Why responsible organisations test what they think they know

Every organisation has opinions. They appear in strategy meetings, staff conversations, headlines and the stories leaders tell about customers, communities and risk. Opinion is not the enemy of good decision-making. It is often where inquiry begins. The danger comes when a plausible story is promoted into a fact, then embedded in policy, budget and practice without being tested.

“Crime is rising.” “Young people have lost resilience.” “Staff do not want to return to the office.” “This community is hard to reach.” Each statement may contain something worth investigating. None is sufficiently precise to guide action. Which crime? Measured by which source? Rising from when, and where? Which young people? What does resilience mean? Who was asked about office work? Is the community hard to reach, or has the organisation used inaccessible routes?

Evidence before opinion is therefore not a demand for leaders to become statisticians or for human judgement to disappear. It is a discipline: state the claim clearly, identify the uncertainty, seek appropriate evidence, listen to affected people, test action and remain willing to change course. In the United Kingdom, public bodies are expected to work in this way. HM Treasury’s Magenta Book describes evaluation as a tool for learning and accountability before, during and after an intervention. The principle applies just as strongly to schools, charities and businesses.

The stakes are high because assumptions distribute resources and consequences. A mistaken view of crime can produce fear, punitive responses and misplaced spending. A fashionable inclusion initiative can consume time while leaving unequal progression untouched. A leadership team can interpret silence as consent and confidence as competence. Evidence does not guarantee a wise decision, but it makes reasoning visible and correction possible.

A statistic is not self-explanatory

Crime offers a useful example because public perceptions are shaped by direct experience, media attention, political argument and multiple statistical sources. The Office for National Statistics explicitly advises using the Crime Survey for England and Wales and police-recorded crime together. They answer different questions.

The Crime Survey asks a representative sample about experiences, including incidents never reported to police. Its consistent methodology makes it better suited to tracking long-term trends in common crimes such as violence, theft and criminal damage. Police-recorded data cover a broader set of offences and are especially important for homicide, robbery and weapon offences, but totals are affected by reporting, recording practices and police activity.

For the year ending December 2025, the ONS estimated 9.6 million incidents of headline crime through the Crime Survey, with no statistically significant change from the previous year. The same measure was 15 per cent lower than in the year ending March 2017, when fraud and computer misuse were first included. Police recorded around 5.2 million crimes excluding fraud and computer misuse, 2 per cent fewer than the previous year but more than in 2015. Those figures can appear contradictory only if we expect one number to describe the whole reality.

The responsible conclusion is not “crime is up” or “crime is down”. Long-term survey trends show decreases in theft, criminal damage and violence compared with the mid-1990s, while fraud has become a major part of contemporary victimisation and particular high-harm offences require separate attention. National trends may also differ from local experience. A specific neighbourhood can face a genuine increase while the national total is stable. Evidence improves the question: what harm, affecting whom, in which place, over which period, and according to which measure?

This precision is not pedantry. If an organisation misdiagnoses the problem, even sincere action can cause harm. A school responding to a general fear of youth violence with blanket surveillance may reduce trust without addressing the small number of situations driving risk. A local partnership relying only on recorded crime may overlook harms that victims do not report. Equally, dismissing fear because a national average has fallen ignores how insecurity is experienced. Data and lived experience must be brought into conversation.

Start with the decision, not the data

Organisations often collect large amounts of information without knowing which decision it should inform. Dashboards expand while judgement remains unchanged. Evidence-led practice begins in the opposite direction: define the decision, then identify the evidence needed.

Suppose a youth organisation is considering an evening programme because staff believe local violence peaks after school. It should make the assumption explicit: providing trusted activity during a particular period will reduce exposure to risk and improve engagement. It then needs baseline information. When and where do incidents occur? Who currently uses provision? What prevents others attending? What alternative explanations exist? Transport, family responsibilities or fear of travelling may matter more than the absence of activity.

This logic forms a theory of change: a transparent account of how resources and activities are expected to produce outcomes. The value is not the diagram itself. The value is exposing the links most likely to fail. If attendance is essential, accessibility must be tested. If trusted relationships are the mechanism, staff continuity matters. If reduced offending is the outcome, a short satisfaction survey will not establish it.

HM Treasury’s evaluation guidance recommends planning evaluation during policy design, not adding it at the end. Before implementation, organisations can examine existing evidence and test assumptions. During implementation, they can check whether delivery is working and identify unintended effects. Afterwards, they can assess impact and value for money. This prevents evaluation from becoming a ceremonial report written after decisions are irreversible.

Use the right kind of evidence

Evidence has a hierarchy only in relation to a question. Randomised controlled trials can provide strong evidence of causal impact in suitable conditions, but they cannot answer every question and may be unethical or impractical. Administrative data can reveal scale and patterns while missing unrecorded experience. Interviews can illuminate mechanisms and meaning but should not be used to estimate prevalence. Professional judgement can identify emerging risks but is vulnerable to selective memory and organisational culture.

Good decisions use triangulation: evidence from different sources that fail in different ways. If staff believe a programme improves attendance, compare records before and after, examine an appropriate comparison where possible, interview pupils and families, and check whether the pupils reached are those most likely to benefit. Agreement across sources increases confidence. Disagreement is not an inconvenience; it is information.

Statistical literacy also requires attention to uncertainty. “No statistically significant change” does not prove that nothing changed. It means the available estimate cannot distinguish a change from sampling variation at the stated confidence level. A percentage increase can look dramatic when the starting number is small. An average can conceal subgroup differences. Correlation can identify a relationship without showing that one factor caused the other.

Leaders do not need to perform every analysis, but they must ask competent questions. What is the source? What population does it cover? What is excluded? Is the comparison like for like? How large is the effect? Is it practically important? Who may be harmed if we are wrong? These questions create a culture in which evidence can challenge status rather than merely decorate a preferred decision.

Lived experience is evidence, not ornament

Organisations sometimes create a false contest between numbers and stories. Quantitative data are treated as objective, while lived experience is treated as emotional; or a powerful testimony is treated as more authentic than any aggregate pattern. Both positions are mistaken.

Lived experience provides knowledge that systems often fail to record. It can explain why a service is not used, how a policy is experienced and which informal practices determine outcomes. A young person’s account of feeling targeted by school discipline cannot alone establish the prevalence of unequal treatment, but it may reveal mechanisms that exclusion statistics cannot show. It can generate hypotheses, improve measures and identify unintended harm.

Participation must be designed ethically. Asking people to retell trauma is not automatically empowering. Organisations should explain how testimony will influence decisions, provide support where necessary, compensate expertise where appropriate and return with an account of what changed. They should include more than the most confident or institutionally fluent voices. Otherwise, consultation selects for people easiest to hear.

There is also a difference between experience and representation. No individual speaks for an entire community. People who share an identity can hold conflicting views, and disagreement should not be edited out to create a clean organisational narrative. Evidence-led inclusion asks which experiences are missing and how power shapes who is invited, believed and quoted.

Bias does not disappear when data arrive

Evidence can be selected and interpreted to protect an existing position. Leaders may demand impossible proof from ideas they dislike while accepting weak anecdotes that confirm their instincts. Metrics may be changed once results become uncomfortable. Data can be collected at an aggregate level that hides inequality or presented without the denominator needed to understand risk.

This is why transparency matters. Decision papers should distinguish fact, interpretation, assumption and value judgement. These categories interact but should not be confused. “Thirty per cent of participants left early” is an observation. “The programme is too demanding” is an interpretation. “Reducing content will improve completion” is a testable assumption. “Completion should take priority over depth” is a value judgement.

Pre-agreeing outcomes and decision rules can reduce hindsight bias. Before a pilot, specify what success, adaptation and stopping would look like. Record changes and reasons. Invite scrutiny from people not invested in the programme. Where evidence is uncertain, use reversible action and proportionate testing rather than organisation-wide rollout.

The UK government’s 2024 Inclusion at Work Panel made a related point about workplace diversity and inclusion. It found that organisations commit substantial resources while often lacking evidence about impact, and recommended focusing on interventions that can be assessed for effectiveness and value. The message is not that inclusion is optional. It is that good intentions do not exempt inclusion work from evaluation. Indeed, the people affected deserve more than activity presented as progress.

Build a test-and-learn organisation

An evidence culture is not created by one research team. It depends on routines. Meetings should begin with the decision and current confidence, not with the most senior opinion. Teams should be rewarded for identifying failed assumptions early. Pilots should be designed to produce learning, not launched as miniature publicity campaigns whose success has already been announced.

A practical cycle has six parts. First, define the problem precisely and identify who experiences it. Second, review existing evidence, including evidence that challenges the preferred explanation. Third, build a theory of change and state critical assumptions. Fourth, test at a scale proportionate to risk. Fifth, examine implementation, outcomes, costs and distribution. Sixth, decide whether to stop, adapt or scale, documenting the reasoning.

This approach is compatible with urgency. Test and learn does not mean endless hesitation. When harm is immediate, leaders may need to act with incomplete information. The response should still make uncertainty explicit, favour actions that can be monitored and changed, and establish review points. Speed without learning repeats mistakes quickly; analysis without decision allows harm to continue.

The College of Policing’s What Works Centre for Crime Reduction demonstrates how research can be translated into practical choices. Its toolkits summarise evidence about interventions and gaps rather than pretending every policing problem has a settled answer. Organisations in other fields can adopt the same posture: use the best available evidence, grade confidence and remain clear about what is unknown.

Evidence and values

Evidence cannot decide what an organisation should value. Data may show that an intervention is efficient, but leaders still decide whether its objective is just. A disciplinary policy might produce orderly corridors while damaging trust and disproportionately excluding particular pupils. Measurement must therefore include the outcomes that matter to those affected, not only those easiest for the institution to count.

Values without evidence can become theatre; evidence without values can become technocracy. Responsible leadership connects the two. It states the intended public or organisational good, examines whether action achieves it, and considers distribution, dignity and unintended consequences.

This is especially important when discussing belonging. An organisation may improve representation while employees continue to feel unable to disagree. A school may reduce formal exclusions while using internal isolation more often. A police force may increase recorded engagement events without increasing trust. Metrics should test the substance, not only the label.

The courage to be corrected

The deepest barrier to evidence-led practice is not technical capacity. It is identity. Leaders become attached to programmes they created, narratives they repeated and instincts that helped them succeed. Contrary evidence can feel like a challenge to competence or moral purpose. Organisations then turn measurement into defence.

Evidence before opinion requires a different account of leadership. Credibility does not come from always being right. It comes from making reasoning inspectable and changing when reality disagrees. A stopped pilot can be a success if it prevents a costly failure. A challenged assumption can be progress if it produces a better question.

Public debate will continue to reward certainty, particularly on emotionally charged issues such as crime, education and identity. Organisations do not have to imitate it. They can distinguish national trends from local harms, prevalence from severity, association from cause and activity from impact. They can respect experience without asking one story to carry the weight of a population.

The practical rule is modest: opinion may open the conversation, but it should not close it. Ask what would need to be true. Find out what is known. Test what is uncertain. Listen to the people who live with the consequences. Then act, measure and adapt. Evidence will not remove disagreement. It gives disagreement a more honest foundation—and gives organisations a chance to learn before their assumptions become somebody else’s reality.

References

Cabinet Office and Equality Hub (2024), Inclusion at Work Panel: Report on Improving Workplace Diversity and Inclusion.

College of Policing (2026), What Works Centre for Crime Reduction and Crime Reduction Toolkit.

HM Treasury and Evaluation Task Force (2026), The Magenta Book: Central Government Guidance on Evaluation.

HM Treasury and Evaluation Task Force (2026), Test and Learn: Supplementary Guidance to the Magenta Book.

Office for National Statistics (2026), Crime in England and Wales: Year Ending December 2025.