← BACK TO THE BLOG DATA 21 SEP 2026

KPIs someone checks on Monday, not the ones that look good

BY ALFREDO MILLÁ·10 MIN READ

Hands holding a clipboard with a printed report of line and pie charts next to a laptop on a wooden table

When we walk into a company and ask about its business KPIs, the same thing happens almost every time: a list of twenty or thirty indicators appears, half a dozen of which everyone knows, while nobody has looked at the rest in eighteen months. No one removed them because removing them feels like admitting they were a mistake. So there they sit, taking up space and draining credibility from the ones that matter.

This guide is about the opposite. How to get to a short list of indicators that someone checks on Monday morning and that change what gets done on Tuesday. What a KPI is and isn't, how to pick them, how many belong at each level, how to write the definition down so the number stops being argued about in meetings, and why the temptation to measure whatever is easy to export ruins more indicator systems than any technical problem ever will.

What is a KPI, and what isn't one?

A KPI (key performance indicator) is a metric that tracks progress towards a specific goal, has an owner who answers for it, and supports a decision. Take away any one of those three legs and it isn't a KPI — it's just a number.

The practical distinction, with examples:

The test we run in the first session is blunt and it works: for every indicator on the list we ask, "what would you do differently if this number got 20% worse next month?". If the answer is "look at it", it isn't a KPI. It can live in a reference report, but not on the leadership dashboard.

A KPI with no decision behind it is a waste of attention. And the leadership team's attention is the scarcest resource in the company.

A second distinction that saves arguments: lagging and leading KPIs. Quarterly margin lags — it tells you what already happened and arrives too late to fix it. Average delivery time, the share of quotes sent within 48 hours, or incidents per 100 orders lead: they move weeks before margin does. A dashboard made only of lagging KPIs is a very expensive rear-view mirror.

How do you choose business KPIs without ending up with thirty?

By starting with decisions, never with the data you happen to have. The order matters because if you start from what the ERP exports easily, you end up measuring what the ERP exports easily.

The method we run in a two-hour session with the leadership team:

1. List the recurring decisions. Each manager writes down the three or four decisions they make each month or quarter: raise prices, hire, increase stock for a product family, drop a customer, add a shift. 2. Name the number that would change that decision. For each decision, one number and a rough threshold. "If this line's margin drops below 18%, we review pricing." 3. Group and remove duplicates. Three people usually ask for the same thing under three different names. Half the list falls away here. 4. Check that the data is feasible. Does the source data exist? In which system? At what quality? This step reorders priorities; it doesn't decide them. 5. Close at 8-15 indicators for the leadership view. The detail doesn't disappear — it moves down to the departmental level.

Step 4 is where people most often get the logic backwards. A missing data point doesn't kill the KPI; it turns it into a small data-capture project. If delivery time is critical to your business and nobody records the actual dispatch date, the fix isn't to stop measuring it — it's to record the date. The alternative, measuring whatever is already there, is exactly the mistake described at the top.

One more rule that prevents decorative dashboards: somebody in the room has to be able to move each KPI this quarter. "Sector market share" isn't something your leadership team shifts in ninety days. "Orders shipped complete first time" is.

How many KPIs, and at which level?

Three levels, each with its own volume rule. This is the structure that holds up best in companies of 50 to 500 people:

| Level | Who looks at it | How many | Cadence | Example | |---|---|---|---|---| | Leadership | Leadership team, managing director | 8-15 | Monthly, light weekly review | Gross margin by line, net customer additions, average delivery time, turnover in key roles | | Department | Heads of sales, operations, finance, support | 5-10 per area | Weekly | Quote-to-order conversion, plant utilisation, days sales outstanding, first response time | | Operational | Team, team leads | Whatever the shift needs | Daily or continuous | Orders pending picking, ticket queue, open incidents by age |

This split solves two things. First, the classic fight between leadership and operations over one screen: they don't share the screen, they share the definition. The "average delivery time" at level 1 and the "orders late today" at level 3 are calculated from the same source data, so nobody argues about the figure — only about how fast to react.

Second, cadence. Checking a monthly KPI every day produces noise and impulsive decisions: the daily variation of a monthly indicator is almost entirely random. The reverse is just as bad — reviewing an operational KPI once a month turns it into archaeology. Review frequency is part of the KPI's definition, not a setting in the tool.

How do you define a KPI so nobody argues about the number?

With a written record of six lines. Boring, essential, and almost nobody does it. This is the minimum format we ask for before connecting a single data source:

| Field | What it holds | Example | |---|---|---| | Name | The one used everywhere, with no synonyms | Orders shipped complete first time | | Definition | The formula in words, numerator and denominator | Orders delivered with all lines complete ÷ orders delivered in the period | | Source | Exact system and field | ERP, delivery notes table, line status field | | Exclusions | What is left out and why | Customer-cancelled orders, samples, internal transfers | | Owner | A person, not a department | Head of operations | | Target and threshold | Target value and the point where you act | Target 95%; below 90% it goes to the leadership meeting |

The "exclusions" field is the one that saves the most money. Nearly every leadership-team argument about whether a number is right is really an argument about exclusions: whether returns count, whether intercompany customers are in the sales figure, whether a partially fulfilled order counts as fulfilled. Write it down once and the argument doesn't come back.

The "owner" field has to be a named person. A KPI owned by "the operations department" has no owner: when the number moves oddly, nobody will explain it. That is the same logic behind data governance, which is why both projects work better run together.

Which KPIs by department make a good starting point?

There's no universal list, but there is a decent starting point that you then prune. These are the ones that, in our experience, most often survive the first review:

That last one is the most ignored and the one that prevents the nastiest surprises. Knowing how many critical processes depend on exactly one person is a risk KPI that appears in no textbook and is felt sharply when that person leaves.

On customer concentration: hardly anyone tracks it, and it explains half the crises at mid-sized companies. If 60% of your revenue sits with three customers, every forecast is fragile, and that shapes everything from the sales plan to your demand forecasting.

What kills a KPI system?

The five failures we see most often, each with its antidote:

If you've tried something similar and it stalled halfway, the diagnosis usually belongs to the same family of causes behind why AI pilots fail: no clear owner, no routine, no decision attached.

How do you get from the list to automated data?

With less technology than people fear and more agreement than they expect. Effort on a first KPI system usually splits like this: 20% defining them, 50% connecting and unifying sources, 20% implementing the logic behind each definition, 10% building the dashboard. The visible part is the cheap part.

A realistic 90-day plan:

One detail about step three: when the new number doesn't match the usual spreadsheet, it's almost never a calculation error — it's different exclusions. Documenting those differences one by one is what turns the dashboard into a single source of truth instead of "yet another number". We cover that layer in what business intelligence is, and the screen design in the executive dashboard guide.

Frequently asked questions

How many KPIs should a company have?

Between 8 and 15 in the leadership view, and 5 to 10 per department. Below 8 you're usually missing a dimension of the business; above 15 the dashboard stops being readable at a glance and each manager only looks at their own slice. Operational detail lives at lower levels, not on the leadership dashboard.

What is the difference between a KPI and a metric?

Every metric is a measurement; a KPI is the metric tied to a goal, with an owner, that triggers a decision when it crosses a threshold. If nobody does anything differently when the number moves, it's a tracking metric, not a KPI.

How often should KPIs be reviewed?

The value is reviewed at whatever cadence its definition specifies: daily for operational ones, weekly by department, monthly at leadership level. The list of KPIs itself is worth reviewing once a year, or whenever strategy changes, to retire the ones that no longer lead to a decision.

Do you need a BI tool to have KPIs?

Not to start. A first version in a spreadsheet is legitimate if the data load is automated and the definitions are written down. The limit shows up with several sources, a large history and multiple users editing: at that point the spreadsheet stops being a reference and becomes a versioning risk.

If your indicator list has grown unchecked and you suspect half of it goes unread, the cheapest place to start is an honest diagnosis: which decisions actually get made, what data exists to support them, and what dashboard can realistically be built in a quarter. That's what the audit is for — and if you'd rather spend half an hour telling us about it and have us say whether it's worth doing, let's talk.

Shall we apply it to your case?

The 360° AI Audit turns these ideas into a concrete plan for your company: three weeks, fixed price and the full picture of your AI before spending a euro.

See the 360° Audit→ Let's talk↗