Product design case study

From struggles to wins using data

Redesigning a candidate management experience of 58k+ monthly active users — and learning, the hard way, what it takes to ship a redesign at scale.

The context

A nuclear reactor

The candidate pipeline is where recruiters live. It's the screen that lists candidates for a job and shows each profile side by side — and it carries more traffic than anything else we build.

58.7k
users / month
11.1M
pages viewed / month
36.7M
clicks / month

When we mapped every action in the product against how often it's used and how many people touch it using the Red Routes matrix, the candidate pipeline dominated the top-right corner: the things people do always, and that most or all customers rely on. That's a thrilling place to improve — and a terrifying place to break. Every design decision here is multiplied by millions of clicks a month.

Before & after

One year of work

Gathering pain points, customer interviews, endless iterations, an alpha release and a closed beta — all to reshape the screen recruiters use most.

The before The original candidate pipeline: a dense two-panel layout with the candidate list on the left and the profile timeline on the right.
…and after The redesigned candidate pipeline: a cleaner two-panel layout with a refined profile, tabbed sections and more breathing room.

Left: the layout recruiters had used for years. Right: the redesign that went through alpha and closed beta before reaching everyone.

The redesign process

How we listened

A redesign this central can't run on opinion. We looked into qualitative and quantitative signals from the first sketch to the open beta.

Gathering feedback & pain points

  • Customer requests routed from support (Savio)
  • Session recordings of real usage (Hotjar)

Metrics

  • Product analytics to prioritise feature importance (Heap)
  • Account & backend data

Release approach

  • Alpha release
  • Closed beta release
  • Open beta release

User research

1
Design Sprint workshop
89
user interviews13 discovery, 6 alpha, 68 beta
24
moderated prototype tests
5
card-sorting iterations
1
tree test
2
rounds of unmoderated tests
1
internal survey9 participants representing insights from 350+ customers
Towards the beta release

The beta that bit back

We'd done the research. We were confident. Then the open beta went live — and the feedback landed like a punch.

“It is useless. I hate it.”

— one of the 1,719 feedback responses we received

We had tested extensively — so why the backlash? Three things we hadn't weighted heavily enough:

👉 Real-life scenarios

Despite all the interviews and testing, day-to-day use of a product is different. People behave in the flow of real work, not in the frame of a study.

👉 We shipped to the least-used area first

To experiment safely, we rolled the redesign out first in a less-used area (the Candidates page) before touching the "nuclear reactor" — the candidate pipeline's job view.

👉 The release process backfired

Because the redesign was already live on the Candidates page, the beta for the candidate pipeline became available without an opt-in. Users were switched over without choosing to be — which caused frustration and even a drop in PSAT.

What this case study is about

Turning a rough launch into a roadmap

Establish a purpose

Improve the candidate pipeline & profile in the weeks and months after the beta release — recover trust and make the screen genuinely better.

Show the journey

Using user feedback and multiple data sources — quantitative and qualitative: recordings, interviews, analytics — we identified real pain points and real user needs. With plenty of ups and downs.

The journey

How improvements affected user feedback

…along with our mood. Each fix moved the needle — some down before they went up. Here's the ride, one problem at a time.

🎉 Screen sizes Timeline default Custom fields Timeline filters
From the release high (🎉), down through screen sizes and custom fields, then a steady climb back up.
Deep dive · Combine data

The mystery of screen sizes

Our first big surprise came from a question we thought we already knew the answer to: how much screen space do recruiters actually have?

Assumption vs reality

Big monitors ≠ big windows

🚧 Assumption

Most of our users work on big monitors, so they have a full and clear view of the candidate list and profile.

✅ Reality

Most users do have big monitors — but they multitask, running several apps in separate windows (e.g. Candidate profile next to Google Meet during interviews). The effective window is much smaller than the screen.

The data

Three sources pointed the same way: Hotjar feedback, session recordings, and analytics on window width vs. screen size.

"Struggle to see the entire layout on a small laptop screen."
"There isn't much screen left to look at each candidate."
"The new format can't handle different page sizes."
5.61M sessions
Window inner width (px)
1340–155932.6% · 1.83M
1780–199926.3% · 1.48M
1120–133913.6% · 760k
1560–177912.3% · 693k
2000–22194.3% · 239k
2440–153604.1% · 232k

The single most common window width was a modest 1340–1559px — not the expansive canvas we'd designed for. Nearly half of all sessions ran below 1560px wide.

The fix
The redesigned candidate pipeline annotated with four fixes: collapsible candidate list, non-sticky stage bar, sticky action bar on scroll, and a more compact header.
  • Collapsible list — reclaim horizontal space for the profile when needed.
  • Non-sticky stage bar — free up vertical room as you read.
  • Sticky action bar on scroll — keep key actions reachable without eating space.
  • More compact header — less chrome, more candidate.
Deep dive · Interpreting feedback

Timeline tab: too smart to be good

Sometimes a helpful idea is too clever. Our "smart" default for the profile tab is a case in point.

Assumption vs reality

When "smart" gets in the way

🚧 Assumption

If someone has already viewed a candidate's details once, next time they'd rather land on the Timeline to see all the latest activity.

✅ Reality

"Smart" system behaviour is hard for users to understand and predict — and unpredictability creates frustration.

The data
"I dislike how, when you click on a candidate profile, it takes you straight to the Timeline tab."
"You need to go back to the Profile tab to remember the CV & the candidate most of the time."

Analytics backed the complaints: tab navigation went up and time-on-candidate went up — people were doing extra work to get back to the view they actually wanted.

The fix
The redesigned profile annotated 'always land here', pointing at the Profile tab set as the default.
  • Always land on the Profile tab. Make the default predictable rather than clever — the candidate's CV and details are what people come for first.
Deep dive · Funnels

Custom fields: the click after click

A pattern that feels fine once can become punishing at scale. Editing custom fields in a modal was exactly that.

Assumption vs reality

One modal is fine. A dozen isn't.

🚧 Assumption

Editing a field in a modal is quick and intuitive — one modal is OK.

✅ Reality

Most users edit custom fields one after another — for example completing several during an interview. One modal becomes a looot of modals, and the flow falls apart.

The data
"Fill out custom fields without having to click each one and respond in a pop-up window. I fear this will take much more time when custom fields are used to pass quick information from recruiters to hiring managers."
"They use custom fields for evaluation — too many clicks, they lose the flow during the interview."

A funnel of users filling in a 2nd & 3rd field within one minute confirmed it — the biggest drop-off happened right after the first answer.

1 · Add answer3,896 users
Step 1
2 · Add answer2,552 users
Step 2
↓ 65.5% convert from step 1 → 2
3 · Add answer1,673 users
Step 3
↓ 65.6% convert from step 2 → 3 · 43.0% total conversion
The fix
The redesigned profile annotated with 'inline editing, optimized to use during interviews' and 'not losing context of the profile'.
  • Inline editing — edit fields in place, optimised for filling in during interviews.
  • Keep the context — no pop-ups, so users never lose sight of the profile.
A/B experiment · ongoing

Optimising the information architecture

The climb continues. We're testing whether a simpler, more versatile structure makes information easier to find across Profile, Timeline, Communication, Evaluation and Comments.

Assumption being tested

Does a simpler structure help people find things?

🚧 Assumption

Offering a simple & versatile structure will make finding information easier.

🔬 Rollout strategy

Three groups — a control group, an optional pilot (10%) and a forced pilot (5%) — with opt-in and opt-out of the feature available throughout.

What we're measuring
  • Pilot retention
  • A UX-lite & subjective satisfaction survey, run for each group separately
  • Time-on-profile

This time, the rollout itself is designed as carefully as the feature — with a control group and a real choice for users.

Lessons learned

The hard way

Everything above distilled into five principles I'll carry into the next redesign.

🧪

Test one thing at a time

Build a habit of testing — and test in live environments whenever you can. Studies don't fully predict real use.

🎉

Design the rollout

Design not only the new feature, but the way it's introduced to users. Consistent, opt-in patterns matter as much as the UI.

👀

Monitor after release

Tracking a feature after launch is vital to see how it really performs — the work isn't done at ship.

🧠

Combine data

A single source can't tell the whole story. Data triangulation — quant plus qual — is the way to go.

📣

Evaluate user feedback

Take feedback with a grain of salt: disappointed users are far more likely to speak up than happy ones. Weight it, don't just count it.