Dave Grant - UX Researcher

About

Insights only matter if they change what gets built.

I'm a senior UX researcher with 15 years in product, 10 of them leading research end to end, the last six at Meta across SMB, business platforms, and AI products. Before that I worked as a designer and design strategist on USPS, American Express, Coca-Cola, and Harley-Davidson.

I work across the full methods stack, from generative discovery through evaluative testing to AI-assisted approaches. At Meta I built a synthetic research platform grounded in real interview data, which was adopted into central research infrastructure. Coming to research from design taught me that a finding nobody acts on is a finding that didn't happen.

I measure success by the decisions my research enables, not the studies I complete.

Methods

Foundational: Interviews, stakeholder interviews, diary studies, field observation, contextual inquiry
Evaluative: Usability testing, concept testing, heuristic review, survey design, intercept studies
Strategic: Journey mapping, persona development, segmentation, information architecture, workshops
Quantitative: Behavioral funnel analysis, telemetry, A/B and multivariant testing, card sorting
Emerging: AI-grounded qualitative, synthetic research, AI trust and adoption psychology

Meta · 2024–2026

Synthetic user research panel

Creative velocity had outpaced research capacity. Teams were shipping work that had never been tested, because recruiting took longer than the production cycle allowed. So I built a panel that could answer in minutes.

Role: End to end. Research design, recruitment, instrument, tool design, prompting, cross-org partnership.
Method: Digital twin modeling, prompt engineering, bilateral validation against a live human panel
Sample: 400+ applicants screened to 100 vetted SMB advertiser profiles
Outcome: Adopted by 24 people across four orgs, 28 tests run, selected as the SMB layer of Meta's central synthetic platform

100 SMB advertiser personas built one to one from real people, screened from 400 plus applicants — Every persona traces back to a real participant's words, context, and constraints. Not survey data, not scraping, not composite archetypes.

How the panel works: browse and filter personas, run a test, review results — Browse and filter the panel, upload creative and pick personas, then read aggregate themes alongside individual responses.

Bilateral test against a live researcher panel reached about 75 percent theme alignment — Run head to head against a live panel of human SMB advertisers, the synthetic panel reached roughly 75% theme alignment and picked the same winning ad.

Extending the panel to non-advertisers, 95 percent of the SMB audience — 95% of the SMB audience were non-advertisers with no personas and no qualitative base. I synthesized nine research streams into a citation-traceable persona system.

Most persona work dies in a deck. This one was built to live in the tool where the briefing actually happens.

Tradeoffs: emotional resonance, persona drift, and stakeholder over-trust — The limitations, stated up front. Synthetic misses emotional resonance, personas drift without re-grounding, and stakeholder over-trust is the real risk once a tool feels authoritative.

Meta · March 2026

AI-first learning: what ad comes next?

For years, advertisers with a problem went to search, help centers, or forums. In about eighteen months that shifted to AI assistants, across the board. The platform was still introducing itself as the first stop.

Role: End to end. Research design, recruitment, instrument, moderation, synthesis, stakeholder readout.
Method: Unmoderated diary study. Screen recordings with verbal narration, typed responses, hand-drawn sketches.
Sample: 19 US-based SMB owners and marketers
Outcome: Five-mindset framework adopted as a creative briefing lens; elevated to three leadership teams

Where advertisers used to go for help versus where they go now — The shift happened fast, in roughly eighteen months.

What the platform assumes versus what is actually true — By the time a business owner sees an ad, the diagnostic conversation already happened somewhere else.

A three step diary study: ask a real AI question, narrate the interaction, sketch the ideal ad — Ask a real question of an AI tool, narrate the interaction live on screen recording, then hand-sketch the ad that would feel genuinely useful right afterward.

Five post-AI states of mind and what to show each one next — Everyone felt something different after using AI, and that feeling determined which message could land next. Five distinct states, each with its own creative direction.

A participant's AI prompt, the wall he hit, and the ad he sketched — The advice was right. The platform know-how wasn't. AI told him to check his metrics; he had no idea how to pull that view.

Tradeoffs: sample skew, pattern versus prevalence, and market coverage — Diary studies skew toward the engaged, 19 participants finds patterns rather than percentages, and the sample was US-only solo and small-team operators.

Meta · April 2026

Three ads, one landing page

Direct response ads with strong click-through were underperforming at the install step. The ads weren't failing and the page wasn't failing. They were misaligned.

Role: End to end. Research design, recruitment, instrument, synthesis, stakeholder readout.
Method: Between-subjects unmoderated study. Three ads, one universal landing page.
Sample: 16 US-based iPhone users
Outcome: Five prioritized fixes split across the page and creative teams, plus a four-experiment test plan

Three ad conditions tested: privacy, connection, and productivity — Three ads, three value propositions, one shared destination.

Study structure: watch one ad, review the page, decide whether to install — Between-subjects, so each participant saw only one ad before the same page.

Four findings on why ads create interest the landing page cannot close — Feature-specific ads created expectations a universal page could not meet, and the question everyone asked went unanswered.

Download intent by ad condition — Positive install intent fell from 80% in the strongest condition to 50% in the one whose hero feature was missing from the page.

How the landing page held up after each of the three ads — Mixed, thematic but cold, and a broken promise. Participants actively searched for the feature the ad had shown them and could not find it.

Tradeoffs: reaction versus reasoning, sample size, and platform coverage — Unmoderated captures reaction rather than reasoning, n=16 across three conditions shows direction rather than significance, and the study was iOS only by design.

More work

Other research I've led

User psychology of AI adoption

Cross-cultural qualitative work in India and the US on how tone, voice, and personality shape whether an AI agent earns trust or reads as a downgrade.

2025

Business AI creative testing

India-market message testing that produced a reusable creative formula: pair operational relief with a concrete growth outcome, and avoid control framing.

2026

Global navigation and IA

Content audit, open card sorts with agency partners, and task-based testing that produced a new site taxonomy and retired 40+ redundant pages.

2022

SMB email performance strategy

Four-day diary study with 17 small business owners across four email concepts, producing a reader segmentation model and seven optimization principles.

2025

Design system typography

Comparative readability and legibility evaluation of a custom brand typeface against system fonts, confirming it met accessibility standards with no cost to reading speed.

2024

Brand perception baseline

Moderated live walkthroughs to baseline how a consultative service was understood ahead of a refresh. Trust landed fast; the offer was hard to describe back.

2026

Full decks, methodology, and detailed findings available on request. Happy to walk through any of this in conversation.