Zack Stewart Get in touch
Senior UX Researcher · AI-augmented research

Research that earns trust,
and gets adopted.

I don’t just surface insights. I translate them into decisions that shape what gets built, how it’s communicated, and how it performs in the market.

Read the case studies 5+ years · AI · E-Commerce · Healthcare · Fintech
Sector fluency

Same rigor, different rooms.

Pick a sector to see what I accomplished there.

Career timeline

A glance at my work.

Career timeline Earlier → later Showing all five career roles.

Helping Hands Community

UX Researcher · Contractor

Helping Hands Community (HHC) is a nonprofit product org (founders from companies like Uber/Google) that built technology so food banks could offer home delivery at scale. The platform connected partner organizations with local end users who carried out deliveries, and used Lyft and Uber APIs for rideshare logistics. My research sat at that intersection: what partners needed to integrate and run the program, and what end users needed in the app to complete deliveries and come back.

When
Role
UX Researcher · Contractor

Selected work from this role

Back to the timeline ↑

Flyhomes

UX Researcher · Real estate & mortgage

As the in-house researcher, I led work across acquisition, onboarding, home search, and AI trust, connecting user evidence to product, design, and brand decisions.

When
Role
Lead UX Researcher
Selected recommendation

What a Flyhomes product leader says

Meredith Han

“Zack is a UX super star. His customer-driven insights made an immediate impact on our entire product team. He pairs strong customer empathy with an ability to pinpoint the most important customer problems for the team to go solve.”

Meredith Han VP of Product

Selected work from this role

Back to the timeline ↑

Winston Francois

UX Researcher · MedTech, B2B SaaS, FinTech

At Winston Francois I owned end-to-end research operations for B2B SaaS, MedTech, and FinTech clients. I independently sourced, recruited, and managed 46 highly specialized participants across three verticals, then led Final Research Handoff presentations that turned raw qualitative data into personas, decision frameworks, and product recommendations for C-suite, Product, and Engineering partners.

When
Role
UX Researcher · Agency

Selected work from this role

Client
MedTech AI generative tooling startup
Study
2023-2024
Methods
International IDIs
Public case study · Client anonymized

Clinical AI Adoption

How interviews with 15 international embryologists and lab managers exposed black-box distrust, HIPAA friction, and a product that graded embryos when clinicians wanted a conversational clinical partner, then redirected the roadmap and cut sales friction by about 60%.

The client is described as a MedTech AI generative tooling company building clinical decision support for IVF and embryology workflows. Identifying brand details, product imagery, and proprietary model specifics have been removed.

Participants
15 embryologists and lab managers
Markets
EU + US clinical experts
Format
60-min semi-structured IDIs
Impact
~60% lower sales friction
Read the full case study
The challenge
Clinical AI was shipping. Trust was not.

The company needed to understand how embryologists and lab managers would adopt generative AI inside IVF workflows. Early product direction treated the model as a reactive binary embryo grader. In the field, black-box scoring, HIPAA anxiety, and slow system speed were already teaching clinicians not to rely on the tool.

The opportunity
Map the mental model before the next roadmap bet.

Rather than validate a single UI pass, I designed research that could answer a harder question: what kind of AI partner would clinical leaders actually trust with high-stakes decisions, and what product, training, and transparency changes would make adoption real.

01 / Framing
Four questions kept the study close to product strategy.
  1. 01

    How do embryologists and lab managers currently make high-stakes clinical judgments?

  2. 02

    Where does AI earn trust, and where does a black-box score actively block adoption?

  3. 03

    Which workflow, privacy, and performance failures force clinicians back into manual work?

  4. 04

    What would make clinical leaders champion the product instead of quietly working around it?

02 / Approach
Deep interviews with the people who cannot afford a wrong call.

I led 60-minute in-depth, semi-structured interviews with 15 international embryologists and lab managers across Europe and the US. The protocol mapped mental models of clinical AI, probed black-box and HIPAA concerns, and pressure-tested whether a binary grader matched how labs actually reason about embryos, risk, and accountability.

  • Specialist sample

    15 embryologists and lab managers across Europe and the US.

  • 60-minute IDIs

    Semi-structured depth interviews built for mental-model mapping, not feature polling.

  • Tier 3 AI evaluation

    Assessed generative clinical architectures for transparency, speed, and human-in-the-loop failure points.

  • Executive handoff

    Final Research Handoff presentations translated findings into personas and roadmap recommendations for C-suite, Product, and Engineering.

03 / Findings
Clinicians did not want a score. They wanted a super brain.

Across markets, experts described the ideal AI as a proactive, conversational partner that educates them and analyzes the whole clinical picture. A reactive binary grader fought that mental model and made distrust feel rational rather than resistant.

Adoption blockers surfaced in interviews
Black-box transparency gaps Core
HIPAA and data-privacy anxiety Core
Slow scores forcing manual double-checks High
Error states driving IT support tickets High

Relative intensity from qualitative synthesis across 15 interviews, not a statistical prevalence survey. The pattern was consistent enough to reframe the product bet.

Defining insight
Users expected AI to act as a proactive conversational partner and "super brain" analyzing the whole picture, not a binary embryo grader that handed down unexplained scores.

That single mental-model gap turned a feature debate into a strategy debate: keep optimizing a grader clinicians double-check by hand, or rebuild toward a holistic, predictive clinical simulator they could reason with.

Trust signal
Training was not a nice-to-have. It was the trust layer.

Clinical leaders did not prioritize software training. When the product failed to explain itself in the language of published science, adoption stalled before the UI ever got a fair trial. Trust had to be earned upstream of the interface.

~60% reduction in sales friction after aligning the tool with published science and vendor-led video training
From hidden friction to a clearer adoption path
Opaque scoreManual checkTicketScience fitChampion

Qualitative adoption path synthesized from interviews; not a percentage scale.

04 / Decisions
The research forced three product moves, not one UI tweak.

Findings were packaged for people who owned roadmap, architecture, and go-to-market, so the study could change more than a screen.

  1. 01
    Roadmap pivot

    Advised shifting from a reactive binary grader toward a holistic, predictive AI simulator aligned to clinical reasoning.

  2. 02
    Transparency and speed debt

    Aligned stakeholders on black-box gaps and slow scoring that forced manual double-checks and inflated clinician workload.

  3. 03
    Error-state cleanup

    Called out high-friction failures that pushed users into IT tickets, and recommended automation for daily administrative paperwork.

  4. 04
    Science-led enablement

    Recommended vendor-led video training tied to published scientific papers so clinical leaders could trust the tool before they ever clicked around in it.

05 / Influence
The handoff had to work for C-suite, Product, and Engineering in one room.
Personas that traveled

Actionable clinical personas that kept expert language intact without drowning non-clinical stakeholders.

Strategy narrative

A clear argument for the conversational clinical partner over the binary grader, grounded in repeated mental-model evidence.

Final Research Handoff

Executive-ready recommendations covering roadmap direction, technical debt, support deflection, and trust-building enablement.

06 / Impact
Research changed the product bet and the sales conversation.
Product movement
  • Roadmap pressure shifted toward a holistic predictive simulator instead of a binary grader.
  • Transparency, speed, and error-state debt became explicit engineering priorities.
  • Support-driving failure states were marked for automation and clearer recovery paths.
Go-to-market movement
  • Enablement was reframed around published science and vendor-led video training.
  • Sales conversations gained a trust story clinicians recognized immediately.
  • Clinical adoption stopped being treated as a UI polish problem alone.
Sales friction
~60% reduction
Experts
15 EU + US clinicians
Format
60-min depth interviews
Scope
Roadmap + trust + ops

Outcomes reflect documented study recommendations and the reported sales-friction reduction tied to science-aligned, vendor-led training. Client name and proprietary model details remain under NDA.

07 / Reflection
What I chose, what I traded off, and what I would add next.
Decisions I made
  • Recruited true domain experts instead of adjacent healthcare proxies.
  • Evaluated the AI architecture and the workflow, not only the interface.
  • Packaged findings as a strategy handoff so Product, Engineering, and GTM could move together.
What I would do differently
  • Add in-lab observation to capture double-check behavior as it happens.
  • Prototype explainability patterns with clinicians before the next architecture sprint.
  • Pair the qualitative mental models with a lightweight trust metric tracked over releases.
NDA boundary

Client identity, proprietary model details, and internal metrics beyond the approved outcomes stay private.

Method tradeoff

Depth interviews revealed mental models clearly, but could not quantify prevalence across every clinic type.

Sample boundary

Embryologists and lab managers were represented; adjacent roles such as clinic administrators were out of scope.

Adaptation strategy

When training emerged as a trust gate, recommendations expanded beyond product UX into enablement and sales motion.

Back to the timeline ↑ Back to the timeline ↑

Applause

UX Researcher · Agency

As a UX researcher at Applause, I run agency research across AI, healthcare, finance, food and beverage, gaming, entertainment, and other sectors. Each team has its own culture, pace, and product maturity, so I adjust my approach to fit the room.

When
Role
UX Researcher · Agency
  • Methodology

    Spans both quantitative and qualitative, including diary studies, heuristic evaluations, user interviews, usability studies, and surveys, supported by tools like Qualtrics, Lookback, Condens, and Google Workspace, alongside whatever else the client already uses.

  • Responsibilities

    I partner on recruitment, run the end-to-end study, lead the analysis and synthesis, meet with participants, write the reports, and lead readouts with stakeholders.

Selected recommendation

What an Applause colleague says

Tree Hines

“Zack is a highly insightful and tech-savvy UX leader known for his easygoing personality and collaborative nature. He helped me understand the Jobs To Be Done (JTBD) framework when I was tasked with applying this method in a study. He is particularly knowledgeable about AI tools and shared his expertise by delivering a presentation on AI use cases to our team as part of professional development initiatives. Zack is truly a delight to work with.”

Tree Hines UX and Consumer Insights Researcher

Selected work from this role

Client
Global mobility platform
Study
Aug–Dec 2024
Methods
Longitudinal mixed methods
Public case study · Client anonymized

Improving Driver Safety

How four months of continuous listening turned navigation friction into a focused safety and retention strategy.

The client is described as a leading global mobility and rideshare platform. Identifying brand details and product imagery have been removed.

An abstract route connecting a pickup point to a safer arrival.
Participants
40 active drivers
Cadence
2× weekly surveys
Duration
4 months
Evidence
471 responses
Read the full case study
The challenge
Driver confidence was slipping.

In mid-2024, retention signals and satisfaction scores were trending in the wrong direction, especially around navigation reliability and safety. With competition for drivers intensifying, the product team needed to understand which recurring problems were worth solving first.

The opportunity
Make a limited budget work over time.

Rather than spend the budget on one snapshot study, I designed a lightweight research pulse that stayed with active drivers across changing routes, conditions, and seasons. Each wave sharpened the next.

01 / Framing
Four questions kept the study close to decisions.
  1. 01

    Which navigation and routing issues repeatedly slow drivers down?

  2. 02

    How do inaccurate real-time traffic updates affect trip completion?

  3. 03

    Which app improvements would most improve safety and satisfaction?

  4. 04

    How do these moments influence motivation and long-term retention?

02 / Approach
Lean by design. Longitudinal by necessity.

Ten-question surveys balanced quantitative satisfaction tracking with open-ended feedback. I adjusted question sets every two weeks as themes emerged, creating depth without exhausting participants.

  • Diverse markets

    40 active drivers across multiple U.S. regions.

  • Repeated pulses

    Two short surveys each week from August through December.

  • Mixed evidence

    Satisfaction metrics paired with qualitative feedback.

  • Adaptive design

    Bi-weekly refinements followed the strongest signals.

03 / Findings
Navigation was not a convenience issue. It was an earnings issue.

Routing errors appeared in every study wave and compounded across a driver's shift, costing time, creating unsafe corrections, and reducing the number of trips they could complete.

Persistent navigation friction
Incorrect entry points 92%
Lost 3+ minutes per trip 67%
Switched navigation apps 30%

Half of drivers encountered incorrect entry points regularly; 19% lost more than five minutes per affected trip.

Defining insight
Every wrong turn created a chain reaction: less trust in the app, more attention split between tools, and fewer earning opportunities per shift.

The 30% who abandoned in-app navigation for another mapping tool exposed both a product gap and a safety risk: drivers were forced to manage competing interfaces while moving.

Seasonal signal
Safety concerns rose as daylight disappeared.

By November and December, drivers described greater difficulty identifying entrances, buildings, and drop-off points after dark. What began as navigation friction became a visibility problem.

77% requested building markers and lighting indicators for night trips
Relative frequency of safety concerns
AugSepOctNovDec

Qualitative trend across study waves; not a percentage scale.

Pattern view
Repeated pulses separated persistent pain from seasonal noise.
Qualitative theme intensity across survey waves from August through December.
Wave Navigation Safety Trip management Earnings Communication
August High Moderate Moderate Very high Moderate
September Very high Moderate High Very high High
October Very high High Very high Critical Very high
November Critical Very high High High Very high
December Critical Very high Very high High Very high

This qualitative synthesis guided follow-up questions; it is not presented as a statistical severity index.

04 / Iteration
Each month had a job to do.

The study moved from discovery to validation without waiting for a final readout. Emerging evidence changed the next question set in real time.

  1. Aug
    Discovery

    Established satisfaction baselines and surfaced navigation as the leading pain point.

  2. Sep
    Exploration

    Tracked routing-error frequency and identified alternate maps as a common workaround.

  3. Oct
    Validation

    Measured time loss and watched safety concerns rise as daylight shortened.

  4. Nov
    Deep dive

    Focused on entry points, night visibility, and detailed incorrect-routing journeys.

  5. Dec
    Synthesis

    Validated the strongest themes and prioritized feature recommendations.

05 / Influence
The roadshow kept evidence moving while the study was still live.
Bi-weekly toplines

Action-oriented trends, driver quotes, and visual metric tracking for product and engineering.

Monthly deep dives

Journey maps, month-over-month patterns, feature priorities, and cross-functional discussion.

Executive summaries

One-page business implications connecting driver needs to competition and company goals.

06 / Impact
The research changed both the roadmap and the role of research.
Product movement
  • Navigation improvements moved up the product roadmap.
  • Mapping integration reduced the need to switch tools mid-trip.
  • Clearer trip information appeared before driver acceptance.
Organizational movement
  • Additional funding was secured for focused follow-up studies.
  • Cross-functional teams aligned around driver-experience priorities.
  • Research became a recurring input to product decisions.
Support tickets
−15% navigation-related
Satisfaction
+20% navigation features
Arrival
+12% on-time rate
Retention
+8% frequent drivers

Reported post-launch outcomes from the client following the Q1 2025 driver-app update.

07 / Reflection
What I chose, what I traded off, and what I would add next.
Decisions I made
  • Used longitudinal surveys to balance budget efficiency with data richness.
  • Combined metrics and open feedback to reveal both scale and context.
  • Prioritized geographic breadth and adapted questions every two weeks.
What I would do differently
  • Add ride-alongs to observe navigation challenges in real time.
  • Use multimedia diary studies to capture in-the-moment conditions.
  • Prototype solutions with drivers and segment results by market and experience.
Resource limits

No budget for contextual inquiry; evidence relied on self-reporting.

Method tradeoff

Short surveys protected completion rates but limited depth in each wave.

Sample boundary

Active drivers were represented; drivers who had already churned were not.

Adaptation strategy

Driver quotes and repeated questions helped triangulate emerging patterns.

Client
Global sports-betting platform
Study
Applause era · 2024–2026
Methods
Unmoderated competitive UX
Public case study · Client anonymized

Competitive UX for a promotional odds feature

How an unmoderated competitive study across six global betting products surfaced the acquisition traps that kill promotional discovery—before a high-stakes feature launch.

The client is described as a global sports-betting and gaming platform. Identifying brand details, product names, and interface imagery have been removed under NDA.

Participants
14 bettors
Competitors
6 global products
Markets
UK · BR · DE · MX
Method
Screen-recorded onboarding
Read the full case study
The challenge
A promotional feature had to feel valuable—not suspicious.

Ahead of a 2026 feature launch, the product team needed to design and position a boosted-odds style promotional experience that maximized value perception without triggering user skepticism.

The opportunity
Learn from the market before shipping the bet.

Rather than invent the journey in isolation, we mapped how similar promotions already worked—and failed—across six global competitors so acquisition decisions could rest on observed behavior, not assumptions.

01 / Framing
Three questions kept the competitive lens decision-ready.
  1. 01

    How do users discover promotional betting offers during first-time onboarding?

  2. 02

    Where do naming, placement, and deposit requirements create drop-off?

  3. 03

    Which competitive patterns increase value perception without inviting skepticism?

02 / Approach
Unmoderated competitive UX, captured in the wild.

I facilitated an unmoderated competitive study in which participants recorded their screens while onboarding and attempting to discover specific betting promotions. The sample spanned multiple markets so patterns could be separated from local quirks.

  • 14 participants

    Screen-recorded onboarding and promotion-discovery tasks.

  • 6 global competitors

    Side-by-side journeys across major betting products.

  • Multi-market coverage

    Including the UK, Brazil, Germany, and Mexico.

  • Acquisition focus

    End-to-end path from entry to promotional offer comprehension.

03 / Findings
Promotions were not hard to build. They were hard to notice—and harder to trust.

Across competitors, three recurring failure modes blocked acquisition long before users could judge whether an offer was valuable.

01
Banner blindness

Promotional placements blended into the noise of sportsbook chrome. Users scrolled past offers without registering them as actionable.

02
No standardized naming

Inconsistent labels for similar promotions forced users to decode marketing language instead of recognizing value.

03
Deposit walls

Funding requirements interrupted discovery and drove severe drop-off before users could evaluate the promotion itself.

Defining insight
Value perception never got a fair test when users could not find the offer, name it, or reach it without hitting a deposit gate.

The competitive map turned three vague promo-UX concerns into a concrete acquisition sequence: notice, comprehend, then access—each step failing independently across the set.

04 / Recommendations
Two moves to streamline acquisition.

Findings translated into testable product guidance rather than a single winning competitor clone.

  1. 01
    Literal naming, A/B tested

    Recommended A/B testing literal naming conventions so users could recognize the promotion without translating marketing copy.

  2. 02
    Centralized offers hub

    Recommended consolidating promotions into a dedicated hub to reduce banner blindness and give acquisition a stable destination.

05 / Impact
Research set the launch constraints before engineering locked the flow.
Product movement
  • Positioning and discovery patterns informed the upcoming promotional feature launch.
  • Naming and hub structure became explicit test candidates rather than late polish.
  • Deposit timing was flagged as an acquisition risk, not a payment afterthought.
Research movement
  • Unmoderated screen recordings made multi-market competitive journeys comparable.
  • Stakeholders received a shared map of pitfalls across six products, not one-off opinions.
06 / Reflection
What the evidence supported—and what it did not claim.
Decisions I made
  • Used unmoderated screen recordings to capture authentic first-run promotion discovery.
  • Compared six competitors so recommendations rested on repeated patterns, not a single benchmark.
  • Focused the readout on acquisition failure modes tied to naming, placement, and deposit timing.
Boundaries of the study
  • Evidence came from task-based competitive journeys, not a live A/B test on the client product.
  • Markets included the UK, Brazil, Germany, and Mexico among the competitive set.
  • Client brand, proprietary metrics, and interface screens remain under NDA.
Method choice

Unmoderated recordings scaled multi-market coverage; they traded live probe depth for breadth.

NDA boundary

Competitor identities and client UI are withheld; patterns and recommendations are not.

Outcome scope

Documented impact is decision support for launch design—not a post-launch conversion metric.

Next evidence

Literal naming tests and hub placement are the natural in-product follow-ons.

Client
Global gaming platform
Study
Applause era · UK market
Methods
Checkout flow investigation
Public case study · Client anonymized

Validating an alternative payment path for UK checkout

A step-by-step Happy Flow investigation across web and mobile web to keep a regional payment method frictionless for a global gaming storefront.

The client is described as a global gaming and entertainment platform. Identifying brand details, payment-partner branding in screenshots, and proprietary checkout imagery have been removed under NDA.

Market
UK
Surfaces
Web + Mobile web
Focus
Alt. payment checkout
Output
Visual journey breakdown
Read the full case study
The challenge
Regional payments only work if the Happy Flow is truly happy.

The platform needed to validate usability of an alternative payment method for the UK market—confirming that purchase and funding felt coherent for regional users, not merely technically integrated.

The opportunity
Document the path before friction becomes a launch surprise.

A detailed flow investigation could surface where pre- and post-video checkout moments diverged across environments, giving stakeholders a shared artifact instead of scattered screenshots.

01 / Framing
The investigation stayed close to the purchase path.
  1. 01

    What is the step-by-step Happy Flow for purchase and funding with the alternative payment method?

  2. 02

    How does that journey differ across web and mobile web?

  3. 03

    Where do pre- and post-video interaction states change what users must do next?

02 / Approach
Flow investigation, screen by screen.

I conducted a detailed flow investigation mapping the step-by-step user journey of the purchase and funding process on both web and mobile web, including the Happy Flow checkout experience before and after video interaction across environments.

  • UK market focus

    Validation scoped to regional payment expectations.

  • Web and mobile web

    Parallel journeys so surface-specific friction could not hide.

  • Pre / post video states

    Checkout documented around the video interaction moments in the flow.

  • Visual breakdown

    End-to-end artifact of the payment process for stakeholder review.

03 / Deliverable
A shared map of the payment journey—not a slide of opinions.

The primary output was a comprehensive visual breakdown of the payment process spanning purchase and funding, built so product and regional partners could inspect the same Happy Flow.

01
End-to-end purchase path

Step-by-step documentation of the user journey through purchase and funding with the alternative payment method.

02
Cross-surface parity check

Web and mobile web flows mapped side by side to keep the UK integration coherent across environments.

03
Pre- and post-video states

Happy Flow checkout captured around video interaction so state changes were visible in the same artifact.

Defining outcome
The alternative payment integration needed to read as frictionless against regional user expectations—not only as a completed engineering hookup.

The visual breakdown gave stakeholders a single reference for whether the UK path met that bar.

04 / Impact
Validation evidence for a regulated regional checkout.
Product movement
  • Produced a comprehensive visual breakdown of the payment process for the UK alternative-payment path.
  • Supported validation that the integration stayed coherent on web and mobile web.
Research movement
  • Replaced ad-hoc screenshots with a structured Happy Flow investigation.
  • Gave regional payment validation a durable artifact for cross-functional review.
05 / Reflection
Credit boundaries and evidence boundaries.
What this case covers
  • Flow investigation and journey mapping for the UK alternative-payment Happy Flow.
  • Web and mobile web documentation of purchase and funding.
  • A visual breakdown used to assess friction against regional expectations.
What this case does not claim
  • No solo credit for adjacent backend-change decks from the same period.
  • No fabricated completion-rate or conversion metrics—none are documented for this study.
  • Client and payment-partner brand marks remain anonymized under NDA.
Method scope

Expert flow investigation and journey mapping; not a moderated usability test sample.

Market scope

UK regional validation only.

Credit scope

Limited to the checkout validation work documented under this engagement.

Artifact type

Visual journey breakdown for stakeholders; interface screens withheld.

Back to the timeline ↑
Selected work

I research the moments that decide whether a product earns trust or loses it.

Learn-Iterate-Launch

Driving Engagement at Flyhomes

Flyhomes · Former employer · 2022

62% of qualified leads abandoned the platform before completing their first search or tour, and the home search feature, Flyhomes' key experience, was only engaged with by 11% of traffic. Usability testing exposed why: almost nobody realized they could search for homes on the site. A second phase tested the rebrand and a redesigned homepage centered on search to better match buyers' mental models.

  • +60%homepage engagement
  • +30%booked calls
  • +20%activation rates
Read the full case study
Two-Phase Research

Disrupting Real Estate Search with AI

Flyhomes · Former employer · Nov 2022 – Jun 2023

Traditional search could filter price, beds, and baths, but not natural light, architectural style, or neighborhood ambiance. We turned unstructured listing language into searchable attributes and paired it with weighted buyer preferences to make every result more relevant and explainable.

  • +83%feature completion
  • +46%tour requests
  • +44%sign-ups
Read the full case study
Flyhomes AI search homepage showing a natural-language home search experience
An AI-assisted search experience built around the way homebuyers describe their lives.
  • RoleLead UX Researcher
  • TimelineNov 2022–Jun 2023
  • ScopeDiscovery through launch
  • MethodsInterviews · Testing · A/B

Overview

A home search should understand more than a spreadsheet can.

Finding the right home is about lifestyle, ambiance, and personal preferences, signals traditional search tools often miss. In November 2022, I led the research and contributed to the implementation of an AI-powered Flyhomes search feature designed to bridge that gap.

We used natural language processing to turn unstructured listing descriptions into structured home attributes, then paired them with a personalized match score. The experience helped buyers surface relevant homes faster and understand why each one matched what mattered to them.

The work was featured by Business Wire and USA Today.

The problem

The details buyers cared about were hidden in paragraphs.

Most real estate websites prioritize structured data, price, beds, baths, and square footage, while overlooking preferences like natural light, architectural style, neighborhood ambiance, or proximity to grocery stores.

Buyers had to read every property description manually, making search feel like trying to spot the right tree in a dense forest. Paid social campaigns were driving traffic, but only a fraction of visitors found homes they loved or booked a tour, the critical next step in engaging with Flyhomes.

The solution

Three connected moments made search personal, fast, and clear.
01
Personalize the experience

Buyers select the home features they care about and set how important each one is, creating a profile with room for both must-haves and nice-to-haves.

View animated preference setup
02
Save homebuyers time

Listings are ranked by a personalized match score, reducing the number of irrelevant homes buyers need to review without hiding lower-scoring options.

View animated match scores
03
Make every listing easier to scan

Feature tags appear directly on listing detail pages so buyers can quickly understand which parts of a home align with their preferences.

View animated listing details

Approach & timeline

Two research phases carried the idea from discovery to launch.
  1. Nov ’22User interviews
  2. Dec ’22Listing analysis
  3. Jan ’23Journey mapping
  4. Feb ’23Questionnaire design
  5. Apr ’23Card sorts
  6. May ’23Usability tests
  7. Jun ’23A/B testing
01

Started Nov 2022 · 3 months

Discovery & design

Identify the preferences missing from traditional search, then build a model that could connect those needs to the language in real listings.

01
User interviews

I interviewed prospective homebuyers and current customers to understand which meaningful attributes typical filters missed. Participants repeatedly cited grocery access, neighborhood ambiance, architectural character, and the feeling of natural light.

02
Listing analysis

We analyzed more than 10,000 property listings, focusing on unstructured agent remarks. This showed how frequently the attributes buyers valued appeared in real inventory and what the product could reliably support.

03
Questionnaire design

I synthesized the evidence into a preference flow that balanced high-coverage attributes such as proximity to schools and modern style with lower-coverage, high-value needs such as natural light and nearby restaurants.

Journey mapping

Mapping the end-to-end experience revealed where personalization could remove effort without taking control away from buyers.

Journey map of the Flyhomes AI search experience from preference setup to listing review

Our hypothesis

Structured preferences would make relevant homes surface faster.
  • Higher engagementRelevant results would give buyers a reason to keep exploring.
  • More tour requestsA tailored shortlist would make the next step feel more confident.
  • Better lead generationA personalized entry point would convert more visitors into leads.

From raw text to relevant homes

The product connected listing intelligence to buyer intent.
Input · Listing data
AI-processed listings

ChatGPT and Scale AI parsed agent remarks for attributes hidden in free-form text. A dedicated back-end service converted phrases into fields such as “lots of natural light” or “big backyard.”

Output · Personal ranking
Match score algorithm

Questionnaire responses became a weighted preference profile. The system scored every property and prioritized top matches without excluding lower-scoring homes, reducing fear of missing out.

02

Started Apr 2023 · 2 months

Iteration & launch

Test how people interpreted the experience, refine the information architecture, and validate that personalization improved real product behavior.

01
Usability testing

I ran three rounds with five users per round. Iteration between rounds simplified question wording and added tooltips for terms such as “open concept,” improving clarity without slowing the flow.

02
Card sorting

I conducted card sorts with 10 homebuyers to organize a broad set of home attributes around user mental models rather than internal real estate terminology.

03
A/B testing

We compared the AI-based search with the existing experience, measuring completion, satisfaction, sign-ups, and in-person tour requests.

Key insights

Buyers wanted personalization without losing control.
Research feedback showing buyers interpreted home tags as restrictive filters
Insight 01 Tags looked like filters.

Participants worried that preferences would hide listings because their established filter mental model implied strict exclusion.

Research feedback showing buyers distinguish must-have and nice-to-have home features
Insight 02 Not every preference is equal.

Buyers naturally separated must-have fundamentals from nice-to-have features, but a binary interaction could not capture that nuance.

Iterative refinements

We changed the model from selection to prioritization.

Multiple design reviews and testing rounds simplified the questions, expanded the attribute set, and made the results easier to understand without overwhelming buyers.

  • Rebalanced match-score weighting so nice-to-haves would not overshadow location and price.
  • Compared completion, satisfaction, and tour-request behavior with the existing search.
  • Replaced binary toggles with importance weighting for every preference.
Iteration diagram showing how the AI home preference experience evolved through research
Design decisions were traced directly to observed buyer behavior.
Final validation

Participants reported that the refined experience could save significant time, the outcome the team set out to create.

Positive participant feedback from final testing of the Flyhomes AI search experience
How the solution works together

Listing details and user-defined priorities combine into a match score for every home. The score ranks recommendations while visible tags explain why each result is relevant.

Diagram showing listing data and buyer preferences combining into personalized match scores

The impact

AI made search feel more human.

By making subjective preferences searchable and showing buyers why a home matched, Flyhomes reduced search effort and helped more people take the next step.

  • 83%increase in feature completion
  • 46%increase in tour requests
  • 44%increase in sign-ups

Clinical AI Adoption, International Research

MedTech AI generative tooling · Anonymized · Winston Francois · 2023 – 2024

Fifteen international embryologists and lab managers did not want a binary embryo grader. They wanted a proactive clinical partner that could explain itself. Black-box scores, HIPAA anxiety, and slow system speed were quietly teaching labs not to trust the product. The research reframed the roadmap and cut sales friction by about 60%.

  • ~60%sales friction reduction
  • 15EU + US clinical experts
  • 60-mindepth interviews each
Read the full case study

The shock

The product was optimized to grade. The users were optimized to reason. Across Europe and the US, embryologists and lab managers described the AI they would actually trust as a conversational "super brain" that educates them and analyzes the whole clinical picture. A reactive binary score fought that mental model and made distrust feel like good clinical judgment.

How we got here

A MedTech AI generative tooling company needed to understand clinical adoption barriers for IVF and embryology workflows. As lead researcher through Winston Francois, I owned the study end to end: specialist recruitment, 60-minute semi-structured interviews with 15 international experts, synthesis, and a Final Research Handoff for C-suite, Product, and Engineering partners.

The investigation

The protocol mapped mental models of clinical AI, then pressure- tested black-box transparency, HIPAA and data privacy concerns, system speed, error states, and training expectations. This was not a preference test on screens. It was an evaluation of whether the product architecture matched the way labs make high-stakes decisions.

What kept blocking adoption: unexplained scores, privacy anxiety, slow results that forced manual double-checks, high-friction error states that created IT tickets, and an enablement gap. Clinical leaders did not prioritize software training, so the product had to earn trust in the language of published science.

What we found

1. Partner, not grader. Experts expected AI to act as a proactive conversational partner. The binary grader framed the product as a judge instead of a collaborator.

2. Transparency is a clinical requirement. Black-box algorithmic opacity and HIPAA concerns were not edge- case objections. They were active adoption blockers.

3. Speed debt becomes workload debt. Poor system speed forced users to manually double-check AI scores, increasing labor instead of reducing it.

4. Trust starts before the UI. Aligning the tool with published scientific papers and offering vendor-led video training created immediate credibility with clinical leaders.

What changed

I advised pivoting the product roadmap toward a holistic, predictive AI simulator, made transparency and performance debt explicit for Engineering, recommended automating paperwork-heavy error recovery, and reframed enablement around science-backed, vendor-led training. The documented result: about a 60% reduction in sales friction once the trust story matched how clinicians actually evaluate tools.

Open the full Winston Francois case study in the career timeline for the complete framing, method, findings, and reflection.

FinTech Rewards, A Measurable Usability Jump

FinTech · Anonymized

A user opens the rewards screen, taps "redeem," and bounces. That single abandoned flow hid a usability problem worth millions. We benchmarked it, redesigned against the data, and proved the lift with the same numbers leadership trusted.

  • 58 → 81System Usability Scale
  • +53%task completion
  • −47%redemption drop-off
Read the full case study

The moment

We watched the bounce happen live, a real customer mid-redemption, giving up three taps from success. Starting there, instead of a clean benchmark intro, made the business cost viscerally clear to stakeholders within the first minute.

What we did

A baseline usability benchmark (SUS + task analytics), a heuristic teardown, and three rounds of moderated testing on redesigned flows. Every change was validated against the same metrics the baseline established.

What changed

SUS climbed from 58 to 81, task completion rose 53%, and redemption drop-off fell 47%. Because we measured before and after with the same yardstick, the redesign was unkillable in review, the numbers did the persuading.

Why Volunteers Deliver Once and Disappear

Helping Hands Community · Nonprofit contract · Jul–Dec 2021

During COVID, Helping Hands Community — a tech nonprofit founded by ex-Google and Lyft leaders — mobilized volunteers to home-deliver food boxes to at-risk neighbors. Deliveries were happening, but volunteers were quietly churning after a single shift. As contract UX researcher (and part-time designer), I stood up the org's first formal research practice, prioritized the work by business and user impact, and treated retention as a journey problem. Interviews, mobile-first usability tests, service blueprints, and synthesis produced six friction themes, POV statements, a future-state volunteer flow, and 18 concrete recommendations the team could ship against.

  • 6friction themes uncovered
  • 18recommendations delivered
  • 50%+volunteers arriving via mobile
  • 1stformal research practice at HHC
Read the full case study

The call to adventure

Home delivery was the program's hardest-working channel and its leakiest. People signed up with good intentions, completed one delivery run, and never came back. Nobody knew exactly why. Recruiting harder would not fix a broken experience — so I treated retention as a journey problem, not a headcount problem, and went straight to the volunteers themselves.

Helping Hands sat at a B2C and B2B intersection: at-risk neighbors receiving food, home-delivery volunteers, partner organizations, and internal ops. My first move was not a single study — it was a research portfolio. I listed every plausible research project in a spreadsheet and scored each on business impact, user-experience impact, resources, and time, then focused on the highest-impact, lowest-waste work. For the volunteer retention study I wrote a full research plan: discussion guide, screeners, process outline, and participant criteria — the first formal research plan the org had run.

Role and constraints

  • Role Lead UX Researcher · part-time designer
  • Timeline Jul – Dec 2021
  • Scope Volunteer journey, onboarding, retention
  • Methods IDIs · remote usability · service blueprints · affinity · 5 Whys
  • Constraint No prior research baseline; lean nonprofit resources

I worked directly with stakeholders from a founding team with Google and Lyft product DNA, which meant high expectations for rigor and a real appetite for journey-level thinking — including problems that lived off the screen.

Trials on the path · how the research ran

Finding the right people was half the work. Some segments were narrow — especially home-delivery volunteers. I recruited through Facebook groups and Reddit, then ran a second filter most screeners miss: a short pre-interview call before any session was booked. That kept the sample honest when the segment was scarce.

Remote usability tests on volunteer sign-up. Analytics showed more than half of volunteers arrived on mobile, so every onboarding test ran on both desktop and mobile. I focused on two segments pulled from prior volunteer data — people aged 18–25 and people 55+ — because those cohorts represented different device habits, physical constraints, and expectations. Sessions were exploratory rather than timed: the goal was conceptual models and failure points, not stopwatch optimization.

Semi-structured interviews, recorded with consent. With no research partner in the room, I recorded Zoom sessions so I could stay present in the conversation and take structured notes from the footage afterward. Because there was no prior research corpus, the guide stayed semi-structured on purpose — tight enough to answer the core questions, loose enough to chase surprising turns. Each interview walked the full arc from “I wonder what volunteering here is about?” to the quiet moment after the last drop-off.

Service design beyond the UI. Drawing on UCLA Extension service-design training, I mapped front-stage and back-stage work in service blueprints across journey stages — consent and rights flows, information checks, handoffs ops owned that volunteers never saw. Several of the retention leaks only showed up once the blueprint made those handoffs visible.

Service blueprint style flow showing consent steps and information handoffs in the volunteer journey
Off-the-screen process mapping — consent, handoffs, and checks that shaped the volunteer experience as much as any screen.

Revelation · what volunteers told us

Synthesis used highlighter coding on transcripts, affinity clustering to connect patterns, and 5 Whys on the sharpest issues to separate symptoms from root causes. Six friction themes held across the journey:

1. Flexibility comes first. Busy people needed control over when they volunteered, where they delivered, and how many boxes they took on. “There's just certain neighborhoods that I'm not comfortable delivering to alone.”

2. Know before you commit. Volunteers walked in blind to what the job actually required. “I don't recall being warned about how heavy the boxes are.”

3. The job needs tools. Heavy boxes, stairs, towing anxiety, and language barriers made shifts physically and mentally harder than they had to be — especially for volunteers over 55.

4. Volunteering felt lonely. “I think the biggest thing missing is a sense of camaraderie. It's great knowing it's not just you and that you're part of a group.”

5. There was no ending. Shifts ended with silence instead of closure or impact.

It was just kind of like you finish and then you drive away. No reporting back. Thought it was a bit odd, it did feel off. I was just like, oh, what should I do?

6. Routing broke trust. “The first time when I put in the link it didn't even show me the stops. And so I'm like, I'm not going to bother with this.”

Research synthesis board highlighting volunteer quotes about camaraderie and community
Synthesis made the emotional gap impossible to ignore — volunteers wanted proof they were part of something, not just a route on a map.

The return · recommendations and how insight traveled

The 18 recommendations mapped each theme to a fix the product and ops teams could act on:

Availability schedules visible to partner organizations; box counts volunteers choose themselves (3 to 10, with time warnings); a four-step explainer carousel and wiki that answer most questions before commitment; dolly-and-partner prompts at sign-up; shared community spaces for volunteers; a “done” checkout that triggers impact stories and recipient thanks; and just-in-time routing with the stop order curated in advance.

Deliverables were built for mixed audiences: POV statements, a future-state volunteer flow, service blueprints, and weekly readouts that mixed video clips, audio, and verbatim quotes so stakeholders heard volunteers in their own voice — a habit that comes straight from my film background. Early, frequent sharing kept founders and ops aligned while the study was still running, not only at the final readout.

The strongest signal in the data was the one the program had never designed for: volunteers stay when they can feel the impact of their work.

When people directly thank you, that makes everything worth it. It gave me the sense of THIS IS WHY.

What I would carry forward

Prioritize the portfolio, not just the study. Scoring possible projects on impact versus cost kept a lean nonprofit focused on the leak that mattered most.

Recruit like the segment is rare — because it is. Community channels plus a live pre-screen beat a polished screener alone when participants are specific and unpaid.

Retention is a service, not a screen. Blueprints and off-the-screen handoffs explained churn that UI-only testing would have missed — especially the missing ending after the last drop-off.

Recommendations

What collaborators say about my work.

Leaders across growth, product, strategy, and AI on turning customer insight into decisions that move teams forward.

Tree Hines

“Zack is a highly insightful and tech-savvy UX leader known for his easygoing personality and collaborative nature. He helped me understand the Jobs To Be Done (JTBD) framework when I was tasked with applying this method in a study. He is particularly knowledgeable about AI tools and shared his expertise by delivering a presentation on AI use cases to our team as part of professional development initiatives. Zack is truly a delight to work with.”

Tree Hines UX and Consumer Insights Researcher
Matthew Helfrich

“He’s the embodiment of turning insights into action and an absolute pleasure to work with. He’s helped unlock insights that’ve turned into company-wide messaging frameworks, personas, communication strategies, and more.”

Matthew Helfrich Sr. Director, Growth Marketing
Meredith Han

“Zack is a UX super star. His customer-driven insights made an immediate impact on our entire product team. He pairs strong customer empathy with an ability to pinpoint the most important customer problems for the team to go solve.”

Meredith Han VP of Product
Andria Jannetti

“Zack isn’t afraid to challenge the status quo. He’s adept at building trust within an organization, digging into the problems the team is facing & surfacing real world, tangible evidence that alter the direction of a product – for the better!”

Andria Jannetti Product Manager
Maggy Wardell

“Zack is a creative, thoughtful, empathetic, and engaging leader. Any company would be fortunate to have him on their team, and I offer my highest recommendation!”

Maggy Wardell Strategy & Operations Leader
Christopher Fryer

“Zack isn’t just good, he is shockingly good. Each time he produced results something about it went above and beyond what I expected, what was shocking was how consistently he did that. He constantly demonstrated a level of experience and expertise that I have only found in the most seasoned veterans.”

Christopher Fryer VP of Product & Brand
Elaine He

“I was always impressed by his great empathy for users, high quality user reports and dedication for the company mission. He is always passionate, curious and willing to contribute more! Zack is truly the best researcher I’ve ever worked with.”

Elaine He AI Product Manager
Method spotlight

The practice behind the studies.

Case studies are the visible output. These are the operating methods that make research findable, shareable, and scalable across a whole organization.

Resource 01 · Inclusive design

Design for more ways of thinking.

An accessible field guide for reducing cognitive load, supporting different sensory needs, and giving people more control over how they experience a product.

Read the 8 principles
Field guide 8 principles

Neurodiversity-Friendly Design Principles

Simple, practical design rules that make digital experiences calmer, clearer, and easier for more people to use.

Download the PDF guide View the print-friendly version Five-page PDF with selectable text
  1. Reduce cognitive load

    Keep it clean and simple

    • Use short, familiar words.
    • Keep each screen focused on one clear purpose.
    • Make the primary action obvious and specific.
    • Remove anything that does not help someone move forward.
  2. Build predictable patterns

    Stay consistent and organized

    • Put similar information in similar places.
    • Use size and weight to show what matters most.
    • Create clear patterns, then apply them consistently.
    • Give every element a logical, expected place.
  3. Support comfortable reading

    Optimize readability and typography

    • Choose highly readable sans-serif typefaces.
    • Avoid all caps and long passages of italic text.
    • Use comfortable type sizes, line heights, and spacing.
    • Maintain strong contrast and a clear text hierarchy.
  4. Lower sensory noise

    Design with sensory sensitivity in mind

    • Use a calm, considered color palette.
    • Avoid flashing, sudden motion, and neon backgrounds.
    • Keep sound off by default and provide clear controls.
    • Reduce competing visual and audio signals.
  5. Remove ambiguity

    Communicate clearly and directly

    • Prefer literal language over metaphors or sarcasm.
    • Give each message one clear meaning.
    • Label controls with specific actions, such as “Send message.”
    • Avoid jokes or wordplay where clarity is essential.
  6. Help people keep their bearings

    Simplify and clarify navigation

    • Keep navigation compact, visible, and predictable.
    • Label every page and destination clearly.
    • Show progress during multi-step tasks.
    • Make the current location and route back easy to find.
  7. Give people control

    Offer personalization and flexibility

    • Let people adjust text size and color schemes.
    • Offer calm and vivid display options where useful.
    • Allow optional content, motion, or sound to be hidden or paused.
    • Remember and respect each person’s preferences.
  8. Design with, not just for

    Emphasize empathy and inclusive practices

    • Learn how neurodivergent people experience your product.
    • Include neurodivergent participants in research and testing.
    • Involve diverse people throughout the design process.
    • Treat inclusion as a product requirement, not a final check.
Better by default

Designing for neurodiversity does not only remove barriers. It creates calmer, more understandable products that work better for everyone.

Resource 02 · Portfolio craft

Build a case study that earns the interview.

A practical field guide for turning thoughtful UX work into a concise, evidence-led story hiring teams can understand quickly.

Opens your browser’s print dialog.
Read the six chapters

Why case studies matter

Strong work still needs a strong argument.

  • Case studies are crucial for landing interviews in a competitive market.
  • Weak studies can actively hurt your chances.
  • Memorable storytelling connects design decisions to business goals.
  • A portfolio with focused, credible studies builds confidence in your craft.

What compelling studies do

Answer the questions behind the hiring decision.

  • Frame it. Name the problem, user impact, and business relevance.
  • Prove it. Emphasize measurable outcomes, not output volume.
  • Connect it. Link the problem, your decisions, and the outcome.
  • Edit it. Aim for a focused read of less than five minutes.
The Minto Pyramid Principle

Lead with the answer. Then earn it.

Structure the story from the most important result down to the evidence. Open each layer to see what belongs there.

01 Start with the answer Present the single most impressive result first.
  • Include the timeframe and business impact.
  • Quantify the result with a key metric.
  • Connect the result directly to your contribution.
02 Support with key arguments Offer two or three insights that explain the result.
  • Show clear cause-and-effect relationships.
  • Include specific observations and supporting metrics.
  • Turn findings into actionable insights.
03 Provide additional clarity Use evidence to validate the argument and result.
  • Include the numbers and data that matter.
  • Explain testing methods and results, including A/B test specifics.
  • Make your methodical approach easy to follow.

Show your thinking. Validate your claims.

  • Explain the why behind decisions, not only the what.
  • Highlight the tradeoffs and challenges you navigated.
  • Back claims with research, analytics, and credible testing results.
  • Include implementation details and metrics when they add context.
  • Treat UX writing quality as part of the experience.

Avoid the familiar traps.

  • Don’t bury the point in an overly long process diary.
  • Don’t force every project into the same generic structure.
  • Focus on business impact, not a catalog of tools and workshops.
  • Write for hiring managers and answer their key questions.
  • Proofread carefully; the details signal how you work.
The wider portfolio

Choose proof that matches where you’re going.

  • Show functional UX work. Prioritize apps, platforms, and meaningful product flows over static pages.
  • Early career? Demonstrate industry-style experience and sound decision-making.
  • More senior? Consider a concise deck centered on the problem, influence, and impact.
  • At every level: quantify business benefits when you can and maintain a professional online presence.

The strongest case study is not the one with the most artifacts. It’s the one that makes your judgment, and its impact, easy to understand.

Mentorship

Research gets stronger when researchers do.

I mentor UX researchers through difficult study decisions, stakeholder dynamics, and career moves, the messy middle between knowing the craft and making it matter.

  • Sharpen the work

    Frame the decision, challenge the method, and find the clearest story in the evidence.

  • Build influence

    Connect findings to what stakeholders need to understand and decide next.

  • Navigate what's next

    Make career and portfolio choices with a practical, experienced sounding board.

Looking for a researcher who helps the whole practice grow?

Let's talk
What mentees say Reviews from ADPList
About

I'm Zack, a researcher who optimizes for adoption.

I've spent the last decade researching AI products, financial tools, clinical workflows, and hospitality experiences across North America, Europe, and Asia. The thread connecting them: I don't consider a study finished until the insight has changed what a team ships.

That means building the operations around research, repositories, stakeholder rituals, and now AI-augmented synthesis, so findings survive long after the report. I work in English with global teams and lead research in plain language any stakeholder can act on.

  • Focus: AI / human-in-the-loop, international, FinTech & MedTech
  • Methods: Ethnography, contextual inquiry, usability benchmarking, co-design
  • Operations: Research repositories, ResOps, mentorship, AI synthesis pipelines
Contact

Let's talk about the question you're sitting on.

Request full portfolio access View résumé LinkedIn

Email: zackdrivesuxinsights@gmail.com

Several case studies are anonymized per NDA. Confidential details remain outside this public repository and deployment.

Beyond the desk

The Human Behind the Professional

When I'm not diving into UX research, you'll find me loving the outdoors, whether I'm rock climbing for fun or recharging in nature's calm. I'm a proud dog daddy, I'm often fueled by too much espresso, and I absolutely love a good breakfast burrito. I also perform in a local improv troupe, indulging my creative side as a cinephile with a passion for storytelling on and off the stage.

Zack holding a freshly caught Dungeness crab on a boat
Crabbing & outdoors
Zack's improv comedy troupe posing with a show poster
Improv comedy
Two dogs, a Husky mix and a Belgian Malinois, on the stairs
Dog daddy