Designers were doing research work they weren’t trained or scoped for

Most designers could no longer run foundational research.
Those who tried spent their limited bandwidth on:

• planning

• note-taking

• synthesis

• report writing

This work sat outside their role, slowed delivery, and produced inconsistent insight quality.

Vendor research was slow and unaffordable

Most teams avoided research entirely because vendors cost USD 40k–60k per study and required 4–6 months to complete.

Company expected insight quality the org could no longer support

Product and designers were forced to ship features with partial insights or none at all.

This created risk, rework, and low confidence in product decisions.

Validated, end-to-end workflow

I created a validated, end-to-end workflow that:

• automates research planning

• generates first-pass insights

• synthesizes across internal + external sources

• produces structured recommendations

• uses a clear validation model for accuracy, consistency, and relevance

It’s now available on the Grab Design internal site, used across teams, and adopted as the reference workflow for AI-assisted research.

Done with minimal engineering support

• Defined the new research workflow from scratch

• Evaluated multiple AI engines + selected the safest high-performer

• Designed the prompt architecture

• Built the insight validation model (accuracy, consistency, relevance)

• Documented the full system for the design org

• Ran pilot sessions with multiple squads

• Guided adoption across teams and design leadership circles

Problem Framing

The sprint direction collapsed after leadership shifted the mandate.
I reframed the problem around the real operational constraints:

• no researchers

• designers overloaded

• vendor research too slow/expensive

• leadership demanding insight at scale

I mapped the opportunity space and defined:
“Fast, reliable insights without specialists or vendor dependencies.”

Exploration & Testing

I benchmarked multiple AI engines and tested them directly against vendor reports.

What I evaluated

• Gemini Deep Research

• DeepSeek (discarded due to compliance)

• GPT-based tools

• Internal approved engines

How I evaluated

• Accuracy: similarity of AI outputs vs vendor findings

• Consistency: repeatability across 5 runs

• Relevance: alignment to the brief and core questions

A few key breakthroughs:

• AI accuracy only increases when the research brief is high-quality → so I built a prompt to generate better briefs.

• Not every research type is suitable → I defined guardrails for human vs AI boundaries.

• The fastest path was not building custom tools → use the internal AI platform as the MVP.

This testing became the foundation of the final workflow.

System Direction

Leadership pivoted the mandate after sprint toward a zero-engineering solution. I merged the original vision (tooling + automation) with the new constraint (workflow-first) into a unified roadmap: a practical workflow now, with full research automation as the long-term direction.

MVP Deployment

I deployed a practical, production-ready MVP:

Inputs:

• Raw research brief

• Context + known data

• Planned scope of inquiry

AI workflow:

• Generate improved research brief

• Run first-pass analysis

• Synthesize across sources

• Structure insights + recommendations

• Validated using accuracy, consistency, and relevance model

This replaced months of manual labor with a minutes-based workflow.

Adoption

The workflow spread through:

• Internal team demos

• Design camps

• Design leadership syncs

• Cross-team requests from product + business

• Pilot support for squads during live projects

It’s now published on the Grab Design internal site as the standard research workflow.

Time, cost, quality, and org-level Influence

• Research cycles cut from 4–6 months → minutes.

• Freed designers to focus on design, not research mechanics.

• Teams avoided paying USD 40k–60k per vendor study.

• Standardized evaluation through accuracy, consistency, and relevance.

• Improved trust in research outputs.

• Became the reference research workflow across teams, with multiple squads requesting to use it during pilots.

Ability to ship high-leverage tools with minimal engineering support

• Systems thinking

• Product strategy under ambiguity

• Evidence-based design

• Prompt architecture

• Workflow design

• Cross-functional influence