B2B analytics SaaS, installed on customers' own servers
Period
2022 to 2026
Context
The product already had paying customers and an interface built over years without a designer. People work in it all day, so every visual change felt like a risk to them. The code was legacy and releases were slow.
Problem
There was no design system, no shared brand, no single set of icons or fonts. Different parts of the product looked and behaved differently. Any change had to be small, safe and easy to roll back.
Visual placeholder: a key screen before and after, recreated with sample data
Before and after. Add: recreate one screen in Figma with fake data.
Constraints
Legacy code and a long development cycle.
Users who dislike surprises and see the product 24/7.
Almost no built-in analytics, because the product is not cloud-based. I could not simply look at what people use.
How I worked
Smallest change with the biggest effect first. I fixed corner radii everywhere. The product immediately looked more modern.
Palettes. Some colors looked great in mockups and broke in production. We rolled them back several times and got a lot of complaints about color.
Dark theme. We shipped it while we were still looking for a light palette that does not hurt the eyes. It softened the reaction.
Contrast checks with real people. Contrast scanners miss things. I invited colleagues who wear glasses to demos, and they saw problems in seconds.
Fonts and icons. We moved to one font family and one icon set, because there was no single pack before.
Cleaning up. We removed dead-end flows and duplicated features, added explanations in the interface, updated the documentation and wrote instructions for every release.
Visual placeholder: palette decisions, what changed and why
Add: a small table of palette rounds without real brand colors, or a recreated version.
Research without analytics
I used what was available: quick hallway tests, short talks with customers, and reading support messages myself. Support knew the real problems best, and most messages were angry, which was useful. The most effective method was also the harshest one: switching a feature off to see whether anyone noticed.
Result
The first reaction was negative, as expected. After the later releases, customer feedback turned positive. The product looked clearly more modern, and there was a system inside it. Different parts of the product now share one core.
What I learned
How a screen looks matters less than how much it costs to build. I stopped drawing expensive things. As the team grew more experienced, we estimated effort before design, using several methods, and made decisions together with developers.