Sharon

I’m Sharon, a senior product designer. I build design systems and the product experiences that run on them. I like finding the simplest way to solve a problem through design, and proving it works with real results.

Selected work

Design System: Color Architecture — cover

Design System: Color Architecture

Read case study
Earning Trust in AI Summaries — cover

Earning Trust in AI Summaries

Read case study
Helping Visitors Decide: Tabs vs. Comparison — cover

Helping Visitors Decide: Tabs vs. Comparison

Read case study

Let’s talk.

Open to senior product design roles. Based in Tel Aviv, working in English, Hebrew and Spanish.

sharon.labaton@gmail.com

LinkedIn

CV (PDF)

I’m Sharon, a senior product designer. I build design systems and the product experiences that run on them.

Selected work

Design System: Color Architecture — cover

Design System: Color Architecture

Read case study
Earning Trust in AI Summaries — cover

Earning Trust in AI Summaries

Read case study
Helping Visitors Decide: Tabs vs. Comparison — cover

Helping Visitors Decide: Tabs vs. Comparison

Read case study

Let’s talk.

Open to senior product design roles. Based in Tel Aviv, working in English, Hebrew and Spanish.

Design System: Color Architecture

Image panel

Cappsool and Better Impression, two comparison-site companies acquired by Natural Intelligence, built the same kind of product separately. I led the creation of one shared design system for both, cutting duplicated work while keeping each brand’s identity and years of A/B-tested design.

Result: One system now serves both companies. Themes ship and A/B test faster, with 72% fewer color tokens and 6 shared themes instead of 29.

COMPANY

Natural Intelligence

TIMEFRAME

9 months · Aug 2024

ROLE

Lead Product Designer

TEAM

2 designers · 2 developers

The problem

Repeated color palettes

Overlapping hue families, defined separately. Two sources of truth that drifted apart.

Visual: near-identical blues

Inconsistent color names

Colors were named by mood (“calm”, “calmer”) instead of a scale (“Blue 300”), so the names didn’t show how colors related.

calmer

→

Blue 300

Duplicated design & code

21 verticals across 2 companies, each on its own platform and its own design system. The same features were built from scratch in both.

Cappsool DS

Better Impression DS

Business impact

Slower delivery: 47% of colors were hard-coded, so theme changes meant editing each vertical by hand.

main challenge

Unify both systems under one token structure, without visible changes or losing the details that had already won A/B tests.

Ownership

I led

Color system

Audit

Token architecture

Accessibility decisions

Rating decisions

Aligning product and engineering

I built

Tokens and components

Tokens: corner radius

Atoms: icons, buttons, navigation

Molecules: badges, score

Organisms: partner cards

Documentation and design QA

Second designer

Foundation tokens

Shadows

Typography

Spacing

spacer

How the pieces I built fit together

spacerBuilt components
Discover — Simplified

Consolidation

35 colors consolidated into 7 base tones. Each one seeds a full scale (100, 200, 300…) of primitive tokens.

Token Mapping (Compact)

Architecture

Primitives → semantic tokens → system tokens. Themes can be swapped without touching any component.

Primitive Tokens

FAMILY

SCALE

Blue

Blue 100

Blue 200

Blue 300

Blue 400

Blue 500

Purple

Purple 100

Purple 200

Purple 300

Purple 400

Purple 500

Green

Green 100

Green 200

Green 300

Green 400

Green 500

Turquoise

Turquoise 100

Turquoise 200

Turquoise 300

Turquoise 400

Turquoise 500

Yellow

Yellow 100

Yellow 200

Yellow 300

Yellow 400

Yellow 500

Red

Red 100

Red 200

Red 300

Red 400

Red 500

Grayscale

Grayscale 100

Grayscale 400

Grayscale 500

Grayscale 600

Grayscale 800

Grayscale 900

Grayscale 1000

Semantic Tokens

The partner card is the core conversion component across all verticals.

Token map · theme tabs (interactive)

Either company can use any theme, as long as the two never use the same theme on the same vertical.

System Tokens

CATEGORY

TOKEN

LIGHT

DARK

Foreground

FG-10

Grayscale 1000

Grayscale 100

FG-20

Grayscale 600

Grayscale 400

FG-30

Grayscale 100

Grayscale 100

Borders

Border-10

Grayscale 400

Grayscale 900

Border-20

Grayscale 500

Grayscale 800

Intent

negative

Red 400

Red 300

positive

Green 300

Green 300

positive-hover

Green 500

Green 400

info

Blue 300

Blue 200

Disabled

disabled

Grayscale 500

Grayscale 600

Tracing one color through the layers

Raw value

#1650b7

Frame

↓

Primitive token (Layer 01)

Blue 400

#1650b7

Frame

↓

Semantic token (Layer 02, Marine)

brand/primary

#1650b7

Frame

↓

Applied

Label

From Figma to code

I built the Figma variables and tokens, with one naming convention for designers and developers from day one, and shared them as the single source. An external design-system consultancy guided the structure and stayed on call for support. Today anyone on the design team builds with it.

Decision: ratings

I owned the decision on how to differentiate the two companies’ ratings.

SEO risk

It had happened before: Google flagged near-identical pages as duplicate content.

vs

Proven performance

Those colors had already won A/B tests. Replacing them meant re-testing from scratch.

How I got there

The problem

Both companies used the same rating colors.

Cappsool

score

Golden

Better Impression

score

Golden

✕

Pages looked like duplicates

Iteration 1: What I tried

Rating colors as tokens inside each company’s theme.

Marine theme

score

Golden, locked

Lavender theme

score

Forest, locked

✕

Changing the rating meant changing the theme

Iteration 2: My decision

Two swappable rating systems, Golden and Forest.

Built from the same validated colors, so nothing had to be re-tested.

CHOSEN

Marine theme

score

Golden

Marine theme

score

Forest

✓

Any theme, either rating

Outcome

Golden started as Cappsool’s rating and Forest as Better Impression’s. Now no theme owns a rating color.

Decision: contrast

With no one owning the shared outcome, conflicts between the two teams stalled. The position badge is one example.

Product

Focused on business results. Protected proven theme details.

vs

Engineering

Focused on delivery speed. Wanted one shared system with no per-theme exceptions.

How I got there

The problem

One shared white foreground on every theme.

Marine

badge

secondary

primary-fg

Lavender

badge

secondary

primary-fg

✕

White text unreadable

Iteration 1: What I tried

Darken the yellow so white text stays readable.

Marine

badge

secondary → #F8B700

primary-fg

Lavender

badge

secondary

primary-fg

✕

Readable, but the proven color changed

Iteration 2: My decision

Original yellow, with its own foreground token.

CHOSEN

Marine

badge

secondary

secondary-fg (new)

Lavender

badge

secondary

secondary-fg (new)

✓

Original color, readable text

Outcome

6/6 themes pass AA, 0 component changes.

Validation

Engineering migrated both companies to one platform, one vertical at a time.

How each theme was tested

Journey

Results

All 21 verticals now run on the shared system, with no drop in performance on any of them.

What changed for the teams

Before · Hard-coded

47%

of color values hard-coded

640

color tokens

29

themes

~3–5

days to ship a new theme (est.)

After · Tokens

0%

of color values hard-coded

180

−72%

color tokens

6

−79%

shared themes

~1

days to ship a new theme (est.)

X-ray stage

Most verticals looked the same after migrating; a few needed small adjustments to fit the shared palette.

What I learned

Looking back, most of the back-and-forth came from deciding each conflict on its own. Next time I’d agree with both teams on scope, decision owners and success criteria before the work starts, so disagreements are settled by rules we’ve already agreed on, not one by one.

Let’s talk.

Open to senior product design roles. Based in Tel Aviv, working in English, Hebrew and Spanish.

sharon.labaton@gmail.com

LinkedIn

CV (PDF)

Design System: Color Architecture

Project details

COMPANY

Natural Intelligence

TIMEFRAME

9 months · Aug 2024

ROLE

Lead Product Designer

TEAM

2 designers · 2 devs

Cappsool and Better Impression, two Natural Intelligence companies, built the same kind of comparison product separately. I led the creation of one shared design system for both, cutting duplicated work while keeping each brand’s identity and years of A/B-tested design.

Result: One system now serves both companies. Themes ship and A/B test faster, with 72% fewer color tokens and 6 shared themes instead of 29.

The problem

Repeated color palettes

Overlapping hue families, defined separately. Two sources of truth that drifted apart.

Visual: near-identical blues

Inconsistent color names

Colors were named by mood (“calm”, “calmer”) instead of a scale (“Blue 300”), so the names didn’t show how colors related.

calmer

→

Blue 300

Duplicated design & code

21 verticals across 2 companies, each on its own platform and its own design system. The same features were built from scratch in both.

Cappsool DS

Better Impression DS

Business impact

Slower delivery: 47% of colors were hard-coded, so theme changes meant editing each vertical by hand.

main challenge

Unify both systems under one token structure, without visible changes or losing the details that had already won A/B tests.

Ownership

I led

Color system

Audit

Token architecture

Accessibility decisions

Rating decisions

Aligning product and engineering

Discover — Simplified

Consolidation

35 colors consolidated into 7 base tones. Each one seeds a full scale (100, 200, 300…) of primitive tokens.

Token Mapping (Compact)

Architecture

Primitives → semantic tokens → system tokens. Themes can be swapped without touching any component.

Primitive tokens

The raw palette: each color family in a numbered scale (Blue 100, Blue 200…).

Semantic tokens

The partner card is the core conversion component across all verticals.

Connector map (mobile, labels above/below)

Where the themes started

Cappsool

Marine

Emerald

Sunshine

Better Impression

Lavender

Denim

Aqua

Today either company can use any theme, as long as the two never use the same theme on the same vertical.

System tokens

Shared UI colors for text, borders and status (like negative or positive), each with a light and a dark value.

Tracing one color

Raw value

#1650b7

Frame

Primitive token (Layer 01)

Blue 400

#1650b7

Frame

Semantic token (Layer 02, Marine)

brand/primary

#1650b7

Frame

Applied

Label

From Figma to code

I built the Figma variables and tokens, with one naming convention for designers and developers from day one, and shared them as the single source. An external design-system consultancy guided the structure and stayed on call for support. Today anyone on the design team builds with it.

Decision: ratings

I owned the decision on how to differentiate the two companies’ ratings.

SEO risk

It had happened before: Google flagged near-identical pages as duplicate content.

vs

Proven performance

These colors had already won A/B tests.

How I got there

Iteration 2: My decision

CHOSEN

Two swappable rating systems, Golden and Forest.

Marine theme

score

Golden

Marine theme

score

Forest

Any theme, either rating

Built from the same validated colors, so nothing had to be re-tested.

The problem

Both companies used the same rating colors.

Cappsool

score

Golden

Better Impression

score

Golden

✕

Pages looked like duplicates

Iteration 1: What I tried

Rating colors as tokens inside each company’s theme.

Marine theme

score

Golden, locked

Lavender theme

score

Forest, locked

✕

Changing the rating meant changing the theme

Outcome

Golden started as Cappsool’s rating and Forest as Better Impression’s. Now no theme owns a rating color.

Decision: contrast

With no one owning the shared outcome, conflicts between the two teams stalled. The position badge is one example.

Product

Protected proven theme details.

vs

Engineering

Wanted one system, no exceptions.

How I got there

Iteration 2: My decision

CHOSEN

Original yellow, with its own foreground token.

Marine

badge

secondary

secondary-fg (new)

Lavender

badge

secondary

secondary-fg (new)

Original color, readable text

The problem

One shared white foreground on every theme.

Marine

badge

secondary

primary-fg

Lavender

badge

secondary

primary-fg

✕

White text unreadable

Iteration 1: What I tried

Darken the yellow so white text stays readable.

Marine

badge

secondary → #F8B700

primary-fg

Lavender

badge

secondary

primary-fg

✕

Readable, but the proven color changed

Outcome

6/6 themes pass AA, 0 component changes.

Validation

Engineering migrated both companies to one platform, one vertical at a time.

How each theme was tested

rail

01

A/B test

rail

02

Compare four metrics

rail

03

Equal or better?

fork (mobile)

Results

All 21 verticals now run on the shared system, with no drop in performance on any of them.

What changed for the teams

Most verticals looked the same after migrating; a few needed small adjustments to fit the shared palette.

What I learned

Looking back, most of the back-and-forth came from settling each conflict on its own. Next time I’d agree on scope, decision owners and success criteria with both teams up front.

Let’s talk.

Open to senior product design roles. Based in Tel Aviv, working in English, Hebrew and Spanish.

Earning Trust in AI Summaries

Image panel

Cappsool’s comparison sites earn revenue when users click through to a partner and convert. I led the research, design and handoff specs for an AI summary of real Trustpilot reviews on the top partner’s card, replacing a bare rating with real customer sentiment, without compromising the business model.

Result: A handoff-ready design for desktop and mobile. It didn’t reach its A/B test: shifting priorities and repeated scope changes kept deprioritizing it.

COMPANY

Cappsool

TIMEFRAME

1 month · Sep 2025

ROLE

Senior Product Designer

TEAM

1 PM · 1 designer · 1 developer

The problem

Our top partner underperformed on conversion in pet food, a category where buyers rely heavily on reviews. The card offered only a Trustpilot badge: users clicked through, but few converted into leads.

The business signal

Journey
Hypothesis + card today

The brief

One scannable overview of what customers say, on the top partner’s card: overall sentiment, recurring themes and the general impression.

Design goals

Real sentiment, fair ranking

Surface real customer sentiment without demoting other partners.

1

2

3

AI text people trust

Make AI content credible and clearly labeled.

✦ Based on reviews created with AI

Few card changes

Lift conversion rate with minimal changes to a proven card.

AI Summary of ★ Reviews ⌃

The challenge

Add AI that builds trust, without losing what drives clicks to our partners.

Real reviews wouldn’t always support our ranking.

User trust

and

Business model

Research

Earlier user testing I ran on Cappsool’s comparison sites, plus a review of how Gemini, Booking.com and Trustpilot present reviews, pointed to one principle: people trust real users more than any other source.

Refs panel

Explorations

I started with the structure: a side overlay first, then the summary inside the card, exploring two review patterns from the research, Trustpilot’s rating breakdown and Booking.com’s quoted reviews.

Iteration 1

Side overlay

A link on the card opened a panel from the right, with the AI summary and real reviews.

Direction A visual

Iteration 2

Summary + ratings on the side

A full breakdown exposes every rating, so red 1–2★ bars would sit right next to the CTA.

Iteration 1 visual

Iteration 3

CHOSEN

Summary + review carousel on the side

The Trustpilot API lets us filter which reviews appear, but not the rating distribution. Positive quotes add a human voice.

Iteration 2 visual

Outcome

On desktop: open by default, with real reviews beside it. The quotes beside it are filtered to positive ones. That was a trade-off between transparency and conversion, so the AI label and the Trustpilot source stay visible on the card.

Decision: the card

With the carousel chosen, the first decision was what the card shows.

On the card: rating line or no rating line

Iteration 4 swapped the breakdown for a single rating line.

Iteration 4

Feature card

✕ The rating line wasn’t always positive

Iteration 5

CHOSEN

Hero mobile

✓ No rating line, labeled as AI

Decision

No rating line on either screen. The labeled summary carries the message.

trade-off

Less detail at a glance, but nothing on the card works against the click. The partner’s score is already on the card.

Decision: mobile default

With the card settled, the last decision was the mobile default. About 70% of pet food traffic was mobile (Google Analytics), where the desktop pattern took too much space.

On mobile: open or closed

An open summary takes the whole mobile screen.

Open

Stage

✕ Only the top partner fits above the fold

Closed

CHOSEN

Stage

✓ The next partner stays above the fold

vs

Decision

Closed on mobile. Expanding rows had outperformed in past tests on our other verticals.

trade-off

Lower summary visibility, in exchange for keeping partner 2 above the fold.

Final design

Built from the card’s existing row and components, with every state and content limit specified for content and engineering.

AB visual

Specs

Mobile

Preview up to 187 characters, full summary up to 379.

Desktop

Summary up to 416 characters, each review up to 307.

States

Closed, open with “See more”, open with “See less”.

Beyond pet food

Built as one component in our shared design system, so any new vertical, on Cappsool or Better Impression, needs only theme tokens and summary content.

See the system behind it: Design System: Color Architecture

Outcome

The design was ready for handoff, but repeated scope changes and a full test queue kept it from reaching its A/B test.

In demand

Product managers across Natural Intelligence asked to bring it to their own products.

The intended journey

Expected journey

What I learned

Design mobile-first

Most traffic was mobile, yet we designed desktop-first. Starting with mobile would have shown earlier what had to fit above the fold.

Agree on scope early

Next time I’d agree with the PM on a fixed scope and a test date first.

Let’s talk.

Open to senior product design roles. Based in Tel Aviv, working in English, Hebrew and Spanish.

sharon.labaton@gmail.com

LinkedIn

CV (PDF)

Earning Trust in AI Summaries

Project details

COMPANY

Cappsool

TIMEFRAME

1 month · Sep 2025

ROLE

Senior Product Designer

TEAM

PM, designer, dev

On Cappsool’s comparison sites, I led the research, design and handoff specs for an AI summary of real Trustpilot reviews on the top partner’s card, replacing a bare rating with real customer sentiment, without compromising the business model.

Result: A handoff-ready design for desktop and mobile. It didn’t reach its A/B test: shifting priorities and repeated scope changes kept deprioritizing it.

AI summary of Trustpilot reviews on a mobile partner card

The problem

Our top partner underperformed on conversion in pet food, a category where buyers rely heavily on reviews. The card offered only a Trustpilot badge: users clicked through, but few converted into leads.

The business signal

rail

01

Visit the ranking

Users compare providers

rail

02

Click the top partner

Our most valuable partner

Users drop off here

rail

03

Convert

Low conversion rate

A lead or sale on the partner’s site

Why pet food buyers dropped off · our hypothesis

01

Users left to research elsewhere

Trustpilot, Reddit and AI tools gave pet owners real opinions in seconds.

02

A number, not a reason to trust

Our card showed a score and a count, not the voice of the customer.

The card today

Card today · panel

The brief

One scannable overview of what customers say, on the top partner’s card: overall sentiment, recurring themes and the general impression.

Design goals

Real sentiment, fair ranking

1

2

3

AI text people trust

✦ Based on reviews created with AI

Few card changes

AI Summary of ★ Reviews ⌃

The challenge

Add AI that builds trust, without losing what drives clicks to our partners.

Real reviews wouldn’t always support our ranking.

User trust

and

Business model

Research

User testing on our comparison sites and a review of Gemini, Booking.com and Trustpilot pointed to one principle: people trust real users most.

Benchmark examples
Benchmark examples (AI)

Explorations

I started with a side overlay, then moved the summary into the card and explored two review patterns: a rating breakdown and quoted reviews.

Iteration 3

CHOSEN

Summary + review carousel on the side

The Trustpilot API lets us filter which reviews appear, but not the rating distribution.

Iteration 2 visual

✓ A real review, with name and stars

Positive quotes add a human voice.

Iteration 1

Side overlay

A link on the card opened a panel from the right, with the AI summary and real reviews.

Iteration 1 visual

✕ Pulled users off the card, away from the CTA

Iteration 2

Summary + ratings on the side

A full breakdown exposes every rating, so red 1–2★ bars would sit right next to the CTA.

Iteration 1 visual

✕ Red 1–2★ bars beside the summary

Outcome

Desktop: open by default, with real reviews beside it. The quotes are filtered to positive ones, a trade-off between transparency and conversion, so the AI label and Trustpilot source stay visible.

Decision: the card

With the carousel chosen, the first decision was what the card shows.

On the card: rating line or no rating line

Iteration 4 swapped the breakdown for a single rating line.

Decision

No rating line on either screen. The labeled summary carries the message.

trade-off

Less detail at a glance, but nothing on the card works against the click. The partner’s score is already on the card.

Decision: mobile default

With the card settled, the last decision was the mobile default. About 70% of pet food traffic was mobile (Google Analytics), where the desktop pattern took too much space.

On mobile: open or closed

An open summary takes the whole mobile screen.

Decision

Closed on mobile. Expanding rows had outperformed in past tests on our other verticals.

trade-off

Lower summary visibility, in exchange for keeping partner 2 above the fold.

Final design

Built from the card’s existing row and components, with every state and content limit specified.

Mobile before-after / Final design

Each screen, its default

On mobile it starts closed; this is after a tap

Beyond pet food

One shared design-system component: a new vertical needs only theme tokens and summary content.

See the system behind it: Design System: Color Architecture

Outcome

The design was ready for handoff, but repeated scope changes and a full test queue kept it from reaching its A/B test.

In demand

Product managers across Natural Intelligence asked to bring it to their own products.

What I learned

01

Design mobile-first

Most traffic was mobile, yet we designed desktop-first. Starting with mobile would have shown earlier what had to fit above the fold.

02

Agree on scope early

Next time I’d agree with the PM on a fixed scope and a test date first.

Let’s talk.

Open to senior product design roles. Based in Tel Aviv, working in English, Hebrew and Spanish.

Helping Visitors Decide: Tabs vs. Comparison

Image panel

Visitors from paid and organic search land on a page ranking online banking accounts, looking for the right account, fast. I researched and designed two ways to lift revenue per visitor, an interactive comparison tool and simple visual tabs; the PM A/B tested each on its own.

Result: Won on desktop on revenue per visitor (+7.5%), conversion rate and revenue per click. Live in Online Banking and Online Degrees.

COMPANY

Cappsool

TIMEFRAME

3 weeks · Dec 2025

ROLE

Senior Product Designer

TEAM

1 PM · 1 designer · 1 developer

The problem

Visitors (B2C) wanted a fast answer. Partners (B2B) wanted their product features visible, and every card had to carry legal information.

Problem statement

Goal

Get visitors to the right partner, faster.

Scope

The list only. Header, layout and legal info stayed fixed.

Success

Revenue per visitor (primary), plus clicks to partners, conversion rate and revenue per click.

Research

What visitors told us, and what leading comparison sites do.

User testing

Interviews and session recordings with US participants.

What a visitor said

“I see a lot of providers and I’m not sure which one I need. It’s clear, but it’s taking me a while to find what I’m looking for. Not sure I would purchase.”

US participant · Recorded session

vs

What we found

It wasn’t only banking: we saw the same pattern in several industries where conversion was dropping.

Visitor journey (before)

Benchmark

Forbes Advisor and NerdWallet.

Benchmark panel (v6, AI layout)

The approach

Two solutions to the same problem: one adds a step for in-depth comparison, the other narrows the list in one tap.

Complex pattern

Comparison tool

Glyph — mini table

Why

More control for the user

→

More confident choice

→

More clicks to partners

Simple pattern

Visual tabs

Glyph — tabs with icons

Why

Less to evaluate

→

Faster decision

→

More clicks and revenue

Interactions a visitor needs before reaching a partner.

4

interactions · comparison tool

1

interaction · visual tabs

Comparison tool

Stakeholders and partners wanted more visibility for product features, so visitors could select up to four partners, compare them feature by feature, then continue to the partner’s site.

For site — Comparison tool: scroll to the tool, th
cursor

Try it · Pick partners to compare

Lost

✕ Lost on desktop. Likely why: it sat below the list, where few visitors scroll, and asked visitors to pick partners before showing any value.

Visual tabs

Tabs above the list take visitors straight to their account type. It took three iterations.

Visual tabs · iterations (interactive)

Decision

Tabs above the list, one per account type, with an icon on each. Separate pages would split traffic and SEO, a dropdown hides the options, and a quiz adds a step.

Trade-off

One more element above the first card.

Final design

The chosen tabs, live on desktop. One tap filters the whole list to the accounts that fit that need. Partners keep their full cards; visitors see the relevant ones first.

Decisions panel

Outcome

Visual tabs shipped on desktop.

+7.5%

Revenue per visitor on desktop, plus +8% conversion rate and +6% revenue per click. Tested on one traffic source first, then rolled out to both paid and organic.

2 verticals

One reusable component, defined with the developer. Live in Online Banking and Online Degrees. Also tested on Web Hosting, where it didn’t win.

What I learned

Bigger brands, different results

Patterns that worked for bigger competitors didn’t always work on our pages, so each one needed its own test.

Next for mobile

I’d test a lighter layout above the fold: fewer elements before the first card, so more partners show on the first screen with less detail each.

What I’d do differently

Next time I’d test a clickable prototype of the comparison tool before building it. It took the most effort, and it lost.

Let’s talk.

Open to senior product design roles. Based in Tel Aviv, working in English, Hebrew and Spanish.

sharon.labaton@gmail.com

LinkedIn

CV (PDF)

Helping Visitors Decide: Tabs vs. Comparison

Project details

COMPANY

Cappsool

TIMEFRAME

3 weeks · Dec 2025

ROLE

Senior Product Designer

TEAM

PM, designer, dev

Visitors land on a page ranking online banking accounts, looking for the right account, fast. I researched and designed two ways to lift revenue per visitor, a comparison tool and visual tabs; the PM A/B tested each on its own.

Result: Won on desktop on revenue per visitor (+7.5%), conversion rate and revenue per click. Live in Online Banking and Online Degrees.

The problem

Visitors (B2C) wanted a fast answer. Partners (B2B) wanted their product features visible, and every card had to carry legal information.

Problem

Low conversion on the online banking page.

The right account was on the page, but visitors had to search the whole list to find it.

WHY

Frame

✕

No account-type label

Only the bullets showed what type each account was.

✕

Scattered details

Key details sat in a different place on every card.

✕

One card above the fold

On mobile, only one card showed on the first screen.

Goal

Get visitors to the right partner, faster.

Research

What visitors told us, and what leading comparison sites do.

User testing

Interviews and session recordings with US participants.

What a visitor said

“I see a lot of providers and I’m not sure which one I need. It’s clear, but it’s taking me a while to find what I’m looking for. Not sure I would purchase.”

US participant · Recorded session

quote bubble

What we found

It wasn’t only banking: we saw the same pattern in several industries where conversion was dropping.

Benchmark

Most useful for us: Forbes’ radio-button tabs, where visitors pick the account type before the list. That was the direction we wanted for our tabs.

Benchmark examples
Benchmark examples (VT)

The approach

Two solutions to the same problem: one adds a step for in-depth comparison, the other narrows the list in one tap.

Complex pattern

Comparison tool

Glyph — mini table

Simple pattern

Visual tabs

Glyph — tabs with icons

Interactions a visitor needs before reaching a partner.

4

comparison tool

1

visual tabs

OPTION A

Comparison tool

Stakeholders and partners wanted more visibility for product features, so visitors could compare up to four partners feature by feature.

Lost

Frame

✕ Lost on desktop. Likely why: it sat below the list, where few visitors scroll, and asked visitors to pick partners before showing any value.

OPTION B

Visual tabs

Tabs above the list take visitors straight to their account type. It took three iterations.

Iteration 3

CHOSEN

Boxed tabs with icons

Desktop

✓ Won on desktop

✕ Lost on mobile, in every round. Likely why: on a small screen, the tabs added one more thing to read before the list, pushed the first partner down, and were partly cut off. Most visitors didn’t use them.

Iteration 1

Lost

Pill tab bar

Desktop

✕ Lost on desktop and mobile. Likely why: the full-width pill bar read as site navigation, not as a choice about the list.

Iteration 2

Lost

Text tabs, right above the list

Desktop

✕ Lost on desktop and mobile. Likely why: small text-only tabs were easy to miss.

Decision

Tabs above the list, one per account type, with an icon on each. Separate pages would split traffic and SEO, a dropdown hides the options, and a quiz adds a step.

Trade-off

One more element above the first card.

Final design

The chosen tabs, live on desktop. One tap filters the whole list to the accounts that fit that need. Partners keep their full cards; visitors see the relevant ones first.

Tabs

✓ One tap instead of reading every card

Desktop shot (full)

Mobile: tested, didn’t ship. The existing list stayed.

Outcome

Visual tabs shipped on desktop.

+7.5%

Revenue per visitor on desktop, plus +8% conversion rate and +6% revenue per click. Tested on one traffic source first, then rolled out to both paid and organic.

2 verticals

One reusable component, defined with the developer. Live in Online Banking and Online Degrees. Also tested on Web Hosting, where it didn’t win.

What I learned

Bigger brands, different results

Patterns that worked for bigger competitors didn’t always work on our pages, so each one needed its own test.

What I’d do differently

Next time I’d test a clickable prototype of the comparison tool before building it. It took the most effort, and it lost.

Let’s talk.

Open to senior product design roles. Based in Tel Aviv, working in English, Hebrew and Spanish.