JOURNAL
Published on
WHICH KPISACTUALLY BELONGON A DASHBOARD
The test isn’t “is this interesting?”. It’s “will anybody do something different?”.
THE ANSWER
On a dashboard I keep only the KPIs that change a decision. How many jobs are stuck, and for how long. How long it takes to answer a new enquiry. Which kind of work leaves a margin. How full the next fortnight is. A number nobody can act on is decoration.
AI-generated image
IN SHORT
Every tile has a decision written beside it.
Every metric has a person’s name on it.
An average with no worst case stays off.
THE TEST
A METRIC EARNS ITS SPACE
Whoever asks me for a dashboard sends the list of what can be measured. Every item on it is true, and hardly any of it useful. The question I cut with is one: who looks at this tile, and what do they do differently if the number moves? If the sentence won’t finish, the metric stays out. It would spend attention daily and give nothing back. (A dashboard nobody opens is a failed project, even when the numbers are right.)
THE FIRST CUT
THE KPIS I LEAVE OFF
They look serious and they change no decision. These are the first ones I remove.
The year-to-date total
It only ever goes up. When the pace changes, the running total notices months later. If a total is needed, it needs a comparable period beside it.
An average with no spread
Average response time hides the cases that lose you clients. A 2-day average can mean everything answered in 2 days, or 5 enquiries forgotten for 3 weeks. Those are 2 different companies, and the tile shows them as one.
Activity counts
Calls made, emails sent, files touched: they measure motion, not whether the work moves forward. They mean something only next to the outcome they are supposed to produce.
Site visits with no next step
Rising traffic is a fact, not a metric. You need how many of those visits turned into an enquiry with a name on it. Without the second number, the first one decides nothing.
The number with no owner
If nobody in the business is responsible for moving it, that tile stops being read. Before adding a metric I ask whose it is. If no name comes back, it stays out.
HOW I CHOOSE
FIVE STEPS BEFORE ANYTHING IS DRAWN
The order matters. Each step uses the answer from the one before, and skipping one means drawing on an assumption.
I start from the decision, not the chart
I write the sentence “looking at this, somebody decides to…”. If I can’t finish it with a concrete verb, the metric stays out. That filter removes more than half of the first list.
I name the reader and the frequency
A metric the owner reads monthly doesn’t take the same shape as one a manager opens every morning. Reader and frequency set the level of detail, before the data source does.
I find the source before promising the number
Some metrics already sit inside the data you have. Others could only be calculated by changing how people work. That difference has to surface now, not on delivery day.
I write down how it is calculated
What counts as a closed job, when the clock starts, what is excluded. Two people reading the same tile have to understand the same number. Otherwise the meeting argues about the definition instead of the decision.
I put a threshold next to the value
A number on its own doesn’t say whether it is a good number. Beside it goes the level at which somebody steps in. That is what turns a display into a working tool.
AI-generated image
WHAT GOES ON
METRICS THAT DECIDE SOMETHING
Each one with the decision it is meant to trigger. If the decision isn’t yours, the metric isn’t yours either.
Jobs stuck, and for how many days
The decision: who you ring today. It is the tile that stays open after the first month. It points at a person and an action, not at a mood.
Time from enquiry to first reply
The decision: where you move attention, or which manual step you remove. Read it with the worst case beside the average. Without that, it tells a calmer story than the real one.
Quotes sent and won, by type of work
The decision: which kind of work you stop quoting for, and which deserves more time. Split by type it says something. As one total, almost nothing.
Estimated hours against actual hours
The decision: which price you correct on the next quote. It is the metric that makes a price range honest, instead of leaving it to a feeling.
Load for the next 2 weeks
The decision: what you take on now and what you push out. It looks forward. It is the only one here that prevents a problem instead of recording it.
THE CONDITIONS
WHEN A DASHBOARD HOLDS UP
- How many metrics
- Few: the ones you take in at a glance, without scrolling. Every extra tile spends the reader’s attention, every day.
- Who reads it
- One named person per metric. With no owner a number becomes furniture.
- Where the data comes from
- Every metric has its source written into the documentation, along with the way it is calculated.
- When data is missing
- Empty states and error states are handled. The dashboard says the data didn’t arrive, instead of showing a zero that looks like a result.
- What it doesn’t do
- It doesn’t decide for you and it doesn’t replace people who know the work. It makes visible early what you would have seen anyway, too late.
A screen full of numbers nobody can move isn’t a tool: it’s a poster.
QUESTIONS
- How many metrics should I have?
- Fewer than you would list. The useful constraint is reading: if you have to scroll to reach the end, nobody reads the top any more. Start with the ones that have an owner and add when somebody asks twice.
- Isn’t revenue a KPI?
- It is a result, and it deserves to be visible. Day to day you need what comes before it: enquiries, quotes, stuck jobs, load. Revenue tells you how it went; the others tell you what you can still change.
- What if the data is messy?
- The dashboard makes that visible instead of hiding it, and that is already half the work. Then you decide whether to clean the source or change how the data gets entered. Correcting downstream is what you pay for every month.
- Does it need maintenance?
- If the dashboard depends on recurring data or an external API, yes — at least minimum monitoring. Support and further development are quoted separately.
If you already have the tiles in mind, write beside each one the decision it is meant to trigger. If the decisions are there, the dashboard can be built. If they aren’t, I’ll say so before we start.