When a metric becomes a target, it stops measuring
In 1975, in an essay on monetary policy, a British economist described an observation that now bears his name: any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes (overview of Goodhart's law). The idea became famous in a shorter version that does not come from him. It goes back to an anthropologist who wrote about it in 1997 in an essay on performance assessment at British universities: when a measure becomes a target, it ceases to be a good measure.
The mechanism: as long as nobody is measured against a metric, it moves with what it is supposed to show. As soon as it goes into a target agreement, a bonus or the monthly report to management, behaviour changes. Whoever is measured looks for the shortest route to the number. That is rarely the same as the route to the result.
A documented extreme case shows how far this can go. In 2016, the US consumer protection authority imposed a fine of 100 million US dollars on a major bank. Under pressure from sales targets and pay incentives, employees had opened around 1.5 million deposit accounts without the customers' knowledge. The target was met. For the customer, the opposite of value had been created.
In mid-sized companies it plays out less dramatically, but along the same pattern. If you count calls, you get calls. Give your inside sales team a contact quota and it will be met. The only open question is how: with a quick callback to a customer who is happy anyway, or with an appointment that looks good in the CRM. The difficult customer who has been ordering less for months takes time and counts for the quota just as much as any other call. Nobody is acting dishonestly here. Everyone is doing what the metric asks for.
In summer 2026, the software vendor ChurnZero, which builds tools for managing existing customers, applied this idea to customer success in an article. Its recommendation: pair every activity figure with an outcome figure. This article applies the idea to the monthly reports of mid-sized companies.
The real problem: health scores measure effort instead of effect
A health score is supposed to answer a simple question: how are things with this customer? Usually it is a traffic light or a score per customer, fed by several individual values. The idea is right. The problem lies in the material the score is built from.
The scores we see often consist mainly of figures that are easy to capture: portal logins, calls made, emails sent, the annual review that has been held. These values sit in the CRM, can be counted cleanly and go up as soon as the team does its work.
That is exactly the flaw in the thinking. These figures show that something is happening. They don't show whether it has any effect on the customer. Ten calls in a quarter can mean a well-looked-after customer. Or a customer with an unresolved problem who has to keep chasing. The score does not tell the two apart.
The article mentioned above puts it like this: health scores seem objective, data-based and reassuring. If you build them on the wrong inputs, they mirror the behaviour of the account managers instead of the customer's situation. If a score relies heavily on emails, calls and mandatory meetings, dissatisfied customers can go unnoticed.
Many managing directors know the consequence: a customer sits on green for months and cancels anyway. Looking back, every activity had been done. But none of them measured whether the customer was making progress with the service. Why even a good score achieves little as long as nobody is allowed to act on its findings is covered in the implementation mandate for a health score programme.
You can check this in your own company. Take the last five cancellations and look at the colour the score had three months earlier. If most were green, your score probably measures effort.
The principle: always report activity and outcome as a pair
The rule: no activity figure stands alone in the report. Next to every figure that shows what your team did, there is a figure that shows what happened on the customer's side. This outcome figure should only move if the activity actually had an effect.
The article justifies the pairing by saying that it creates a warning signal as soon as the team hits its activity target but the results fail to follow.
An activity figure on its own can be pushed up almost at will. More calls, more reminder emails, tickets closed faster: this takes effort, but does not require anything to change for the customer. With the outcome figure next to it, you get a comparison in the same row. Calls rise by a third, renewals stay flat? Then the extra activity achieved nothing. The pair shows the gap by itself. Whether a measure really works only shows in a comparison with a control group, as proving the value of CX with a control group explains.
The principle does not abolish activity figures. It gives them a different role. If the renewal rate drops in a customer group, only the activity figure next to it shows whether nobody has called these customers for months or whether they were looked after regularly and are leaving anyway. Those are two different problems with different countermeasures. The outcome tells you whether you have a problem. The activity helps you understand what is causing it.
The following five metrics are examples. Each section shows how the figure gets met without having an effect, and which counterpart is missing.
Metric 1: Logins and frequency of use
The system supplies logins automatically, and they seem to give a direct view of the customer relationship. That is why they end up in the health score so often.
A constructed example shows what happens when the figure becomes a target. The team is told to increase monthly logins and sets up a reminder email: "You haven't logged in for two weeks." Customers click, take a quick look and close the window again. The login count rises. Order entry carries on in Excel, and nobody produces the analysis the customer bought the tool for. The report shows a green light. At the next renewal meeting, the question on the table is what they are actually paying for.
The login count shows that someone opened the door. It does not show whether they did any work in the room.
The missing counterpart: the value achieved from the customer's point of view. In practice, this means: what share of your customers actually run the use case agreed at the time of purchase in the system? The article makes a suggestion in the same direction: supplement an activity-heavy score with frequency of use and the depth of the features used. If logins rise after a reminder campaign but the share does not, the emails produced logins without producing use.
The prerequisite: you record for each customer what they wanted to use the product for. This information is usually in the quote or in the minutes of the kick-off meeting. It is rarely in the CRM.
Metric 2: Number of customer contacts and calls
"Customer contacts per account manager" or "calls to existing customers" are rows we regularly come across in monthly reports. An example: an inside sales team is told to call every A customer twice a week. After a few weeks the quota is met and the calendars are full. Which of these customers will still be buying next year is not in that row.
A contact quota rewards the call, whatever its effect. So the team does the obvious thing. It calls the customers who like talking on the phone first and pushes the difficult ones back. It holds short conversations that can be logged ("Just wanted to check everything's all right"). And it also counts conversations the customer started themselves, for example because of a complaint. The total shows how much your team worked, and that is all it shows.
The article points to a further consequence: customers who are contacted often but with little substance are more likely to go quiet or bypass their account manager.
The missing counterpart: renewals or follow-up orders among the customers being looked after. With fixed-term contracts, it is the share of expiring contracts, framework agreements or subscriptions that were renewed. In project or order-based business, it is the number of customers looked after who ordered again or held their volume. The data is usually in accounting or the CRM.
A pragmatic addition is a mandatory field in every CRM entry: "What changed for the customer as a result of this conversation?" If the field stays empty permanently, you already know a lot.
Read side by side, contacts and renewals produce three patterns:
| Contacts | Renewals / follow-up orders | Reading |
|---|---|---|
| rising | stable or falling | The team is working more and the customer doesn't notice. Check what the conversations are about. |
| stable | rising | The conversations are working. Instead of raising the quota, understand what is working. |
| falling | falling | A genuine early signal. Look at these customers first. |
Metric 3: Emails sent and open rates
The number of emails sent depends only on the sender. One click in the sending tool, and the monthly report shows a new campaign, a newsletter, a reminder to all existing customers. It says nothing about whether anything happens on the recipient's side.
The open rate looks like the matching counterpart. It isn't. It too only describes that an email was displayed. If it becomes a target, the team optimises subject lines that spark curiosity and send times that look good. The customer gets nothing out of it.
On top of that comes a technical problem. Opens are measured through content that is loaded remotely. Apple's mail privacy feature loads such content in the background by default, regardless of whether the recipient reads the email at all, and so prevents senders from learning when and how often an email was opened. Some of the counted opens are therefore not a person who read anything. The open rate fails as a steering metric on two counts: it measures the wrong thing, and it measures it inaccurately. Ask your email service provider how an open is counted in your setup.
The missing counterpart: the action triggered. What did the customer do after the email? Reply, order, book an appointment, request a document? These figures are usually already in the CRM, the shop or the calendar. If sending goes up while replies, orders and bookings stay the same, more activity achieved nothing. The open rate can stay in the report as explanatory background. You should not steer by it.
Metric 4: Ticket handling time in customer service
Handling time is one of the most popular figures in the service report. Every ticketing system supplies it automatically, and it drops as soon as the team works faster. It also drops when the team merely closes faster.
A customer reports that a delivery was invoiced incorrectly. The service employee sends a standard reply, sets the ticket to "done" and meets her target time. Three days later the same customer gets in touch again because the invoice is still wrong. A new ticket is created and is again closed quickly. The report shows two cases handled promptly. On the customer's side there is an unresolved problem and the experience of having explained it twice.
If handling time becomes the target, closing gets rewarded. Solving does not show up in the figure.
The missing counterpart: the share of requests that are reopened, or that receive a follow-up request about the same problem within a set period. This figure only moves if the problem was really fixed the first time. If handling time drops while reopenings rise, tickets were closed faster without being solved any better.
For the figure to be reliable, count follow-up requests as well as reopenings. Many customers don't reopen an old ticket; they write a new email. Define what counts as "the same problem", for example the same customer and the same order within 14 days. The period is our own choice; there is no standard for it. And report both figures in the same row. Handling time on page one and reopenings in the appendix do not make a pair.
Metric 5: Annual reviews held and surveys conducted
"Annual reviews held with key customers, customer survey sent, responses analysed": rows like these suggest closeness to the customer. But they only show that a meeting took place or a questionnaire came back.
You know the mechanism from the calls. If the meeting quota becomes the target, meetings get held. They pile up at the end of the year, follow a standard agenda and end with minutes recording that a conversation took place. What was agreed, who implements it by when and whether it happened remains open. The next year, the conversation starts from zero again. The quota is still at one hundred per cent.
The same goes for surveys. A good score on the question of whether the customer would recommend you is a snapshot. The article describes a deliberately constructed scenario: a customer gives a nine in January and cancels in March. The survey was done, the score was good, and nobody saw what shifted in between.
The missing counterpart: it has two parts. First, the share of agreements implemented. Record specific actions from every annual review, with an owner and a date, and report how many of them have been implemented by the next contact. A meeting without an agreement then counts as an indication rather than a success. Second, the development of revenue with this customer: follow-up order, renewal or decline are in accounting or the CRM.
The two-column grid: check your reports in 30 minutes
Only your own report shows which pairs are missing in your company. We use a simple grid for this. It is our own working tool and has not been validated as a standard. You need last month's report, a blank sheet and half an hour without interruptions.
In the left column, write every metric in the current monthly report. All of them, without preselection. In the right column, enter the outcome figure that only moves if the activity on the left had an effect on the customer. If there isn't one, the field stays empty.
This is what a typical report might look like on the first pass, before the gaps are filled. The empty fields are exactly the places for which the sections on metrics 3 and 5 suggest a counterpart:
| Metric in the report | Matching outcome figure |
|---|---|
| Logins per customer | Share of customers running the agreed use case |
| Number of customer calls | Renewals and follow-up orders among customers looked after |
| Emails sent, open rate | empty |
| Average ticket handling time | Reopenings and follow-up requests about the same problem |
| Annual reviews held | empty |
Three test questions per row
1. Can the figure be increased without the customer benefiting? If yes, it is an activity figure and needs a counterpart. If no, there is probably already an outcome on the left, such as revenue from existing customers. Mark the row and move on.
2. Which figure would give it away? Look for the value that stays still if the figure on the left was merely met. It arises with the customer rather than in your team: an order, a renewal, a reply, a problem that stays solved.
3. Do we have this figure today? If yes, enter it. If no, note where it could be, for example in accounting, the CRM or the ticketing system. A missing figure counts as a finding rather than an omission.
Empty fields are your action list
Every empty field on the right marks a place where your team can meet a target without anyone noticing whether it has an effect on the customer. Start with the activity figures that targets, bonuses or the discussion in the weekly meeting depend on. That is where the pressure to hit the number, whatever the effect, is greatest. For each field, decide who will obtain the missing outcome figure and from which source. Don't fill everything at once. Two or three completed pairs in the next monthly report say more than ten planned ones.
What changes in the health score once the pairs are in place
With the grid, every activity figure has its counterpart. That lets you rebuild the health score into two levels instead of one blended figure.
At the top are the outcome signals, and only they determine the traffic light: renewals and follow-up orders, revenue development with the customer, agreements implemented, reopenings about the same problem. Below them are the activity figures, meaning logins, contacts, emails and annual reviews. They explain why an outcome moves or doesn't, but they no longer colour the score.
This makes visible what used to get lost: a lot of activity on a customer whose outcome figures are standing still. That exact combination is the warning signal. In the old blended figure, it looked green. Which signals need a fixed action behind them is shown in early warning signals as on Flight 401.
You need neither a BI team nor a new tool for the rebuild. Most outcome figures are already in the company. Accounting knows revenue per existing customer over time and the follow-up orders. The CRM holds renewal dates, deals closed and, if documented, the actions agreed in customer meetings. The ticketing system supplies reopenings and repeated requests. The work consists of laying these figures side by side for each customer. A spreadsheet is enough to start with. Where a figure is missing, the gap goes on the action list from the grid.
The first step for Monday morning: take your ten largest existing customers by revenue. For each, enter two values: the current health score and a single outcome figure you already have today, such as revenue for the last twelve months compared with the same period a year earlier. Mark every customer with a green score and a falling outcome.
We are happy to review your reports with you in the Reactivation Audit.
How many marks are on your list at the end?
Frequently asked questions
What does Goodhart's law mean for customer management? The principle comes from the monetary policy of the 1970s. In short: when a metric becomes a target, it no longer measures reliably. In customer management, this means: if you set a number of calls, you get those calls, but not necessarily more loyal customers.
Should we get rid of activity metrics such as calls or logins altogether? No. They explain why an outcome moves. They only become a problem when they stand alone and become targets.
Why do so many health scores fail? Because they are often built mainly from emails, calls and meetings. Then they reflect the behaviour of your own team, and dissatisfied customers sit on green until the cancellation arrives.
Which outcome metrics are suitable if we don't have a BI team? The ones already available: renewals, follow-up orders and revenue from existing customers from accounting or the CRM, reopenings from the ticketing system. A spreadsheet is enough to start with.
How often should we review the metric pairs? Continuously in the monthly report, where the figures appear anyway. Once a year it is worth asking whether each pair still fits, for example after changes to the offering, the sales channel or the targets. Every new metric gets its counterpart as soon as it is introduced.
How do we stop outcome figures from being massaged too? You can't rule it out completely. Any figure that becomes a target comes under the same pressure, as the bank case in the first section shows. So draw outcome figures from systems that the reporting team does not maintain itself, for example from accounting instead of self-reported data. The pairs also keep each other in check: anyone who closes tickets quickly without solving the problem shows up in the reopenings.