Roman
Larionovskii

Product Manager / E-commerce Analyst

For four years I have run my own store on Ozon, Russia’s largest marketplace, and built an analytics service for it that the team uses every day.

Latest version of this CV and the live project page: larionovskii.tech/en

Message on Telegram

Download PDF

Monthly store revenue; I launched it on Ozon from scratch: ~₽2M
Of sales moved to Ozon from retail and Avito: 75%
More add-to-carts after the product card redesign: ×1.5
Saved for the team every week by my analytics service: 5 h

Why product

What I enjoyed most in the store was finding problems in the numbers and solving them, and my best project turned out to be my own service. I want to do this in a product team, so I am studying for a master’s degree in Digital Product Management at HSE University.

Experience

E-commerce and Product Development Manager

AutoMarket, own store on Ozon

October 2022 - present, 4 years

Store on Ozon

Launch and growth

Researched the market and launched the store on Ozon from scratch; before that the business sold only in retail and on Avito. Now 75% of sales come through Ozon. Grew the range to 2,000+ active SKUs.
Analysed competitors’ prices and ranges, looked for open niches and expanded the range into them.

Product cards, brand and pricing

Improve product cards based on customer reviews and questions and test hypotheses with A/B tests. After the card redesign, shoppers add items to the cart 1.5 times more often.
Registered a trademark and enforce it: removed 50+ stores from Ozon that sold copies of my product. Set up pricing and product card SEO.

Logistics and suppliers

Moved key SKUs from FBS (shipping from our warehouse) to FBO (Ozon’s warehouses). Delivery to Moscow went from 3-4 days to 1-2, coverage grew, and sales of these SKUs rose by 75%.
Manage 2 product categories and work with 2 large foreign suppliers and a logistics partner: set tasks, agree deadlines, track certification. Handed product assembly, labelling and packaging over to the supplier.

Finance

When revenue passed ₽20M, modelled the tax options and moved the business from a 6%-of-revenue regime, which now required VAT on top, to a flat 8% of revenue with no VAT. This cut the tax burden by almost a quarter (≈24%) against the best alternative.
Plan purchases from sales and stock levels.

Cases from the store

Three decisions I made from the numbers. Each follows the same pattern: problem, hypothesis, what I did, result.

×1.5

more add-to-carts

Product card redesign

Problem: Product cards got views, but shoppers rarely added items to the cart. In questions they kept asking about things that were in the description but hard to notice: compatibility, manufacturer and other specs.
Hypothesis: If the answers to frequent questions and the key specs move into the card design, the title and rich content, more visitors will add the item to the cart.
What I did: Went through reviews and questions, redesigned the cards, improved titles and rich content. Ran Ozon’s built-in A/B test on selected products for a month and only then rolled it out to the whole range.
Result: Across all cards, items are added to the cart 1.5 times more often on average.

+75%

sales of key SKUs

Moving products from FBS to FBO

Problem: Key products shipped from our own warehouse, and delivery to Moscow took 3-4 days. Shoppers saw next-day offers right next to ours, and Ozon ranks faster delivery higher.
Hypothesis: If key SKUs are stored in Ozon’s warehouses, delivery gets shorter, the products rank higher, and extra sales outweigh storage and shipment costs.
What I did: Picked SKUs by demand, margin and size, modelled unit economics with FBO costs upfront and planned shipments by sales velocity. Later I automated this calculation in SellerAnalytics.
Result: Delivery to Moscow dropped to 1-2 days, coverage grew, and sales of these SKUs rose by 75%.

≈24%

lower tax burden

Choosing the tax regime

Problem: Revenue passed ₽20M. From 2026, small businesses paying 6% of revenue must also pay VAT above that level, so the tax burden was set to rise noticeably.
Hypothesis: A flat 8% of revenue with no VAT, a special regime for small businesses, would cost less than 6% of revenue, 15% of profit or the standard corporate regime, each with VAT on top.
What I did: Calculated the tax burden myself on our revenue and cost structure for all four options. Checked the eligibility rules for the 8% regime: revenue cap, headcount, bank. The business met all of them.
Result: Moved the business to the flat 8% of revenue with no VAT. The burden fell by almost a quarter against the best alternative.

Projects

2025 - presentin production

SellerAnalytics

Reports and purchasing in the store were done by hand in Excel, and there was no ready-made analytics for Ozon. I built my own service and took it to production: the whole team uses it, and it saves about 5 hours a week.

Ozon data every 5 minutes: stock analytics, automatic supplier orders, financial reports with tax under each small-business regime, labels to EAEU and Ozon standards, team roles.
Processing weekly exports went from 1 hour to 15 seconds. Defended as my bachelor’s thesis with an “excellent” grade.
Wrote a PRD and 8 user stories for the demo mode, with acceptance criteria, requirements and success metrics.
Run the backlog in Jira: epics, tasks estimated in story points, sprints.
Before building FBS returns, checked the Ozon API, found the required methods missing and cancelled the epic.

TypeScript, NestJS, Next.js, PostgreSQL, TypeORM, Jest, Nginx

Live project page with epics and sprint: larionovskii.tech/en/seller-analytics · demo: seller-analytics.ddns.net

About the project Open the demo

2026 - present

Telegram bot and mini app

The bot sends XLSX reports and PDF labels to the chat on a schedule.
Customer questions arrive in the chat, and a reply from the chat is published on Ozon.
The mini app opens without a password and shows revenue, orders, FBO stock and a label editor.
Demo mode: a new user explores the product on a demo store without signing up.

TypeScript, NestJS, Telegram Bot API, Next.js

Telegram bot@seller_analytics1_bot

No sign-up needed: the bot opens a demo store with made-up data right away. The bot speaks Russian.

HSE study projects

Master’s in Digital Product Management, Strategic Analysis course. Six team projects on real companies; in each, a team of five writes a report and defends it.

8.5

out of 10 for the report, “excellent” on the HSE scale

Yandex Lavka: growing beyond Moscow without losing its edge

Project 1 of 6 · September 2026

Task: Reconstruct the strategy of Yandex Lavka, Yandex’s rapid grocery delivery service, from its actions, test whether it holds up as the environment changes, and frame the company’s strategic challenge.
My part: Found and checked the data, did the calculations and pulled the facts into one picture. Identified three megatrends that hit order economics: an expensive last mile, customers moving into ecosystems and AI assistants, and demand for ready-to-eat food. At the defence I presented the strategic challenge, the response options, the roadmap and how to test the diagnosis.
Key finding: Lavka is looking for growth where its model works worst. Its edge, own brand and ready-to-eat food, was built in dense Moscow, but it now wants to grow in the regions through franchising, just as Ozon fresh enters the same cities. Revenue growth has already slowed from 45% to 23% year on year.
Recommendation: Grow in depth, not breadth: 5–7 large cities with its own kitchen factories, and sell its food through Yandex Eats in less dense ones. We named in advance the data that would disprove the diagnosis.

You laid out the trade-off between fast growth through partners and control over execution well. The plan to compare own and partner locations by city, age, demand and season is especially useful.

from the instructor’s feedback, translated

What I would improve

Our hypotheses explained why Lavka chose franchising but did not prove that quality is worse at partner locations. Next step: compare the quality of own and franchised dark stores on data, and frame the challenge without assuming growth through partners.

Methods PESTEL · VRIO · strategy diamond · scenarios · competing hypotheses · falsifiability test

How I can help

Give engineers clear tasks; I took my own service from PRD to production
Work out a feature’s economics before launch
Automate manual routine
Test hypotheses with data and A/B tests

How I work

An AI agent writes the code for SellerAnalytics, and I own everything around it: what to build, why, and how to check it worked. Here is the process on my own service.

Find the problem in the numbers
I look at store and service data and estimate what the problem costs and what a fix would bring. That is how automatic processing of weekly exports appeared: it takes 15 seconds instead of an hour.
Describe the solution
A PRD with goals, metrics and non-goals; user stories with Given / When / Then acceptance criteria. Docs live in Confluence, tasks in Jira with story point estimates.
Brief the AI agent
I give the agent context, requirements, constraints and the story’s acceptance criteria. I don’t write code by hand, but I read every change and understand what it does.
Verify
Change review, unit tests and a run on the demo store before release. If something doesn’t match the acceptance criteria, the task goes back to the agent.
Ship and watch the metrics
Release to the server, then watch errors and the PRD metrics. If the hypothesis doesn’t hold, I revisit the decision: that is how the FBS returns epic was cancelled.

Skills

Analytics and data

SQL
Python (pandas)
Excel / Google Sheets
Working with APIs
Reporting and dashboards
Forecasting

Product

RICE, ICE
PRDs and user stories
Jira backlog
A/B tests
Unit economics
Funnels

Engineering

Python
C++
JavaScript / TypeScript
REST API
Database design
Git
Docker
Linux
Nginx
Figma

Education

2026-2028

HSE University, Graduate School of Business

Master’s, Digital Product Management

2022-2026

Saint Petersburg State University of Aerospace Instrumentation

Bachelor’s, Software Engineering. Thesis: SellerAnalytics, graded “excellent”

First-degree diploma, 2025

For the paper “A method for optimising logistics chains”

Languages

Russian native, English C1

Looking for a role in a product team

Telegram is the fastest way to reach me. Email works too.