Types of measures
Did the change help—and how would we know?
Choose a small set of measures before introducing a change. Together, they should show whether results improve, whether the change is being used, and whether it creates problems elsewhere.
For example, suppose a business checks delivery capacity before promising customers a delivery date.
1. Outcome: are results getting better?
Measure the result that matters to customers, patients or the service—not just completed activity.
Example: the percentage of orders delivered by the date first promised.
A training session or a new policy is an action. Neither, by itself, shows that deliveries improved.
Coverage is different from outcome. A dementia diagnosis target counts how many people have a recorded diagnosis; it does not say how long they waited, whether the diagnosis was right, or what support followed. See the dementia diagnosis analysis.
2. Process: is the change happening?
Measure whether the intended way of working is actually being used.
Example: the percentage of orders whose delivery date was checked against available capacity before being promised.
This helps distinguish an idea that did not help from one that was never properly implemented.
3. Balancing: are we making anything worse?
Check for harm, extra work or problems shifted to another part of the system.
Example: track customer cancellations and staff overtime. More reliable promises would be less useful if customers walk away or staff must work excessive hours to meet them.
Measure fairly over time
- Keep definitions consistent. Specify what is counted, which group is included, and the period covered. Do not improve the figure by changing who or what counts.
- Allow for variation. Nearly everything varies. Use run charts and, where suitable, step-change analysis to look for changes beyond routine—or common-cause—variation. Do not judge from one good or bad result.
- Check the data and context. Missing records, changes in demand, seasonal patterns and observations that move together can affect interpretation and the choice of chart.
Allow enough time
Agree what should change, roughly how much, and when. Allow time to implement the intervention, for effects to develop, and for enough observations to detect them. Some outcomes may take several years.
Use leading indicators—earlier signs of progress or trouble—to help anticipate a later result. These may be process measures or intermediate outcomes.
Example: the percentage of orders ready for dispatch two days before their promised delivery date. This gives an earlier indication than waiting to count late deliveries. Check whether it actually predicts on-time delivery; it is not a guarantee.
A chart can suggest that performance changed. It cannot, by itself, prove why. No detected change does not automatically mean no benefit.
What’s your next useful question?
What small set of measures would tell us whether our change helped—and whether we caused problems elsewhere?
Agree the measures, definitions and review dates with the people doing the work.
Return to right problem, right intervention, right measures →
Explore your data with the analyser →