01 · Start with decisions rather than tiles
Start with decisions rather than tiles
A dashboard brings selected measures together for an audience that needs to make related decisions. Its purpose is to direct attention, provide context and connect evidence to action. Begin with the decisions the audience owns. A dispatch lead deciding where to place support needs different information from a sponsor deciding whether the improvement’s benefits have persisted. Putting every available metric on one screen usually weakens both decisions.
Define the customer outcome, its process drivers and the constraints that must be protected. Outcome measures often lag the work, such as wrong-service dispatches or missed collection commitments. Driver measures can provide earlier warning, such as unverified parcels approaching the collection cutoff. A leading indicator is useful because of a plausible relationship to an outcome, not because it updates frequently. Test that relationship and avoid treating it as proof of future performance.
Choose a manageable set covering quality, delivery, safety, workload and cost where relevant. Do not optimise dispatch speed without seeing errors and employee burden. A balanced dashboard makes tradeoffs visible; it does not require every category to have the same number of metrics.
02 · Build a measure definition before a display
Build a measure definition before a display
Specify each measure’s numerator, denominator, unit, population, timing, source and owner. “On-time dispatch” might mean scanned before the internal cutoff, loaded before vehicle departure or delivered by the customer promise. Those definitions are not interchangeable. Decide which event corresponds to the intended decision and retain the other events where they provide explanatory context.
Define exclusions in advance. If cancelled orders are excluded from on-time performance, describe when cancellation becomes effective and retain the excluded count. Prevent late orders being reclassified after the fact to improve the percentage. Document the observation window and treatment of open orders. A day’s incomplete records should not silently appear as a final green score.
Check data lineage from the work event to the dashboard. Verify identifiers, timestamps, duplicate handling and joins between systems. If collection confirmation arrives the next morning, show the measure as provisional with its last refresh time. A highly automated dashboard can distribute an incorrect definition more efficiently; automation does not remove the need to reconcile a sample against the underlying work.
03 · Worked example: denominators change the story
Worked example: denominators change the story
Fictional Northbank dispatches 980 of 1,000 eligible orders on time on Monday, yielding 98%. Tuesday records 190 of 200, yielding 95%. The percentage is lower, but the number late falls from 20 to 10. Neither statement alone is enough to judge the operation. Tuesday’s smaller volume may reflect a planned demand change, while a different carrier mix may explain the lower rate. The dashboard displays numerator and denominator beside the percentage.
Across both days, the pooled on-time rate is 1,170 / 1,200 = 97.5%. Averaging 98% and 95% gives 96.5%, which gives the low-volume day the same weight as the high-volume day. Use the pooled rate when the intended question concerns all eligible orders. If the question concerns a typical operating day, a daily summary may be useful, but label that different quantity explicitly.
Tuesday’s driver panel shows 14 unverified parcels within 30 minutes of collection, compared with the usual two to five. The dispatch lead deploys authorised verification support. A workload measure confirms that the action did not require unsafe stacking or excessive overtime. The dashboard has supported a practical decision while preserving the distinction between an early warning and the final outcome.
| Period | On-time / eligible | Rate | Late orders |
|---|---|---|---|
| Monday | 980 / 1,000 | 98% | 20 |
| Tuesday | 190 / 200 | 95% | 10 |
| Pooled | 1,170 / 1,200 | 97.5% | 30 |
04 · Use trends and signals honestly
Use trends and signals honestly
Show enough history to distinguish a recurring condition from a single result. A run chart can reveal the sequence, and a suitable control chart can provide a statistical reference when assumptions and sampling are sound. A target line answers whether the desired level was met; control limits answer a different question about variation. Label these distinctly. A red target miss is not automatically evidence of a special cause.
Northbank has a 99% on-time goal. Its stable pre-improvement rate is around 97.5%. The dashboard should identify a performance gap requiring process improvement instead of demanding a new explanation from each shift whenever the daily number is below 99%. After a deliberate improvement, annotate the date and examine subsequent evidence before declaring a durable change.
Small denominators deserve visible caution. One late order among ten produces 90%, but its precision differs greatly from 900 on-time orders among 1,000. Show counts and avoid ranking teams using percentages stripped of case mix and exposure. If missing data are material, display completeness as a measure or status rather than hiding the uncertainty in a footnote.
05 · Make the review an action system
Make the review an action system
Agree a review cadence matched to the decision. Near-cutoff queue information may need immediate review; realised labour savings may need monthly validation. Give every material abnormality an owner, next action and expected review date. Separate containment from investigation. The team can protect today’s collection while opening a deeper study of recurring release delays.
During Northbank’s daily review, the lead chooses three issues requiring action rather than reading every tile aloud. An unresolved barcode exception is assigned to a supervisor for immediate resolution. Repeated late collection is escalated to transport planning. A stable but elevated mislabel rate enters the improvement backlog. The next review asks what changed and what evidence supports closure.
Keep an action record linked to the relevant observation. Avoid counting “actions closed” as the primary success measure; a quick administrative closure may produce no benefit. Check whether the corrective action was implemented and whether the problem recurred. Escalation should request a decision or resource that the current owner lacks, not simply forward a red indicator up a hierarchy.
06 · Protect integrity and adapt the dashboard
Protect integrity and adapt the dashboard
Common errors include percentages without counts, stale data, decorative gauges without thresholds, excessive metrics and metrics whose owner cannot influence the result. Individual league tables can encourage avoidance of difficult work. Review potential gaming before release: could a team improve the number by moving work to another queue, delaying registration or changing the denominator? A balancing measure or stronger event definition may be necessary.
A dashboard should evolve when operating decisions change, but historical comparability needs protection. If Northbank changes “on time” from an internal scan event to actual vehicle loading, mark the definition change and avoid claiming that an apparent jump represents improvement. Keep a data dictionary and version the measure logic. Where useful, calculate both definitions during a transition period.
Advanced dashboards can support drill-down by carrier, service and shift, provided access protects sensitive information and users understand small-sample limitations. Drill-down should help investigate a question rather than invite endless searches for flattering subgroups. Retire metrics that no longer support decisions. A small, trusted dashboard with responsive ownership is more useful than a comprehensive display that nobody can explain.
07 · Distinguish a rate from a workload forecast
Distinguish a rate from a workload forecast
The near-cutoff queue is a count, but its implication depends on remaining time and available authorised capacity. Fourteen unverified parcels may be manageable with thirty minutes and two trained staff, yet critical with five minutes and one person. Northbank therefore displays the count beside time to cutoff and expected processing capacity. It does not label every queue above ten as an emergency regardless of context. Capacity estimates remain assumptions and are checked against actual work, especially for exception-heavy batches. This makes the leading indicator a basis for a specific support decision rather than a generic red tile.
Connections
Related methods and phase use
Placed under Control because of the outcome it supports. Use the method in other phases when its purpose and assumptions fit the work; the phase list is a guide to selection, not a compulsory sequence.
Sources
Professional foundations and further reading
The explanation and fictional teaching examples are written for LeanTrainingLab. Linked primary sources provide professional foundations; example values and local conventions are stated in the guide. The accepted toolset draws on Six Sigma+Lean Toolset, second edition (Lunau, editor; Springer, 2013). No book text or examples are reproduced here.