The Ozon Seller Knowledge Base was already well maintained. It had hundreds of articles, regular updates, and even its own CMS. But there was a problem: sellers still preferred contacting support. Even when articles existed, they didn’t always lead to resolution.
So we asked ourselves:

• Why do sellers turn to support when answers are in the KB?
• How do they search, and where do they drop off?
• Can we make the KB not just usable, but useful?

This mindset shift was critical: we stopped treating the knowledge base as just a content repository. Instead, we started thinking about it as a product.

Our first move wasn’t to rewrite articles. It was to look at the research that already existed. Several product teams had conducted their own studies around onboarding, promotions, and seller experience. We synthesized findings from six previous studies:
What we found:

• Promotions and advertising are the most common pain points. Sellers want to act fast and need simple, actionable guidance.
• Onboarding and first-sale steps confuse both new and experienced sellers.
• Tariffs, commissions, logistics are relevant across all lifecycle stages.

These findings helped us prioritize content topics. But they also pointed to a more fundamental issue: sellers couldn’t find what they needed, even when it existed.

To validate and expand on previous insights, we launched a two-track research initiative:

• A quantitative survey with 300+ respondents (138 completed it fully)
• A series of moderated usability tests and in-depth interviews across desktop and mobile

We didn’t just ask whether users liked the KB. We traced how they searched for information, where they gave up, and how they interacted with structural elements like cards, search results, and toggles.

We identified two distinct behaviors in how sellers use search:

• Confident users ignored search suggestions and typed full queries based on their own vocabulary.
• Uncertain users heavily relied on suggestions, iteratively tweaking keywords until relevant results appeared.

But both groups hit friction:

• Linguistic mismatches in inflected Russian terms made the system brittle.
• Low relevance or mismatched mental models led to irrelevant articles.

On mobile, suggestions were often hidden or overlooked due to UI issues like keyboard overlap or visual design.

When sellers didn’t know what to ask, they used the homepage cards. But the card structure revealed several issues:

• Vague or unfamiliar section titles deterred exploration.

Overly broad clusters of articles forced sellers to read through too much text to find the answer.

Even when sellers landed on the right article, they often struggled to extract value:

• Many missed key information hidden inside toggles or long paragraphs.
• Most scanned using Ctrl+F, headings, or visual cues — and were frustrated when those didn’t match their query.
• Sellers consistently requested step-by-step formats, screenshots, and video examples.

This exposed the gap between how the knowledge base was structured (by internal logic) and how users actually looked for help (by tasks, errors, or business goals).

We approached redesign like a product team: combining UX principles with product metrics. Our framework had three layers:

After rolling out the first round of improvements:

• Support contacts on well-covered topics dropped
• More sellers completed tasks without leaving the KB
• We saw improved engagement with in-article navigation and visuals

We also improved seller sentiment. Open responses in feedback forms started referencing clarity, speed, and structure — the same issues that originally sent them to support.