Why measurement needs a plan
Centralizing a function, whether IT, finance processing, procurement or reporting, usually comes with a promise: things will be more consistent, cheaper, faster or easier to control. Without measurement, that promise cannot be checked. With too much measurement, the organization collects dozens of numbers, argues about them, and acts on none.
The practical goal is a small set of measures that connect to what the organization is trying to achieve, that people define the same way, and that get looked at on a regular schedule. This article gives an overview of that process. Individual steps, such as writing definitions or designing a dashboard, have their own articles; here the focus is on choosing well.
Start from goals, not from available data
It is tempting to measure whatever the current systems already report. That tends to produce numbers that are easy to collect rather than numbers that matter.
A more reliable route runs in the opposite direction:
- State the goal in plain words. For example, “Employees get their access requests handled quickly” or “Supplier invoices are paid correctly and on time.”
- Ask what a reader would see if the goal were being met, and what they would see if it were not.
- Choose a measure that would show that difference.
- Only then check whether the data exists, or can be collected at a reasonable effort.
If a goal cannot be tied to any observable measure, it is worth restating. If a measure cannot be tied to any goal, it is a candidate for dropping.
Keep the set small
There is no magic number, but a short list is easier to remember, explain and review. Many teams find that a handful of measures per function is enough for a regular review, with more detail kept available for investigation when something looks off.
A useful balance is to cover several angles rather than many measures of the same angle:
- Speed or timeliness: how long things take.
- Quality or accuracy: how often work is right the first time.
- Cost or effort: what resources the work uses.
- Experience: how the people served rate the service, such as customer satisfaction scores.
- Risk or stability: how often things fail or drift outside agreed limits.
Using a mix guards against a common trap: improving one number at the expense of another. A team measured only on speed may cut corners on quality. Pairing measures is a simple defence, and the idea is explored further in the article on target side effects.
Check each candidate measure
Before adopting a measure, it helps to ask a few questions.
- Does it link to a decision? If nobody would act differently whether the number went up or down, it is probably decoration.
- Can it be defined precisely? Two people should calculate the same value from the same data. Clear metric definitions are what make that possible.
- Is the data reliable enough? Known data quality issues can make a trend meaningless. It is better to fix or acknowledge the gap than to chart it with confidence.
- Who owns it? A measure with no named owner tends to drift or be quietly abandoned.
- How could it be gamed or misread? Every measure has side effects. Thinking them through in advance is cheaper than discovering them later.
Mix outcome and early-warning measures
Some measures report results after the fact, such as customer complaints last quarter. Others give an earlier signal, such as the age of the oldest open request. A balanced set usually includes both, because outcome measures arrive too late to steer by, and early signals on their own may not reflect what really matters. The distinction is covered in more detail in the article on leading and lagging indicators.
Establish where you stand first
A measure is only useful for judging a change if you know what it looked like before. Collecting a few periods of data before reorganizing anything makes later comparisons fairer. Without a performance baseline, it is hard to say whether a centralization project improved things, made them worse, or simply coincided with a seasonal swing.
Decide how and when to review
Measures that nobody looks at do not change anything. Agree on a review rhythm that matches how fast the underlying work changes: weekly for daily operational queues, monthly or quarterly for broader outcomes. The structure of those meetings matters as much as the numbers, and is the subject of the article on performance review meetings.
It also helps to review the measures themselves from time to time. Goals change, systems change, and a measure that was useful two years ago may no longer reflect what the function does. Retiring a measure is a normal, healthy outcome.
Be honest about limits
Numbers describe, but they do not explain. A rising average can hide a few very bad cases; a stable figure can hide a change in what is being counted. Context from the people doing the work is part of reading any measure correctly.
Centralization is also a trade-off, so measuring only the benefits it was meant to deliver gives a one-sided picture. It is reasonable to include measures that would reveal costs, such as slower local responses, workarounds appearing in business units, or dependence on a single team or system, which is how single points of failure tend to form. If the evidence shows a function works better closer to the people it serves, that is a legitimate finding, and a hybrid arrangement may be the answer.
A simple selection checklist
- Each measure is tied to a written goal.
- The list is short enough to review in one sitting.
- Several angles are covered: speed, quality, cost, experience, risk.
- Every measure has a written definition, a data source and an owner.
- A starting point has been recorded.
- A review date is in the calendar, along with a date to revisit the measures themselves.
These examples are general guidance rather than a prescription. The right set depends on the organization, its goals and the function being measured.