accepted edits are saved and taken into account in later reviews.

Health insurance · back office

Broker accounting: from 1C to the browser

A web system for tracking contracts, payments, insurer reports and managers' commission. It reproduces the workflow of 1C “SB Companion” but works from anywhere and does a lot on its own.

Status: in use by the teamRoles: head, curator, manager
Contract card
What was optimized

Before

  • 1C opened only through a remote desktop.
  • One person calculated the commission by manager.
  • Insurer reports and payments were reconciled by hand.

After

  • Work happens in the browser; everyone has their own login and permissions.
  • Commission is split between the curator, the manager and the bureau automatically, and the manager sees their own income.
  • The system reconciles reports and bank receipts with contracts.
How it works
  1. Contract

    Policyholder, insurer, amounts, rate, commission and participants' shares, checked to total 100%.

  2. Payment schedule

    The system reminds about overdue payments and contracts due for renewal.

  3. Insurer report

    You upload the report as PDF, Excel or CSV and the system finds the contracts by number.

  4. Bank

    You paste the bank statement and the system finds the receipts and matches them to the right report.

  5. Profit calculation

    Commission for the period is distributed among managers with a breakdown by contract.

How it was moved from 1CThe configuration was available only on screen, with no export. The logic was reconstructed from the forms and checked on a real contract: premium UAH 440.80, commission 110.20, curator 73.28, manager 31.41. It matched 1C to the kopeck.
Screens

Screenshots were taken on an empty demo database, with no client data.

ReactNode.jsDockerTelegram notifications
Villa rentals · Koh Phangan

An AI manager for two villas on Airbnb

One agent looks after guests, the cleaner and the pool technician. It runs in the cloud around the clock and writes to the owner only when a decision is needed.

Status: runs on its own, 2 villasLanguages: Russian, English, Thai
What was optimized

Before

  • The owner answered guests on Airbnb personally.
  • Arranged every cleaning with the cleaner personally and checked it was done.
  • Scheduled the pool technician personally so that he would not disturb the guests.

After

  • The agent drafts replies to guests; the owner just taps “OK”.
  • Cleanings are arranged, reminded and checked without the owner.
  • The owner learns at once, at any time, about a missed cleaning, a breakdown or a refusal.
How it works
  1. Booking

    The agent sees the calendar and guest messages for both villas.

  2. Guests

    Instructions the day before arrival, a selection of places on the second day, answers to questions. Sent after the owner's OK.

  3. Cleaning

    A couple of days ahead it proposes a time to the cleaner, reminds the day before, and on the day waits for “done”.

  4. Pool

    Once a week, at a convenient time and with the guests as the priority. It forwards the pool technician's photo to the owner.

  5. Owner

    Reports and anything that needs a decision go to Telegram.

Safeguards in code, not in the promptQuiet hours, the daily message limit and the ban on repeats are checked by the program before sending. A model can change its mind; a check in code cannot. The agent has been tested on refusals, reschedulings, counter-questions and false reports of “already cleaned”.
For guests

Online villa books

Each villa has its own page for guests: how the house works, how to get there, beaches, where to eat and what to do if something breaks. It opens via a QR code in the house.

PythonClaudeLINETelegramHospitableSupabaseGitHub Actions
Health insurance · B2B sales

A broker's website that shows where clients come from

A new B2B website for the insurance broker Companion, replacing a site built on a website builder. Every inquiry arrives in Telegram together with its ad source.

Status: workingLanguage: Ukrainian
Website home page
What was optimized

Before

  • A website-builder site, inquiries without a source.
  • No way to tell which advertising pays off.
  • Presentations for clients were assembled by hand.

After

  • Forms by line of business, the inquiry goes to Telegram instantly.
  • Each inquiry shows the campaign, source and keyword.
  • Personal presentation pages with video for a specific client.
Pages
AI video clips

Explanations of the tender and the app in short clips with AI voice-over, without a studio.

Calculator and checklist

Health insurance cost and a program-review checklist for HR.

Fast site

The home page is pre-rendered and opens even without JavaScript.

Small projects

AI helps with small things too

Not everything has to be a big system. Sometimes a small tool that saves half an hour every day is enough.

The Maths Pets game
Maths Pets

A game that makes the child ask to practise

The task. The founder's daughter, in year 3 of a British school, finds it hard to add two-digit numbers with carrying, and also has to learn the multiplication table.

How it works. A game in the tablet's browser, in English. Solved problems grow a collection of animals and their little house. After a mistake the problem is worked through by tens. Dad gets a short report once a day: how long she practised and which mistake is getting in the way.

Improv House

A website for an hourly-rental hall

A page for a hall on Koh Phangan with booking through WhatsApp. The next step in progress: agents that manage bookings themselves.

LayoutFix

Keyboard layout fix on Mac

Typed “ghbdsn” instead of “привіт” (Ukrainian for “hello”)? The utility converts the text to the right layout by itself.

Tell us what eats your team's time. In 30 minutes, for free, we'll show you what to hand to AI.

Book a review → Telegram: @EgorMuratov