Due diligence data ✦ 5× conversion 12 months
(01) CASE STUDY
⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
ARBOLUS · 2023 / 2024
PRODUCT / UX DESIGN

Data that buyers
actually
trust.

Arbolus is an expert insights platform. Private equity firms and consultants use it to capture attributable buyer perspectives on tech companies before, during and after a deal. I joined as one of two product designers, paired with Diego in Italy, and led design across the platform redesign, the company's first real design system, and four new product surfaces, Canopy, AI Summaries, Channels and Connect.

Lift in insight-to-decision conversion
after the platform redesign
ROLE Product / UX designer
SCOPE Platform, system, 4 new products
PLATFORM Web, responsive
YEAR 2023 / 2024
TEAM Two designers, in-office eng pairing
(02) Context

What Arbolus
actually does.

Arbolus turns conversations with real B2B buyers into structured, attributable data. Private equity firms, consultants and corporate development teams use it to validate a thesis before a deal, track companies post-investment, and monitor portfolio health. The platform spans Canopy async video Q&A, scheduled Calls, Surveys & Panels, the Channels library of pre-recorded expert answers, Connect for native expert calls, and an AI layer that summarises responses at scale.

Two audiences share the same product. Experts are the operators, executives and former buyers contributing time, often paid $200–$500 per call. Clients are the PE and consulting firms paying for the insight. They have very different mental models of what trust looks like, and the platform has to earn it from both sides.

Company snapshot
FOUNDED 2018, London / Barcelona
STAGE Series B, $20M, late 2024
CUSTOMERS PE, growth equity, consulting
EXPERT NETWORK 200k+ B2B professionals
TEAM ~150 across four global offices
DESIGN MATURITY Low at start
(01)
Two-sided product

Experts contributing buyer perspectives on one side, PE associates and consultants consuming them on the other. Different mental models, different trust signals, the same underlying database.

(02)
AI in the critical path

Generative models drive Canopy summaries, transcript search, expert matching and scoring. The design problem was making probabilistic output feel honest to a managing director who has never read a model card, and tying every claim back to an attributable quote.

(03)
Compliance is not optional

Financial services regulation, expert vetting and GDPR meant no shortcut was free. Every change to a label, attribution or flow went through compliance before it shipped.

(03) The problem

A capable product
that no one could
read.

The platform had two years of feature work behind it and no design system holding it together. Every team had shipped what they needed in the way that made sense to them. The result was a product with most of the right capabilities and very few of them visible.

Three patterns kept showing up in research. None of them were about missing features.

3
Failure modes worth fixing
before any new pixels
Failure mode 01
Too much, all at once

Customer profiles and Canopy responses surfaced over forty fields with no hierarchy. The insight a PE associate needed for a Monday partner meeting was sitting next to metadata that was never read.

Failure mode 02
Trust signals out of reach

Attribution, vetting and recency were buried three clicks deep. Sales kept having to confirm by email what the UI should have said on its own. Without those signals, partners would not quote a finding in a deal memo.

Failure mode 03
No shared design language

Twelve button styles, three modal patterns, no tokens. Engineering kept building the same things twice, and small inconsistencies were stacking up into a confidence problem.

"We're losing deals not because of the product, but because partners can't see the product in the ten minutes they have between meetings."

Sales feedback, Q1 2023
(04) Role

A design pair.
Back to back.

Two product designers, working back to back from the Barcelona office. We split the platform by surface, ran daily critique, and pair-reviewed everything that shipped. There was no system to inherit and no playbook to follow, just a backlog and a lot of customers waiting on improvements.

I owned the platform redesign, the Canopy AI surfaces, and the project dashboard, plus the design system underneath. Full cycle. Discovery, strategy, interaction and visual, handoff, post-launch validation. Plus the room time. Roadmap calls, stakeholder reviews, customer presentations.

How we worked
BCN OFFICE
CADENCE Daily critique, back to back
SPLIT By surface, paired on systems
STAKEHOLDERS Weekly reviews, sales loop
CUSTOMERS PE / consulting interviews
HANDOFF Engineering pairing, no walls
Q1 2023
Audit and discovery
Q2 2023
System foundation
Q3 2023
Platform redesign
Q4 2023
Canopy AI & Surveys
Q1 2024
Project dashboard
Q2 2024
Handoff and impact review
(05) Initiatives

Five bets.
One direction.

Rather than a list of tickets, here is how the year clustered into strategic threads. Each one started with a hypothesis and ended with something we could measure.

(01) Deep dive ↓
Platform & system
End-to-end rethink of discovery, attribution and request flow across the platform, on top of the company's first real design system, tokens, grid and component library.
↳ 5× conversion rate
(02)
Canopy
Async video Q&A that replaces 10–30 expert calls with a single launch. I shaped the interaction model for how clients author surveys, review video and text responses, and pull out the answers that move a thesis.
↳ Hours of insight per launch
(03)
AI Summaries
As Canopy scaled, qualitative volume outpaced analyst attention. I designed the UX for AI-generated summaries that synthesise dozens of video responses into one readable report, with every claim linked back to its source quote.
↳ Probabilistic output, made trustable
(04)
Channels
A searchable library of pre-recorded expert answers in video and audio. Instead of commissioning a new call, clients search what has already been asked. I designed the discovery experience and the in-page media interactions around transcripts and quotes.
↳ Insight without a new call
(05)
Connect
A native expert-calling platform built inside Arbolus, replacing the third-party video tools clients had been stitching together. I worked through the call experience, scheduling, recording and compliance, paired in-office with engineering to ship it.
↳ One product, one transcript
Deep dive 01
Platform redesign

From cluttered
to converting.

Insight discovery was the company's main revenue lever. PE associates who could pull a credible buyer perspective into a deal memo kept renewing. The ones who couldn't, churned without ever feeling the product work.

The flow was technically functional but experientially broken. Too many steps, the wrong things on the wrong screens, and attribution signals filed in places no one looked.

The research
12 USER INTERVIEWS
6 client side (PE / consulting), 6 expert side

80+ SESSION REPLAYS
Insight workflows in detail

FUNNEL ANALYSIS
Where users actually dropped off

"Trust is established before the first read. The insight page was doing the work of a CV, not a pitch."

Insight from user sessions
01
Lead with relevance, not credentials

Insight cards and customer profiles were restructured to answer "why this perspective for this thesis" before showing former titles and tenure. The drop-off was at the profile view, not at checkout. The content order was the problem.

Trade-off. The ML team had to expose matching rationale through a new API, costing two sprints of engineering. The conversion gain made it pay for itself.
02
Move attribution into context

Vetting, attribution and recency had their own page. We pulled them inline, surfacing them at the moment a client was deciding whether to quote a finding. Trust signals work best when they appear exactly when trust is being weighed.

Trade-off. Every copy variation needed legal review, which slowed iteration. The lift in trust scores justified it.
03
Cut the request flow to three steps

The original Calls and Surveys request flow had seven steps with compliance gates in the middle. The new one had three, with compliance handled async after submission. The friction was never regulatory, it was the order things happened in.

Trade-off. This was a product policy change, not just a design change. It was the hardest internal sell of the year.
Project dashboard
three views

From cluttered v0 to a dashboard the project lives on.

The project dashboard is where a deal team spends most of its time. The redesign restructured it around the next decision, surfaced attribution and recency inline on every insight, and unlocked a heat-map view that lets a partner read a thesis in under a minute. Same data, same project, three points in the work below.

Before: three competing primary actions, insight cards identical whether vetted, in-progress or expired
After: project header summarises status and the one decision that is overdue; attribution, vetting state and recency live on every card
Heat map: the same dataset, restructured for partner-level scanning, with every cell drilling back to its source quote
Conversion lift
90 day post-ship comparison
3 mo
Research to ship
Including compliance review
(06) Selected screens

Across the
full surface.

A sample from the year. Two user types, one product. Each surface had its own constraints: video on the expert side, scannability on the client side, attribution everywhere. Screens are presented as shipped, then iterated on quarterly after launch.

(07) Design system

The invisible work
that made everything
else possible.

The system was not a side project. It was a survival tool. Without it, every new screen meant deciding which of three button patterns to use and guessing which spacing scale engineering had picked that month.

It was built incrementally, alongside feature work, starting with the parts that hurt most. Type, colour and spacing tokens first, then the grid. Previous grid was undocumented and inconsistent. New grid was twelve column, eight pixel base, three breakpoints.

System scope
8 Semantic colour tokens 2 Accent ramps (purple, orange) 4 Type scales 3 Breakpoints 12 Column grid 200+ Components built ~40% Less time on handoff
bg
#f4f2ec
paper
#fbf9f4
ink
#0a0a0a
hairline
#d9d6cc
accent
#6157FC
accent-soft
#ECEAFF
accent-2
#FF6B2C
accent-2-soft
#FFE9DD
(08) Impact

The work,
measured.

Lift in conversion
Discovery to insight requested
FUNNEL ANALYTICS
90 DAY PRE / POST
~40%
Less rework on handoff
ESTIMATED FROM
SPRINT RETROS
12
User interviews run
6 CLIENT SIDE
6 EXPERT SIDE
3 mo
Research to shipped
Platform redesign
INCLUDING COMPLIANCE
REVIEW CYCLES
30+
Screens redesigned across
two user types
WEB AND RESPONSIVE
CORE AND ADMIN
NOTE. Some figures are directional because analytics maturity was limited at the start of the project. Where exact numbers are not available, ranges reflect estimates that stakeholders signed off on.
(09) Outcomes

What changed.
What it earned.

Before After
Twelve button styles, no token setOne component system, one token set
Seven step request flow with compliance gatesThree step request, compliance handled async
Raw model output, scores without explanationAI Summaries with reasoning and source quotes
Dozens of repeat calls to ask the same questionsCanopy and Channels, async by default
Third-party video stitched in for expert callsConnect, native calls with transcript and compliance
Compliance buried in documentationTrust signals woven into the flow
No design system, no shared gridEight tokens, twenty plus components, twelve column grid
No structured design processDiscovery, test, ship, measure
(10) Reflection

What I learned.

What worked
Starting with the system, not the screens

Without the grid rebuild and the first round of tokens, every screen after would have been slower and messier. The compounding value of this kind of work is consistently underrated.

Treating sales as a research channel

Customer-facing teams hear things formal research often misses. A monthly loop with sales became one of the cheapest and most reliable feedback mechanisms I had.

Making decisions legible

Framing each design call as a hypothesis with a measurable outcome changed the quality of every stakeholder conversation, and eventually changed my seat in the room.

What I would do differently
Instrument before launch, not after

Some impact claims are directional because analytics were not in place at baseline. I would push harder for instrumentation as a pre-launch requirement, not a post-launch task.

Bring legal in earlier

I treated compliance review as a gate rather than a collaborator. Earlier loops would have prevented a few late-stage redesigns and probably unlocked a couple of better patterns.

More structured A/B testing

The conversion lift is directional. With proper testing infrastructure I could have shown harder numbers, and built a stronger case for the next initiative.

A year of work and a product that finally reads as one thing.
Artur Lopez Zarytskyi · Arbolus · 2023 / 2024
Next case, Finario All work