← Blog

Evidence-Based Management: Making Better Decisions Under Uncertainty

An operational decision rarely comes with complete evidence. A hiring process needs to change, a cash forecast needs rebuilding, or a new system needs to go live. Waiting for certainty is not an option. Neither is treating a persuasive presentation as proof.

Evidence-based management offers a practical discipline: use the best available information, make uncertainty explicit, and check whether a decision delivers the intended result.

What Management Can Learn from Medicine

Evidence-based medicine brings research evidence together with clinical expertise and the circumstances of the individual patient. Its original formulation was not a demand to replace judgement with a hierarchy of studies. It was a way to improve judgement by making its basis more explicit.

The useful parallel in management is to combine relevant research, organisational data, professional experience, and the needs of the people affected. A published finding is one input. Whether it applies to your company is a separate question.

This means asking about study quality, plausible mechanisms, alternative explanations, and context — not simply whether an article reports a statistically significant result.

Start with the Base Rate

Consider a hypothetical diagnostic test with 90% sensitivity and 90% specificity, used in a population where 1% of people have the condition.

Among 10,000 people, approximately 100 would have the condition. The test would identify 90 of them. Of the remaining 9,900 people, around 990 would receive a false positive. That is 11 false positives for every true positive; only about 8.3% of positive results would be true positives.

The lesson is not that the test is useless. It is that the starting probability matters when interpreting a result.

In management, the same discipline encourages useful questions: how often does this problem occur, what usually causes it, and how plausible is the proposed explanation? A surprising finding deserves examination, not automatic acceptance or rejection.

Examine the Strength of the Evidence

Publication and selection bias

Published success stories can be an incomplete sample. Positive findings may be more likely to appear than null results, while commercial case studies may emphasise favourable outcomes.

Look for transparent methods, pre-specified outcomes, independent evaluation, and negative findings. These features do not guarantee a correct conclusion, but they help assess how much confidence to place in it.

Replication and context

An initial result may become smaller, disappear, or hold up in later studies. Replication is informative precisely because the outcome is not guaranteed.

Ask whether an effect has been observed by independent teams and whether the settings resemble your own. A process that helps a large, established company may not suit a 20-person biotech startup.

Statistical significance and practical value

A p-value does not tell you how large an effect is or whether a change is worth implementing. Examine the effect size, uncertainty, costs, and consequences.

For example, an increase in risk from 0.1% to 0.2% is both a doubling of relative risk and an increase of 0.1 percentage point in absolute risk. Whether it matters depends on the severity of the outcome and the number of people exposed. The numbers alone do not settle the decision.

The operational equivalent is straightforward: time saved, errors avoided, and cash preserved matter more than a headline percentage without a clear baseline.

Combining studies carefully

A meta-analysis can provide a useful synthesis, but pooling results does not remove differences in study quality, populations, or outcome definitions. Examine those differences before applying an average effect to your organisation.

A Practical Decision Process

  1. Define the problem. “Our onboarding is poor” is too vague. “New hires wait several days for essential system access” gives you a problem to investigate.

  2. Record the baseline and assumptions. Measure the current situation and describe why you expect a proposed change to help. Distinguish observed facts from working assumptions.

  3. Set an outcome and a review date. For an onboarding change, you might track access readiness, time to the first independent task, and the administrative effort required.

  4. Run a proportionate test. A reversible, low-cost change may justify a small pilot. An expensive or difficult-to-reverse decision deserves stronger evidence and closer scrutiny. Where feasible, use a meaningful comparison rather than a before-and-after anecdote alone.

  5. Review and update. Decide whether to continue, adapt, or stop. Record unexpected effects as well as the intended result. A successful pilot supports further testing; it does not automatically establish a universal rule.

Rigour Without Paralysis

The goal is not to reproduce a clinical trial for every business decision. It is to match the strength of the evidence to the stakes of the decision, and to remain honest about what is still uncertain.

Good operational leadership needs both action and revision: make the best decision you can now, create a way to learn from it, and change course when the evidence warrants it.

Further Reading


Lionel ARNAUD draws on senior roles in biotech to provide hands-on support to CEOs through Dinnizer SAS, across finance, funding, HR, team leadership, and operations. Discuss your operational priorities or write to contact@dinnizer.com.