- What Is Your Dashboard Actually Showing You?
- Lie One: The Average Hides the Failure
- Lie Two: The Rollup Cancels the Problem
- Lie Three: The Period Boundary Moves the Result
- Lie Four: The Definition Changed Quietly
- Why Doesn’t Anyone Notice?
- How Do You Fix a Lying Dashboard?
- What Does an Honest Dashboard Look Like?
- Frequently Asked Questions
- About the Stagnation Assassin
Your dashboard is green and your business is not fine. That is not a paradox and it is not a reporting failure in the sense people usually mean. Nobody entered a wrong number. The systems are working. The dashboard is green because of what happens to information on the way up, and the answer is almost always arithmetic rather than dishonesty.
Four operations run between the event on the floor and the tile on the screen: numbers get averaged, they get rolled up, they get cut at a period boundary, and they get defined by somebody. Every one of those operations destroys information, and three of the four destroy specifically the information you needed. By the time the number reaches a screen it has been through all four, and it is entirely possible for a genuinely serious problem to survive the journey as a green tile with a small upward arrow.
What Is Your Dashboard Actually Showing You?
It is showing you a single statistic, calculated over a chosen population, across a chosen period, under a definition somebody wrote down once. Change any of those four and the number changes, often dramatically, without a single underlying fact changing at all. That is not a criticism of anybody’s reporting. It is a description of what summarization does, and it is unavoidable. What is avoidable is reviewing the output of that process as though it were the process itself.
The four hidden choices
Most people reading a dashboard believe they are looking at a fact. They are looking at four decisions. Which statistic (mean, median, count, percentage). Which population (all orders, shipped orders, orders excluding the ones flagged as customer-caused). Which period. And which definition. Nobody is hiding these. They just are not on the screen, and the screen is what gets reviewed.
Why the tile format makes it worse
A tile has room for one number, a comparison, and a color. That format is not neutral. It forces exactly the summarization that destroys the information, and then it applies a color that tells the reader whether to think about it. A red tile at least gets investigated. A green tile is an instruction to move on, which is why a green tile over a bad reality is the single most expensive object in a reporting system.
Lie One: The Average Hides the Failure
The average is the default statistic on nearly every dashboard and it is the wrong one for most operational questions, because averages are built to suppress variation and variation is usually the thing you are managing.
Why the mean is wrong for a promise
Hypothetical: a plant ships 100 orders in a month. Ninety-two ship on the promised day. Of the eight that miss, six are one day late and two are eleven days late. Average days late across all 100 orders is 0.28 days, which rounds to nothing and reports as excellent. Meanwhile two customers waited eleven days past a commitment, and those two customers are the ones deciding right now whether to keep single-sourcing with you.
An average is a statement about a population. Your customer is not a population. Your customer is one order, and they experience only their own order.
What to show instead
Show the distribution, or at minimum show the tail. On-time percentage plus the worst decile is two numbers instead of one and it tells a completely different story than the mean. For any metric where the customer or the constraint experiences individual events rather than the aggregate, the tail is the metric and the average is context.
Lie Two: The Rollup Cancels the Problem
Aggregation across units does something worse than averaging within one. It lets problems offset each other, so two real failures moving in opposite directions report as stability.
Offsetting variances
Hypothetical: three plants. One improves margin by two points because of a favorable mix shift that will reverse next quarter. Another loses two points because of a genuine pricing failure that will compound. Rolled up, the business reports flat margin and nobody looks. Both plants had news. The rollup deleted both.
This is why the most useful discipline on any aggregate number is to break it down one level and check whether the components are moving together. If they are moving in opposite directions, the aggregate is not a summary of anything. It is a coincidence.
Where aggregation is legitimate
The same logic applies across time as well as across units. A quarter that nets out to flat can contain a month of genuine deterioration followed by a month of unsustainable recovery, and the quarterly view will report stability for a business that was never stable at any point inside it. Wherever an aggregate is used to reassure rather than to decide, check whether its components ever pointed in opposite directions.
Aggregation is fine when you are making a decision at the aggregate level. Total cash, total capacity, total spend for a corporate negotiation. The failure is using an aggregate to decide whether a component-level problem exists, which is what a dashboard implicitly does every time a leadership team reviews it and moves on.
Lie Three: The Period Boundary Moves the Result
A period boundary is an arbitrary line, and any arbitrary line that people are measured against will be managed.
What month end does to the numbers
Shipments get pulled into the last three days. Orders get held for the first week of the next period. Maintenance gets deferred past the line. Purchase commitments get timed around it. None of these require dishonesty and most are individually rational responses to being measured on a boundary. Collectively they mean your monthly number describes the boundary as much as the operation.
How to detect it
Plot the metric daily across the period instead of viewing the period total. If the shape is flat with a spike at the end, the monthly number is partly an artifact of the calendar. Rolling periods are the structural fix: a trailing four weeks recalculated weekly removes the boundary entirely and is usually a two-hour change in the reporting layer.
Lie Four: The Definition Changed Quietly
This is the least discussed and the most corrosive, because it breaks comparison across time, which is the entire purpose of tracking anything.
Silent redefinition
Somebody decides that orders delayed for customer-supplied reasons should not count against on-time performance. That is a defensible position. It is also a redefinition, and after it happens, this year’s number is not comparable to last year’s number. If it happened without a dated note, the trend line on your dashboard is now a composite of two different metrics drawn as one continuous line.
A metric nobody owns the definition of will drift toward whatever makes the reporting easiest, and it will do it quietly, one reasonable exclusion at a time.
The metric dictionary
One page per metric: formula, population, exclusions, source system, owner, and a dated change log. It is the cheapest control in the reporting system and almost nobody has it. Every metric change gets a dated entry, and any chart spanning a change gets a visible marker at that point.
Why the fix is not more data
The instinctive response to a boundary problem is to add granularity: daily reporting, then shift reporting, then hourly. That helps with detection and does nothing about the incentive, because the incentive lives at whatever line people are measured against. A plant reported daily and rewarded monthly will still manage the month. Move the measurement to a rolling window and the line disappears, which is a cheaper and more permanent fix than any amount of additional resolution.
Why Doesn’t Anyone Notice?
Because a green dashboard is confirming rather than alarming, because the people who understand the calculation are not in the meeting where the number gets interpreted, and because nobody has been given the job of owning definitions.
The interpretation gap
The analyst who built the metric knows about the exclusions, the boundary effects, and the mix sensitivity. That analyst is usually not in the leadership review. The leader reading the tile has none of that context and no reason to suspect it exists. The gap between those two rooms is where most reporting failures actually live, and it is a structural problem rather than a competence problem.
The objection worth answering
People push back that this is over-thinking, that directionally correct is good enough for management. Directionally correct is fine. The problem is that these four operations do not produce directionally correct numbers. They produce numbers that are wrong in a consistent direction, always toward calm, because every one of them suppresses variation and variation is where the bad news lives.
How Do You Fix a Lying Dashboard?
Six steps, and they take about two weeks for a mid-market manufacturer. Do them in order, because the definitions have to be locked before anything else is worth recalculating.
- Pick the five numbers your leadership team actually acts on. Not the forty on the screen. The five that change behavior.
- Write one page per metric: formula, population, exclusions, period convention, source system, and a named owner.
- Replace every average with a distribution, or at minimum add the worst decile alongside the mean.
- Break each rollup down one level and check whether the components move together. Where they offset, retire the aggregate or publish it alongside its components.
- Convert period totals to rolling windows wherever a boundary effect is visible in the daily plot.
- Recalculate the last four quarters under the locked definitions and compare to what was reported at the time. The gap is your credibility problem, and finding it yourself is considerably better than a customer finding it for you.
What Does an Honest Dashboard Look Like?
Fewer numbers, each carrying its population and period on the face of it, distributions instead of means wherever individual events matter, components visible next to any aggregate, and a definition anyone can pull in one click.
The practical version of this is narrower than it sounds. You are not rebuilding the reporting stack. You are adding a population and a period label to the face of each tile, swapping four or five means for distributions, publishing components beside the two or three aggregates that survive, and putting the definitions one click away. In most businesses that is a fortnight of work in the reporting layer and none of it requires new data capture, because the underlying detail is almost always already being collected. What is missing is not information. It is the refusal to compress it before anyone has looked.
It is less comfortable to look at, which is the point. A reporting system that produces calm is doing a different job than a reporting system that produces decisions, and most dashboards were built, without anyone intending it, to do the first one. Which numbers deserve a permanent place on it is a separate question, and I have set out the twelve manufacturing KPIs worth a scorecard slot in detail. How often each one should be reviewed sits inside the operating cadence that surrounds it.
Frequently Asked Questions
Is a dashboard that shows averages always wrong? No. Averages are appropriate when the decision is made at the aggregate level, such as total capacity or total spend. They are wrong when the customer or the constraint experiences individual events, because then the tail is what matters and the average conceals it.
How do I know if my dashboard has a period boundary problem? Plot the metric daily across the period rather than viewing the period total. A flat line with a spike in the final days means the monthly figure is partly an artifact of the calendar rather than the operation.
Who should own a metric definition? The person accountable for the outcome the metric describes, not the person who produces the report. Definition ownership placed in the reporting function will drift toward whatever is easiest to calculate.
How many metrics should a dashboard show? Roughly the number your leadership team acts on, which in most businesses is around five at the top level and twelve at the business unit scorecard level. Anything beyond that consumes response capacity without changing decisions.
About the Stagnation Assassin
Todd Hagopian is a Fortune 500 transformation executive who has generated $3B in shareholder value across Berkshire Hathaway, Illinois Tool Works, Whirlpool, and JBT Marel, where he serves as Global VP of Product Strategy and Marketing. Known as The Stagnation Assassin, he is the author of two published books: The Unfair Advantage: Weaponizing the Hypomanic Toolbox and Stagnation Assassin: The Anti-Consultant Manifesto. His blog is published in 15+ languages and read by operators worldwide. Bring him to your stage via his speaking page or connect with him on LinkedIn.
Stagnation Genome scoring. Send me the five numbers your leadership team reviews most often, with their definitions, and I will score where your reporting is manufacturing calm instead of decisions. Free, and there is no pitch at the end of it. Request your scoring here.