← Back to case study

How this
case study
was made.

This page documents the team, the process, and the decisions behind The Market Beneath the Model. It exists because how research gets built matters as much as what it finds.

Why build this.

I relocated to Brazil with a U.S.-formed product lens and noticed that most AI market analysis defaults to U.S./China framing. Brazil is either absent or reduced to an “emerging market” footnote.

The question wasn’t “what AI tools exist in Brazil” but “what does a product leader need to understand about this market that won’t show up in a Statista report?”

I decided to build the analysis as an interactive case study — not a slide deck, not a blog post — because the format itself demonstrates product thinking. I chose consumer and creative AI because those categories expose locally specific dynamics: culture, labor, monetization, rights, dignity, platforms.

Eight agents,
five leads,
one thesis.

I designed and managed a structured research operation with specialized roles, quality gates, and evidence standards — the same skills that run a product org.

Product Owner & CEO

Sally Kellaway

Thesis, team design, product briefs, all strategic and framing decisions, final narrative approval

Direct reports — equal level

Product Manager

PM Lead

Decision docs, scorecards, review agendas

Technical Director

Engineering Lead

Schema, build, data layer, code delivery

Data Architect

Agent 00

Entity model, types, filters, quality standards

UX Designer

Visual & Interaction

Cards, layouts, mobile, visual language

Research Team — Parallel Execution

Researcher 01 — Creative Economy
Researcher 02 — Consumer/Creator AI
Researcher 03 — Policy & Legal
Researcher 04 — Digital Behavior
Researcher 05 — Qualitative
Researcher 07 — Market Landscape

Agent 06

Research Editor & Synthesis

Turns 6 streams into one coherent case study

Quality

QA & Launch

Accessibility, performance, launch readiness

Milestones
and gates.

Objective-based milestones with quality gates. Each milestone produces a reviewable artifact before the next begins.

M1 Data Architecture & Schema

11 entity types, TypeScript interfaces, filter taxonomy, evidence classification, data layer utilities.

M2 Research Execution

6 researchers working in parallel. 381 structured records. First-round memos from each agent.

M2.5 Research Review Checkpoint

19 PM decisions resolved. Hypothesis scorecard reviewed. Claims approved or caveated.

PM Decision Point
M3

Design & Interaction

Visual language, card formats, mobile-first

M4

Explorer Build

Svelte components, filters, real data

M5

Strategic POV & Narrative

Sally’s voice, not committee output

M6

QA & Launch

Accessibility, performance, the hiring manager test

8

Specialized Agents

381

Structured Records

159

Cited Sources

19

PM Decisions

11

Strategic Claims

7

Milestones

Every claim
has a grade.

The case study uses a six-level evidence classification system. Every finding, card, and strategic claim carries a confidence level and a classification. This prevents overclaiming and makes it clear what the research actually supports.

Validated Claim — 2+ independent sources. Strong enough for the narrative.
Emerging Hypothesis — Promising, not yet validated.
Open Question — Important but not safe to assert.
Contradictory Signal — Conflicting evidence.
Product Opportunity — Buildable direction.
Risk / Constraint — Barrier or limitation.