How analytics enabled Halyk Market to scale its product range threefold

Aida Akhmetkereeva, Editor-in-Chief of WE Media Group

The end of August 2024. Walmart is gathering thousands of sellers for its Let's Grow! summit and reporting a 30% growth in its marketplace for the quarter. The number of sellers on the platform has grown by 20% over the year, and the assortment is expanding into new categories — from premium cosmetics to collectibles. Analysts predict that by 2030, marketplaces will drive more than half of all e-commerce. The global dropshipping market is valued at $226 billion and is growing by 23% annually.

Everyone is talking about growth. But almost no one explains how it happens technically — especially when you don’t have an Amazon-sized budget, yet your ambitions are quite comparable. Western companies announce plans, allocate millions for expanding teams and technologies. And somewhere in Kazakhstan, while everyone else is discussing this, a small team is already doing it. And showing results that many major players could only dream of.

From January to June 2024, Halyk Market — the e-commerce platform of Kazakhstan’s largest bank — tripled its assortment. It broke its annual KPI of one million SKUs in June, six months ahead of schedule. These figures are in the bank’s financial reports; no one embellished or fabricated them. They simply took action and got it done.

The question is how.

30 people or one system

The usual scenario for such growth looks predictable: you hire an army of content managers who manually create product cards, fill in descriptions, and upload photos. A team of up to 30 people is normal for a marketplace with moderate ambitions. The process is slow, expensive, and poorly scalable. If you want to grow faster, you hire more. But what if you tried a different approach?

In early 2024, a new Head of Analytics joined Halyk Market. Roman Beseda — a professional with a background at Yandex, Delivery Club, and Palta, accustomed to solving problems with systems rather than people. After studying the current processes, he saw that manual product card assembly was slowing down assortment expansion and creating a bottleneck that affected plan fulfillment.

The alternative was full process automation. This was not about point improvements, but about rebuilding the entire chain: from data acquisition to displaying the product card on the storefront. The concept included collecting data from external sources, loading it into the database, automatic matching, basic quality checks, and publication. Development required a compact team — no need for 30 new employees, just one developer-analyst and systematic work on architecture. And Roman himself as the architect of the entire system.

The analysis of bottlenecks began in the first weeks, and by May, the new system was already functioning: Four months for full implementation of a solution that replaced dozens of people and allowed for threefold growth.

Scrapers, LLMs, and no magic

The technical solution looked like a classic ETL pipeline, but with a few tricks. External scrapers collected product data from open sources with assortment information — names, descriptions, prices, and specifications. Then this data went through a unified pipeline with automatic cleaning and validation. The result was a ready-made product card that entered the database and was mapped to the storefront via API.

The most interesting part began during the mapping stage. Matching thousands of products from different sources with existing categories and attributes is a non-trivial task. Roman used LLM models that analyzed descriptions and automatically determined where each product should be placed. This reduced manual adjustments by 70%. Previously, content managers spent hours manually verifying positions. Now the system did this itself, leaving people only with the final verification of disputed cases.

But technology was only half the battle. The main difficulty lay elsewhere: integrating numerous internal and external services, connecting scrapers with databases, databases with the storefront, wrapping all this in a reliable layer of microservices with centralized logging and monitoring. Meanwhile, the team on the project was minimal: two people, clear architecture, and a clear understanding of what needed to be done. In parallel, Roman was building the analytics department from scratch — hiring people, building infrastructure, launching A/B tests, but he pulled off the content automation project almost single-handedly.

Roman took on the coordination of all processes: from writing the technical specification to final testing. Throughout the project cycle, Roman held regular demos for e-commerce management and the CIO. When stakeholders provided feedback, he made adjustments promptly. When the system went live, he prepared documentation and conducted workshops for the content and support teams. The handover went smoothly.

x3 instead of x2, one million in June instead of December

The result exceeded all expectations. Instead of the planned doubling of the assortment — a tripling. Instead of one million SKUs by the end of the year — one million in June. Operational costs for content went down. The speed of adding new products went up, and significantly so. And this was not just a one-time success: the system remained in the form in which it was implemented. It works, scales, and does not require constant modifications.

But Roman looked at this more broadly than just “launched a project.” For him, this was a story about how analytics stops being a service function and becomes a business driver. When he joined Halyk Market, analysts created dashboards on request. Someone from the business came with a task — they executed it. The classic model: analytics as a support department.

Roman changed this. He taught the team not to wait for requests, but to find bottlenecks themselves and propose solutions. To immerse themselves in processes, figure out where things break, and come to stakeholders with ready hypotheses. The story of content automation is a vivid example: no one asked analytics to solve this task. Roman himself saw the problem, formulated the solution, and brought it to production.

When analytics stops serving and starts leading

In parallel, he was building a culture within his department. Focus on people and strategy. Without a strong team, it is impossible to build something sustainable. Without infrastructure, it is impossible to develop solutions and ensure their stability. Without quick wins, it is impossible to demonstrate the value of analytics and prove its real benefit. Roman started by forming the core team, defining areas that could quickly show results. Three blocks: infrastructure, major product projects, and systematic efficiency improvement.

Hiring turned out to be one of the most difficult challenges. A new country, no ready-made network, specifics of the local market. But eventually, the team came together — with a good balance between technical and product competencies. Roman focused on “quick deltas”: they optimized operational expenses, built infrastructure on open-source solutions, and fully automated reporting. Then they launched the first A/B tests in e-commerce, implemented systematic metric tracking, and realized several large-scale projects. By the end of the year, the majority of the company’s assortment onboarding processes were running on their solution.

The main principle that Roman preached: the ability to motivate the team and give them interesting, complex challenges. Analysts should move away from the role of request executors to the position of product development partners. When they understand the business goal and see the impact of their decisions, they start thinking not about numbers, but about growth. Second — systematic interaction between analytics, data engineering, and ML. The team works as a single loop: from quality data to product insights and automated solutions. And third — a culture of speed and iterations. Rapid hypothesis testing, measuring effects, and readiness to revise decisions.

Small teams overtake on the turn

The Halyk Market story shows what is rarely discussed at loud summits: marketplace growth is not only about attracting new sellers and expanding categories. It is about infrastructure that allows scaling without proportional cost growth. It is about teams that do not wait for ready-made solutions but invent them themselves. And it is about leaders who understand: analytics is not about reports, but about changing how the business works.

While large companies discuss strategies and allocate budgets, small teams with the right approach can overtake them on the turn. The main thing is not to be afraid to rethink processes and see the solution through to the end. Roman Beseda did exactly that. And the result speaks for itself.