Studio floor, Billingford. We read pipelines before we decorate dashboards.
Dev Flowgrid · Norfolk teaching studio
A KPI is a contract, not a chart.
Most teams already have a stream. What they lack is a written agreement about freshness, grain, and who is allowed to change the definition at 11pm. We teach real-time app KPI tracking as that agreement: slow enough to be honest, live enough to be useful.
How a week actually runs
We begin with the join that nobody named
01 · Monday
Bring a live metric that already embarrasses you
Not a greenfield design. You arrive with a number the board already watches. We reconstruct its grain, late events, and the last unreviewed change to the SQL.
02 · Midweek
Write the freshness window in English
Streaming is not a personality trait. We force a sentence: “this tile is trustworthy if events older than N minutes are marked stale.” That sentence becomes the owner’s job, not the vendor’s.
03 · Clinic
Assign an on-call for definition drift
Alerts fire for volume. Definitions drift in silence. The clinic ends when someone named on a rota can explain why yesterday’s conversion is not comparable to last Tuesday’s.
Flagship
KPI Grid Clinic
Eight weeks · live markup sessions · fee listed on the syllabus
A working programme for product, data, and mobile engineers who already emit events and still argue about whether “active” means a heartbeat, a foreground session, or a paid receipt. You leave with a grid: primary tiles, guardrails, and a written lag budget.
From a recent seat
“The clinic would not let us keep ‘real-time’ in the tile title until we published the lag histogram. That felt pedantic in week two. By week five the board stopped asking why the number jumped at 09:00.”
Rhys P., analytics lead, grocery delivery · KPI Grid Clinic, spring 2026 · Billingford remote seat
Field notes
Recent writing
12 May 2026
Lag windows are not latency
Teams collapse ingest delay, processing time, and the human decision to wait. The tile then lies with perfect confidence.
3 April 2026
The KPI that moved because of a join
A left join to a slowly changing user table shifted conversion by 6 points. Nobody shipped a feature that week.