Portfolio

2026

Numbers. Rules.
A system that runs without you.

Management systems for businesses whose revenue depends on their clients and their team.

Metrics tied to profit. Processes people actually follow. Tools, automation and AI agents that keep the system running.

Katerina Bokova

Business management systems architect · strategy · data · automation · AI development

I design and build business management systems, from metrics and incentives to tools, automation and AI agents.

10+ years in tech · departments built from scratch · products shipped without a development team

The four-layer method

Selected cases

The same four layers every time. Only the scale changes.

Numbers

Metrics tied to profit. A handful, not dozens.

Rules

Procedures and incentives built around those metrics. Clear to the team.

Tools & automation

One workspace instead of chats and spreadsheets: website, CRM, portal. Repetitive steps run on their own. Built around the process.

Agents

AI handles the routine: replies, reminders, reports. People make the decisions.

Case 01

Operating model for a service business

A department from scratch: metrics, app monetization, pay for results, processes built for growth

Case 02

Team management platform

40+ screens, 3 roles, 3 languages, AI interviews

Case 03

AI agent system

One coordinating agent and two specialists, with checkpoints

Case 04

Analytics that change decisions

What really drives revenue, and what only seems to

Case 05 · in progress

Lead-handling loop

Website + CRM + AI agent: no lead gets lost

Case 01 · Operating model

A department from scratch: from a few metrics to a monetization model

Emotional support companion service · role: head of department, owner of metrics and processes

Starting point

Revenue depended on how the companions worked: how fast they replied, how good the conversations were, whether clients came back to them. None of this was measured. Reports were put together only on request.

What was built

Top-level metrics: consumption volume, customer retention, cost of service relative to revenue
The department's analytics foundation: metrics, weekly reporting, data quality checks
App monetization: a model built from scratch
Hiring from application to first day on the job, a supervisor tier, a conduct policy and a process for handling breaches

Why it worked

The team works on tasks it understands.

Incentives are tied to customer retention, not to hours logged. The team works towards LTV even when no one is watching.

For your business: a few metrics the team influences every day, with incentives tied to them.

Case 02 · Development

Building a team management platform

Same department · role: product owner and developer, from concept to launch

The rules from Case 01 needed a tool: hiring, onboarding, training and oversight lived in email and spreadsheets. The result is a portal that brings AI interviews, hiring, agreements and internal communication together in one place.

40+ screens
3 roles
EN · ES · RU
AI interviews
2FA

What's inside

Two ways in for candidates: a personal invite link, or an open application with an AI interview whose results go straight into the candidate's profile
From applicant to team member in one step: create the account and send the login details
An internal Slack: tasks with comments and reactions, direct messages, a Q&A wall, notifications
Versioned agreements: each new version is signed at the next login
A personal dashboard for each team member: a transparent earnings model, KPIs, a knowledge base

For your business: order in tasks and hiring without buying expensive software. Built around your process.

Case 03 · AI agents

A system of three AI agents that runs sales and content while a human makes the decisions

In-house agent infrastructure · role: system architect and author of the operating rules

Architecture

Coordinating agent

Manages tasks, priorities and daily checkpoints. Delegates to two specialist agents and pulls their output into a single report.

Project agent

Runs the sales pipeline: finds and qualifies prospects, drafts emails, maintains the contact database and tracks the weekly metric.

Content agent

Runs a media project: short-form video scripts, copy, editing, the publishing schedule.

Infrastructure

A dedicated server, an approval interface, long-term agent memory on PostgreSQL + pgvector, backups in Git. My files, my rules, my control.

Rules that keep the system safe

Three checkpoints a day: agents report with deliverables, not with “everything's on track”
Task cards: Backlog → Proposed → Approved → In Progress → Done → Blocked; priorities P1–P3
For internal tasks, silence means consent: the system doesn't wait for a human to keep moving
Any external action (an email, a post, a message) only after an explicit “Approve”

For your business: the same setup can run your process, from the first reply to a client to weekly reports. A human approves; the agent does the work.

Case 04 · Analytics

What really drives revenue, and what only seems to

Large online product · role: lead of a 9-person team, owner of the area's KPIs

The challenge

The team's KPIs measured things the team couldn't influence. Hypotheses were argued on gut feeling. Financial models for leadership were assembled by hand, with no cross-checking.

What the data showed

The metric that actually explains revenue and retention. And fashionable metrics that explain nothing
A costly hypothesis, “cut the weakest performers”, that the data showed would lose money: most revenue came from the mass segment
An error in the financial model, caught before it reached leadership: the total was understated by almost half
The main reason partner activity fell: the penalty system, not the market

What changed

Team KPIs rebuilt on data, not opinions
Investment in partner incentives justified and approved
Reporting methodology fixed: the reports stopped “lying” where segments overlapped
A redesign of partner incentives proposed instead of tighter control

For your business: a single source of truth instead of scattered reports, and confidence that the numbers behind your decisions aren't lying.

In progress · Lead-handling loop

Renovation and interior design: a system where no lead gets lost

Companies with high-value projects and long sales cycles · product: website + CRM + AI agent working as one loop

Where the money leaks

In renovation and interior design, clients rarely walk away over price. They walk away because nobody called them back within the hour, the chat thread got lost, or nobody followed up a week later. The leads are there. What's missing is a loop that carries them all the way to a signed contract.

How the loop works

Numbers: a loss calculator shows the owner how much revenue slips away every month
Rules: every lead gets a reply within N minutes and has a status and an owner; none is closed without a reason
Tool: a website that captures leads and a CRM where they don't get lost
Agent: the first reply to the client, reminders for the sales manager and a weekly report: how many came in, how many were lost, and why

For your business: the calculator runs your numbers in two minutes. If the result isn't zero, the loop pays for itself with the very first contract.

Ways to work together

Three formats, one goal: a system that runs without you

“Order in 30 Days” diagnosis

For owners who feel they're losing money but can't see where. In one month: your numbers, your bottlenecks, and rules the team will actually follow.

Diagnosis

“Website + CRM + agent” loop

For those who already know leads are slipping through the cracks. Built around your process: capture, handling, reminders, reporting.

Describe your challenge

AI agents and tools built around your process

For those who have order but no time. Agents take over the routine; a human approves.

Request terms