Business Dashboards: Choosing Metrics That Drive Decisions

Almost all business dashboards die the same way: they get looked at for two weeks, then someone asks where a number comes from, no one knows for sure, and from that point on meetings go back to spreadsheets. The problem isn't the chart: it's that the metrics were chosen from what was available rather than from what's needed to decide. This guide will help you build a few reliable ones.

A useful dashboard can be recognized by one simple sign: when a number moves outside expectations, everyone knows who looks at it and what they do. If the number gets worse and nothing happens, that metric isn't useful, no matter how well designed.

Start From the Decision, Not From the Available Data

The fastest way to get it wrong is to ask the company what data it has. The right way is to ask managers what decisions they make, and how often.

• "Every Monday I decide which crews to assign." "Every month I decide whether to accept new job orders." "Every quarter I decide whether to revise the price list." These are decisions: each one needs two or three numbers, not twenty.
• For each decision, ask what is currently looked at to make it. Often the answer is a spreadsheet that someone updates by hand: that's the real metric, and it needs to be formalized.
• If a decision doesn't exist, the metric isn't needed. Revenue by day of the week is interesting; if no one can change anything based on that data, it just takes up space and attention.

This work typically produces five or six metrics for management and about as many for each department. If thirty come out of it, no choice was made: a list was made.

A Metric Sheet: Late Job Orders

Every metric needs a written sheet, shared and approved before it becomes a chart. Here's an example, with illustrative values, for a business that works on job orders.

The last row is the most important and it's almost always missing. A metric without its stated limits gets used to answer questions it can't answer, and at that point it leads to worse decisions than having no data at all.

The row on the owner is the second most important. If delivery dates are entered incorrectly, no dashboard can fix them: the owner is responsible for the quality of the data, not just for reading it.

Three Kinds of Data That Shouldn't Be Mixed

Numbers of very different natures coexist on the same screen, and showing them in the same typeface is the most elegant way to lead to a wrong decision.

• Consolidated data: comes from a system, refers to a closed period, and will never change again. It's the only kind you can discuss without caveats.
• Incomplete data: the period is still ongoing, or some entries will arrive later. It must always come with the date and time of the last update, or it will end up compared with a closed period.
• Estimate or forecast: comes from a calculation or a judgment, not from a recorded entry. It must be flagged as such, with the underlying assumption written next to it.

The trickiest case is comparing a month in progress with the same month the year before: the first is incomplete, the second is closed. The apparent decline doesn't actually exist, but it gets discussed in meetings as if it were real.

Linking a Deviation to a Check and a Person

A metric becomes a management tool when it has a follow-up already decided in advance. There's no need for a formal procedure: three lines written next to the threshold are enough.

1. When the metric crosses the threshold, who looks at it first and at what point in the week.
2. What check that person carries out: not "understand the problem," but a concrete action — open the list of job orders involved, call the two site managers, check whether the delay is concentrated on one vendor.
3. What they report and to whom, if the check confirms the problem. A two-step path is worth more than a ten-step plan.

Being able to drill down from the number to the list that makes it up is the feature that makes a dashboard usable. A value that can't be clicked on to see the rows behind it gets challenged at the first doubt, and at that point the meeting shifts to the method instead of the problem.

What This Guide Doesn't Cover

This guide defines your company's numbers: which metrics are needed, how they're calculated, and who is responsible for them. Using artificial intelligence to comment on or explain those numbers — what can be entrusted to it and what needs to stay under human control — is covered in a dedicated guide. The data structure of a job-site management system has its own guide as well.