Evidence-based decisions connect claims to information
Evidence-based decision making means using relevant information, assessing how trustworthy it is and making uncertainty explicit before acting. It does not mean that every decision requires a statistical study, or that measured numbers automatically outweigh informed observation.
This is one of the seven quality management principles. Its practical value is discipline: distinguish what happened from what you think caused it, and decide what further information would materially change the action. Judgement still matters, particularly where consequences are significant and evidence is incomplete.
Begin with the decision to be made
Define the question before collecting data. Are you deciding whether to change a supplier, investigate a process, release an output or adopt an improvement? Different decisions require different evidence. A broad dashboard may contain much information yet answer none of those questions clearly.
State the alternatives, consequences and timing. A reversible trial may justify acting with limited information. An irreversible decision with major consequences may require stronger assurance. The principle supports a proportionate approach, rather than an endless search for certainty.
Write down the main assumption behind the preferred action. If that assumption proves wrong, would you choose differently? This question helps identify the most valuable additional observation or check.
Check relevance before precision
A precise measure can answer the wrong question. Fast dispatch does not establish timely delivery. A high completion count does not establish acceptable work. Ask whether the evidence represents the outcome the organisation is trying to achieve.
Quantitative and qualitative evidence can complement each other. A pattern of repeat contacts may show scale; interviews and case review may explain the confusion behind it. Neither should be accepted uncritically. A memorable complaint can be unrepresentative, while an average can hide a serious minority of cases.
| Evidence question | Why it matters |
|---|---|
| What is being counted? | Definitions can change the meaning of a trend |
| What is the denominator? | Counts alone may reflect changes in volume |
| Which cases are missing? | Incomplete coverage can distort the picture |
| When was information collected? | Different time periods may not be comparable |
| Who recorded it and how? | Collection practices affect reliability |
| What else changed? | Several explanations may fit the same result |
Examine reliability and limitations
Review whether the collection method is suitable for the decision. An instrument reading may depend on measurement conditions; a service category may depend on employee interpretation. If people apply different definitions, apparent differences between teams may be partly differences in recording.
Look for missing, duplicated or retrospectively changed records. Do not quietly clean data without understanding why defects occurred. The same defect may reveal a process problem worth addressing separately.
Explain limitations in the decision record. A small sample can still be useful, but it should not be presented as comprehensive. A recent process change may mean older information no longer represents current performance. Qualifying a conclusion makes it more usable by showing where further checking is warranted.
Hypothetical example: deciding about a supplier
A hypothetical distributor receives eight complaints about damaged items from one supplier and three from another. A manager proposes removing the supplier with eight complaints. The quality team checks purchase volumes and discovers that the first supplier delivered many more items.
The team then separates damage likely to originate in packaging from damage associated with handling after receipt. It reviews photographs, product types and delivery routes. The information suggests a packaging weakness for one fragile item, rather than poor performance across the supplier’s range.
The distributor agrees a focused packaging trial and defines what evidence would support wider adoption. This hypothetical example does not establish that complaint rates always resolve supplier decisions. It shows why counts, exposure and plausible causes should be considered together.
Separate association from explanation
Two events moving together do not necessarily establish that one caused the other. A fall in errors after training may coincide with lower workload, a changed product mix or increased supervision. Consider these alternatives before attributing the result entirely to the intervention.
Where practical, compare suitable periods or groups, inspect relevant cases and look for a credible mechanism linking the change to the result. The strength of analysis should match the decision. Avoid complex calculations that add apparent authority without improving understanding.
The article on improvement as a quality management principle explains how to build an effectiveness review into the plan for a change. Deciding what evidence matters before implementation reduces the temptation to select whichever later number looks favourable.
Act with a stated confidence level
A useful decision record can be concise: the question, evidence considered, important limitations, decision owner, action and review trigger. It need not become a separate form where an existing meeting record or operational system already serves the purpose.
When evidence is incomplete, specify the conditions for revisiting the decision. That might be the next representative group of cases, a failed trial criterion or new information about an assumption. Assign someone to notice and respond to that trigger.
Review a recent important decision by asking whether another informed person could follow the reasoning. They need not agree with every judgement, but they should be able to distinguish observations, assumptions and the basis for action. That visibility makes decisions easier to challenge constructively and improve over time.