Tamirlan Askar · Product Design Lead
From repainting buttons to real discovery: rebuilding how a design org does research
As a newly promoted Design Lead, I set out to turn design from "repainting buttons" into a real discovery practice. I built an AI-assisted research-prep system and a reusable respondent database, and put Double Diamond + HADI into three product teams. Research prep dropped about 90%, recruitment effort about 85%, and discovery became a standing part of how the teams work, solid enough that it outlived my move to Yandex.
Impact
- ~90% less time on research prep (2h → ~15 min)
- ~85% less recruitment effort via a respondent DB
- 3 teams onto a real Double-Diamond discovery practice
- Durable the practice outlived my move to Yandex
Situation
Design was treated as repainting buttons. The pattern across teams: a hypothesis got handed down, the team built it, and it shipped to nowhere, because it was just a hypothesis, never validated. There was no real discovery and no trodden path for research; I was one of the first to even try, and I hit every wall.
The operational friction was brutal. To run a single study you had to:
- pull a user base from the analysts, and keep it current;
- clean it in an ancient system: strip dead numbers, B2B corporate numbers, and Beeline employees so you didn't SMS your own colleagues;
- prepare an SMS blast and a screener questionnaire, then convert the link into the official `b.gg` Beeline shortcut;
- do it in both Russian and Kazakh (two languages in Kazakhstan);
- then call respondents and run them through a Zoom session.
And recruitment was wasteful: of ~100 people who responded, you'd take ~10, and the other ~90 just fell away every time.
Task
The goal: make discovery a continuous, repeatable practice the design org could run itself, so design produces validated insight instead of executing someone's guess. Concretely that meant fast prep, reliable recruitment, designers who could run their own research, and discovery built into product teams' actual process, not a thing you do "if there's time."
The hardest part wasn't the tooling. It was adoption and protected time: the standing objection was "no time for your research, I'm busy designing, designing, designing."
Action
Listen first
I started by asking designers directly what they hated about the current research process, then rebuilt from their feedback rather than my assumptions.
AI-assisted prep
Our head designer had built a "design-GPT," a bot primed with the memory of being a senior designer that helps produce artifacts. I loaded it with the right instructions and templates so it could generate research plans, interview questions, and prep docs to our standards. Writing those used to take ~2 hours (≈1.5h of it focused); with the templated bot it dropped to ~15 minutes.
A reusable respondent database
The waste bothered me: about 90 of every 100 respondents were thrown away. So we started banking them in a reusable respondent base, and future studies recruited from people who had already raised their hand. That cut recruitment effort by about 85%.
The fraud scare that gave us VoIP
While rolling this out, calling respondents from personal and test numbers nearly got me reported for fraud: people screenshotted the calls, suspicious of who was phoning them. I negotiated with the call center to get us VoIP and per-person licenses, so every call now shows up as "Beeline Research." Recruitment went from sketchy to legitimate overnight.
Building it into how teams work
A faster pipeline means nothing if teams don't adopt it. So I drove the operating model:
- Implemented Double Diamond + HADI in two teams and with two designers, plus my own team: three teams total.
- Ran one-to-ones and mentoring until designers could go and research on their own.
- Took it to the product heads (I remember the conversations with the Simply and Beeline Shop leads) and their product managers: "let's build this into the process together." Everyone was on board; the real work was making it stick.
- Split teams into Discovery and Delivery and ran discovery sprints with protected research time, which finally killed the "no time for research" objection. Real insights started arriving instead of more buttons.
(Earlier, as a mid/senior, I co-wrote a research guide with my lead that became the basis for research across all designers. This initiative grew from that seed.)
Results
- About 90% less prep time and about 85% less recruitment effort. Research went from a multi-day ordeal to something a designer could actually sustain.
- Three teams running Double Diamond + HADI, with designers able to run research on their own.
- When I was promoted (and later left for Yandex), my old position was filled by someone who stepped into working processes rather than starting over. The system outlived me, which is the real test of an operating-model change.
What I can't claim is the downstream product metric: I moved to Yandex before those studies' decisions played all the way out. This case is proof that I can change how a design org works, not proof of one shipped number.
What I'd do differently
- Wire each research cycle to a product decision and a metric from day one. The gap here is exactly the thing I couldn't measure, so I'd instrument the link from insight to decision to shipped change to result, and make the system prove its value in product outcomes, not just process speed.
- Build the respondent database and VoIP first, not mid-flight. Both emerged as reactions to pain (waste, a fraud scare). Standing them up at the start would have saved months and a very stressful week.