Previous KPI
The reference value that gives the change meaning.
Selected work
Case study 03 · Performance analytics
Engineering a clear view of performance after coaching.
I led the full stack delivery of Coaching Effectiveness, defining how filters, KPI calculations, aggregation rules, API responses, and chart behaviour worked as one analytical contract.
- Role
- Senior Full Stack Engineer
- Ownership
- Technical direction across backend, analytical contracts, and interface delivery
- Decision
- How did KPI performance move after coaching?
- Stack
- React, TypeScript, ApexCharts, Python, Django Ninja, PostgreSQL
01 · The decision behind the dashboard
Coaching effectiveness required a consistent comparison contract across calculations, APIs, filters, and charts.
The experience shows observed KPI movement in context; it does not treat correlation as proof that coaching caused the change. Time context, calculation rules, and incomplete evidence remain consistent across the API and interface.
- 01
What moved?
Compare KPI performance before and after the selected coaching context while keeping both values visible.
- 02
By how much?
Express movement without hiding the baseline or confusing points with percentages.
- 03
Over which period?
Keep the comparison legible as the view moves between daily, weekly, and monthly evidence.
02 · The comparison contract
One answer needed three values and several conditions.
I defined the metric contract before designing the chart so the summary cards, series, tooltips, and backend response could not disagree about what improvement meant.
New KPI
The result for the selected comparison window.
Relative change
The percentage movement calculated from the same two values.
Improvement formula((New KPI − Previous KPI) ÷ Previous KPI) × 100
GuardrailWhen the baseline is missing or zero, the product must not manufacture a percentage. It should explain that the comparison cannot be calculated.
03 · One decision surface
The interface made context, comparison, and time grain visible together.
The reading order was deliberate: choose the coaching context, see the headline evidence, then inspect how the result changed over time.
- 01
Path
Choose the coaching journey or population being evaluated.
- 02
KPI
Choose the performance measure that defines success.
- 03
Period
Start with a useful default while keeping the comparison configurable.
Time grain and tooltip context
Keep individual dates visible
Useful for short windows where each day carries enough context to stand alone.
Tooltip14 Aug
Show the range, not only a week number
The tooltip names the start and end dates so an aggregated point remains traceable.
Tooltip11 Aug to 17 Aug
Reduce noise without losing direction
Longer windows group evidence into a stable trend while preserving the selected KPI meaning.
TooltipAugust 2026
04 · Full stack contract
The interface and service shared one decision contract.
I governed both sides of the boundary so filters, calculations, grouping, and incomplete states remained consistent from PostgreSQL to the rendered chart.
Product questionInterface responsibilityService responsibility
Which evidence is being compared?
Make Path, KPI, and period explicit, with stable defaults and clear selected state.
Validate the request and return one comparison contract for the selected context.
How should time be grouped?
Render daily, weekly, and monthly labels without changing the meaning of a point.
Aggregate at the requested grain and return the boundaries required by labels and tooltips.
How is improvement interpreted?
Show KPI value and percentage movement together, each with explicit units.
Apply the same baseline and current formula and return a clear unavailable state when the calculation is not valid.
What if evidence is incomplete?
Distinguish no result, insufficient comparison data, and a genuine zero.
Preserve missing values instead of converting absence into a misleading result.
05 · Interpretation boundaries
The dashboard remained useful because its claims stayed precise.
- 01
Observed movement is not proof of causation
The dashboard presents a before and after comparison. It does not claim that coaching alone produced the change.
- 02
Percentage movement is not a point difference
The two axes keep KPI values and relative percentage movement connected without presenting them as the same unit.
- 03
No evidence is not zero performance
An unavailable baseline, an empty result, and a failed service response remain distinct instead of collapsing into a misleading zero.
Verification focus
Testing the interpretation, not only the rendering.
- Default Path, KPI, and period behaviour
- Positive, negative, zero, and unavailable comparisons
- Daily, weekly, and monthly grouping
- Tooltip context plus loading, empty, and failed responses
06 · What the product made explicit
Leaders could inspect KPI movement without mistaking missing evidence for zero.
- Explainable movement
- the result, its baseline, and the calculation stayed visible together
- Consistent meaning
- summary cards, chart series, tooltips, and APIs followed one contract
- Flexible time context
- daily, weekly, and monthly views stayed interpretable
- Honest attribution
- the experience showed observed change without overstating what the data could prove
A chart is only as trustworthy as the comparison contract behind it.
Making units, time grain, missing evidence, and attribution explicit mattered more than choosing the visualization.