Product growth — engineering and analytics

Huichuan Li

I ship the surface, wire the tracking into it, and own the numbers it produces.

Most growth work breaks at a handoff — the person who builds the page isn't the person who defines the event, and neither one owns the funnel. At OpGov I joined with no measurement in place and did all three. M.S. Statistics, B.S. Quantitative Economics.

Ship Production pages and components in Next.js, React, TypeScript
Instrument Event taxonomy designed and wired into the components themselves
Measure Funnel and path analysis, cohorts, controlled tests
Iterate Ship the winners, write up the losers, change the definition when it's wrong

One person, four steps, no handoff.

Selected work

Finding where signups died

Nobody could see into the funnel, because there was no funnel to see — no event schema, no journey view, no agreed definition of what a converted user was.

I designed the event taxonomy from scratch and implemented the tracking directly in the components, then ran funnel and path analysis across 30M+ interactions to isolate the two largest drop-offs. Partnered with product on the flow redesign that followed.

Next.js · React · TypeScript · GA4 · SQL · BigQuery

43% lift in visitor-to-signup conversion, 2.1% to 3.0% over two quarters

Ad-click tracking, end to end

Sponsored placements were being sold against numbers nobody could verify, because click data never made it anywhere durable.

I built the capture flow across the whole path: the interaction handling in React components, a Redis-backed API to write the events, and the downstream tables that made the result queryable for campaign and conversion analysis.

React · REST API · Redis · SQL

35% lift in sponsored-placement CTR after reallocating toward high-intent surfaces

Spending less to get better users

Acquisition spend was spread evenly across an audience that didn't behave evenly.

I segmented users into five behavioral cohorts on geography, behavior, and engagement, measured retention by cohort, and moved spend toward the two that held. Separately ran controlled tests on placement and messaging against success metrics defined before launch — newsletter signups rose about 20%.

Cohort analysis · retention curves · A/B testing · Python · SQL

15% reduction in effective CAC, quarter over quarter

Deciding what the AI was allowed to ship

The company wanted to publish AI-generated content and had no way to judge whether it was good enough.

I defined the quality rubric, scored AI output against human-authored output against it, and the analysis set the policy for what reached production and what didn't. Separately evaluated a production RAG system on live interaction data — resolution, fallback, retrieval relevance — and traced failure patterns to specific fixes in retrieval and prompting.

Evaluation design · structured-output validation · Python

Policy the evaluation governed what shipped, not a recommendation someone could ignore

How I work

Build
Next.js App Router, React, TypeScript, JavaScript, MUI, Tailwind, Framer Motion, Clerk, JSON-LD. Production pages and reusable components — pricing and donation flows, advertiser surfaces, city pages, shared component library.
Instrument
GA4 event taxonomy design and implementation, REST APIs, Redis, metric definition and standardization, completeness and conformance checks so a number that looks wrong is wrong for a findable reason.
Measure
Funnel and path analysis, cohort and retention analysis, A/B test design and analysis, significance and proportion testing, regression and time-series forecasting.
Data
SQL across BigQuery, PostgreSQL, Oracle and SQL Server. Python with pandas and statsmodels, R, Databricks, Tableau, Power BI.
Scope
Frontend and instrumentation are where I'm strongest. Backend work has been APIs and data pipelines.

Background

2025 — now
Data Analyst, Product & Growth Analytics OpGov.ai / OpGov.News — early-stage civic information platform, San Francisco Bay Area
2024
Data Analyst, contract Boston Congress of Public Health — forecasting that caught an under-enrollment trend mid-program, and a monthly reporting cycle cut from three days to four hours
2024
Data Analyst Fellow, GIS CA Consortium for Public Health Informatics & Technology — county-level geospatial analysis for regional resource allocation
2021 — 2023
Data Analyst, Operations & Quality Analytics Gemological Institute of America — raised reporting accuracy from ~92% to 99%+, and got three teams to agree on one set of metric definitions
2025 / 2021
M.S. Statistics · B.S. Quantitative Economics CSU East Bay · San Francisco State University

Get in touch

Looking for growth roles where the same person builds the surface and owns the number. Happy to walk through any of the work above.

huichuan.li.ds@gmail.com LinkedIn