The Differentiator
Two ways to run product.
I chose the faster one.
Same lifecycle, same rigor — a different engine underneath. Toggle to compare, then see a real pipeline below.
6–8 weeks / cycle
Discovery to launch readiness, run manually
↓ Roughly 80% cycle-time compression · same quality bar
Discovery & ResearchMarket, users, competitors
~2 weeks
Claude ProjectsPerplexityDeep Research
Synthesis & InsightsInterviews, feedback, signals
~1 week
LLM synthesis promptsMixpanelSQL copilots
Specs & PRDsRequirements, stories, acceptance criteria
4–5 days
ClaudeCustom prompt systemsNotion AI
Prototyping & ValidationFlows, mocks, usability signals
1–2 weeks
Figma AIClaude ArtifactsAI prototyping
Experimentation & AnalysisA/B design, stats, readouts
3–4 days
SQL + LLM analysisOptimizelyLooker
GTM & Stakeholder CommsNarratives, decks, launch docs
~3 days
ClaudeAI deck generationLoom
Timings reflect my typical cycles and vary with scope — the compression ratio is the point, not the precision.
A real pipeline: how my discovery cycle actually runs
Not a claim — a system. Inputs go in, a prompt architecture I've iterated through three versions does the heavy lifting, and decision-ready artifacts come out.
Inputs
- Customer interviews & support tickets
- Mixpanel funnels & SQL extracts
- Competitor & market signals
- Stakeholder context docs
→
AI System
- Claude Project with structured research instructions
- Multi-step synthesis prompts — clustering pain points, scoring severity × frequency
- LLM-assisted SQL to validate patterns against real usage data
- Every output cited back to source — auditable, not hallucinated
→
Outputs
- Opportunity map, ranked by evidence
- Scored backlog candidates
- PRD first draft with acceptance criteria
- Open questions for human judgment
The same engineering shipped to production: a RAG-based knowledge system I led at Accenture cut information-retrieval time and lifted support productivity 22% — validated by A/B test, not anecdote.