正在加载页面 · Loading page
Data visualisation · Shanghai
Every chart is making a decision for you.
Four years in product analytics and roughly two hundred charts. The useful ones share a trait: people finish reading them and say "so we should" rather than "wow".
By the numbers
Four years, the countable part
Own audit, 2022–2025
About 30% of the charts I built, I later deleted myself
A chart nobody has ever acted on still spends the attention of everyone who reads it. Deleting it is part of the job.
4
Years in analytics
31
Dashboards shipped
9
Dashboards later deleted
Work
Three dashboards
Three examples. The real ones live inside other people's BI tools, so these are descriptions only.
Growth · 2026
Retention, reduced to one line
Eleven metrics nobody read, replaced by a single weekly cohort curve. The first time anyone argued about it in a weekly meeting counted as success.
View project →
Product · 2025
A funnel as one sentence
Anyone can draw a four-step funnel; the hard part is saying that each drop-off is relative to the step before. Saying it revealed two of the steps were defined wrong.
View project →
Method · 2024
A chart that admits ignorance
Segments below the sample threshold render grey and labelled "too little data", which is far more honest than a confident-looking line.
View project →
A chart nobody acts on is noise.
From a weekly note
Tools
What I use
The tools matter least, but people always ask.
Query
- SQL
- dbt
- Python
- pandas
Chart
- D3
- Observable
- Vega-Lite
- ECharts
Ship
- Metabase
- Superset
- Notion