A dashboard: what fits on one screen

A manager opens the dashboard: 34 metrics, 11 charts, five colours. A minute later the manager closes it and still asks, "How is the warehouse doing?" A dashboard is not there to display data; it is there to answer "is everything all right?" in five seconds. For that you must put on one screen not everything, but only what is needed for a decision.

First identify the reader

One dashboard will not suit everyone.

• The warehouse manager looks daily: stock accuracy, order fill rate, today's delays.
• The director looks weekly or monthly: days of stock, the value of idle stock, losses in so'm.
• A shift worker sees the number they can influence themselves: uncollected orders, faulty counts.

Do not put a monthly metric on a daily dashboard: it does not change during the week and only takes up space.

How to choose the metrics

Draw up a list of candidates, say 22. Put four questions to each:

1. Does it have a specific owner?
2. What will be done if it gets worse?
3. Is the data reliable, and were the balances checked by a count?
4. Is it refreshed as often as needed?

One "no" and the metric stays off the dashboard. Usually 8-10 of the 22 remain and the rest go into a separate report. Mind the comparison period too: for seasonal items compare not only with last week but with the same week last year. Compare December growth with November and you may mistake the holiday effect for a problem.

Five to nine, each in three parts

Every metric should carry three things: the current value, the change against the previous period and a threshold. Nine metrics give 27 elements and twenty give 60; the eye cannot take in that much at a glance. An example of a weekly dashboard:

• Stock accuracy: 96.4 percent, last week 97.1 (−0.7 points), threshold at least 95. Status: normal.
• Order fill rate: 91 percent, last week 93 (−2 points), threshold at least 90. Status: watch, because it is near the threshold and falling.
• Share of days without the item: 3.2 percent, last week 2.4 (+0.8 points), threshold at most 2. Status: problem.
• Late dispatches: 5.8 percent, last week 4.1 (+1.7 points), threshold at most 5. Status: problem.
• Idle stock (not sold for 90 days): 119 million so'm, that is 14 percent of 850 million so'm of stock (also 14 last week), threshold 15 percent. Status: normal.
• Days of stock: 48, last week 46 (+2), target 45-50. Status: normal.

Write the status rule down in advance. Normal: the metric is on the right side of its threshold. Watch: normal, but 1 point or less from the threshold and worse than in the previous period. Problem: the threshold has been crossed. Without a rule the colour is set by mood, and trust in the dashboard disappears.

The result is three normal, one watch and two problems: at first glance the manager knows where to look.

Divide the screen into layers

The top row: the three or four most important numbers, in large type, with a status colour. The middle row: a line of those same numbers for the last 12 weeks and a threshold line. The bottom row: a "where is the problem" table, for example the five zones or suppliers with the most delays or shortages. Each card can also hold a small line chart (a sparkline). Then the dashboard answers the question and points the way to the next one. You can build it in Excel, Google Sheets or another tool: the tool does not change the rules.

The five-second test

Show the finished dashboard to a colleague for five seconds and ask, "What got worse?" If they cannot answer, the screen is overloaded. Remove the following:

• Metrics without an owner: who can influence them?
• Totals whose change means nothing, such as "total number of documents".
• Decoration: three-dimensional charts, shadows, speedometers.
• Blocks that nobody has opened for a whole month.
• Numbers with no refresh time. A line saying "Updated: 08:30" prevents doubts about how fresh the data is.

Conclusion

A good dashboard consists of few numbers, each with its change and threshold, and the worst spot is visible at first glance. It rests on reliable balances and cleaned data: CountLab supplies stock-count results as verified figures and gathers them into a short management report.