Give the productivity dashboard one job and a metric dictionary

A business productivity software reporting dashboard should help a named audience make a named decision. Build its data contract and checks first. Layout comes later.

This method is for organisations in England. It does not establish accessibility or data-protection compliance. Analysts, privacy specialists and affected users should review the real implementation.

What to take away

1. Give the page one job

Write the decision at the top of the design brief. An operations view might expose unassigned work and ageing exceptions today. A leadership page might support a monthly choice about capacity or process change. Do not force those jobs into one dense screen.

Name the owner, audience, review interval and action route. Remove any chart whose movement would not prompt a question or decision.

2. Create a metric dictionary

For each tile, record the formula, unit, source fields, eligible population, exclusions, time zone, period, refresh schedule and definition version. State whether waiting states pause a clock and how reopened items behave.

GDS guidance on setting service metrics distinguishes dashboards, reports and alerts according to purpose and urges teams to give measures context. Government KPI duties do not apply generally to private firms, but the decision-led distinction is useful.

3. Trace and test the data

Draw the route from operational event to displayed figure. Mark manual entry, automated rules, integrations, transformations and corrections. Reconcile a sample to source records.

Add automated checks for expected dates, row counts, unique identifiers, missing values and impossible event order. Stop or visibly downgrade the report when a material check fails. The Government Analysis Function's reproducible analytical pipeline strategy supports auditable calculations, version control, testing and documented manual steps. A small business can apply those principles without adopting a government-scale technical stack.

4. Sketch hierarchy before charts

Lead with the question and the few indicators needed to answer it. Put diagnostic breakdowns below. Show reporting period, extraction time and data status near the figures, not in a distant methodology note.

Use a line only for genuinely comparable time points. Use bars for categories and a table when readers need exact values. Never use colour as the sole signal. Government Analysis Function chart guidance asks for specific sources, clear statistical titles and an accessible table or text description where appropriate.

5. Explain changes on the page

Annotate launches, outages, staffing changes, backlog clearances and definition revisions. Show missing periods. An unexplained gap must not be silently joined by a trend line.

Add a concise limitation beside the affected result. If privacy controls suppress a small group or late records make the latest period incomplete, say so where the reader encounters the number.

6. Protect drill-down access

Aggregates may be suitable for a wider audience while item or employee data needs tighter permission. Give users only the detail needed for their role. Log exports and restrict sharing according to the approved design.

Do not introduce individual rankings merely because assignee fields exist. Have privacy and employment reviewers assess purpose, fairness, necessity and transparency before using workflow data to evaluate workers.

7. Run a release test

Ask a second person to reproduce two figures, identify the period and unit, locate the source, explain an annotation and complete the intended decision. Test keyboard and screen-reader journeys with competent assessors. Government guidance on testing dashboard design and accessibility recommends machine-readable downloads and warns that quality notes can be lost in automated interactive views.

Publish only after the owner accepts remaining limitations. Store the released extract, calculation version and decision note together. At the next cycle, check whether anyone used the page to act; a dashboard that attracts views but changes no work may need redesign, not more tiles.

Before you act

Common questions

What should a metric dictionary record for each tile?

It records the formula, unit, source fields, eligible population, exclusions, time zone, period, refresh schedule and definition version. It also states whether waiting states pause a clock and how reopened items behave. This gives every figure a documented meaning before layout work begins.

How should a dashboard handle a failed automated check?

Add automated checks for expected dates, row counts, unique identifiers, missing values and impossible event order. When a material check fails, stop the report or visibly downgrade it. This keeps unreliable figures from being read as normal output.

What does the release test ask a second person to do?

A second person reproduces two figures, identifies the period and unit, locates the source, explains an annotation and completes the intended decision. Keyboard and screen-reader journeys are tested with competent assessors. Publish only after the owner accepts remaining limitations.