Evidence-based leadership development means choosing who to develop, promote and back on measured signals that converge over time and that someone outside the original conversation could check, instead of on a feeling that a person seems ready. It does not remove judgement, and I would not want it to. It requires judgement to point at something specific, and it is honest about how strong that something is.
This page covers what counts as evidence. The companion pieces cover the next questions: how a CHRO proves development changes business performance is the chain from demand to outcome, and how to measure leadership capability across an organisation covers the instruments. Read this one first if you are deciding what you are prepared to call proof.
A fair starting point is that the research base is itself uneven. Geerts's 2024 review in Behavioral Sciences describes the scholarly knowledge base on leadership development as predominantly anecdotal and seldom involving objective data, and notes that programme evaluations typically only assess participant satisfaction. The CIPD's 2023 evidence review finds a moderate positive effect of leadership training overall, larger effects on learning than on performance, and an economic return that is unclear. Evidence-based development has to start from that humility.
Why intuition feels objective and usually is not
Intuition-based leadership decisions rarely feel like guesses to the people making them. A senior leader who has promoted people successfully before believes their read on the next candidate is sound judgement earned through experience. Sometimes it is. Often it carries the usual hazards of judgement: favouring people who resemble us, overweighting recent and vivid performance, and mistaking a confident communication style for capability. I cannot hand you a measured rate for those in your company, and I would distrust anyone who did. What I can say is that those hazards are hard to see from inside the decision.
What counts as evidence in evidence-based leadership development
Evidence is not a personality score treated as gospel, and it is not a single 360 survey, though both can be legitimate inputs. It is a combination of signals that converge, repeatedly and over time, on a consistent picture. One strong data point is an anecdote. Three or four independent signals pointing the same way over a meaningful period are evidence.
The most useful way to grade a signal is the way evaluation research does. The CIPD review describes the four Kirkpatrick levels: reaction, learning, behaviour and results. Geerts's review divides each level further into subjective and objective, where objective means verified by external sources, tangible results or statistics. The table applies that grading to the signals a leadership team can actually collect.
| Evidence type | What it can show | Its weakness | Cost to collect |
|---|---|---|---|
| Participant satisfaction | That the experience was well run | Says nothing about behaviour; easy to please without changing anything | Very low: a short survey |
| Knowledge or skills test | That something was learned | Learning effects are larger than performance effects in the CIPD review, and knowledge fades | Low |
| Multi-point 360 or upward feedback | Whether others see behaviour change over time | Subjective; one reading is a snapshot | Moderate: raters' time at each point |
| Decision records and outcomes | Decision quality against the information available at the time | Needs a record kept at the time of the decision | Moderate: discipline more than money |
| Pressure-tested behaviour | How the person acts under real difficulty | Hard to observe; depends on who is watching | Moderate to high |
| Promotion and retention data | Whether development shows up in careers and stay rates | Slow; confounded by many things besides development | Low if the records exist |
| Business results | Whether the capability moved a measure the business cares about | Hardest to attribute to development alone | High |
SHRM's 2025 guide to measuring return makes a practical point about cost: build data collection into the design early, use data you already hold in the performance system or 360 assessments where you can, and use an anonymous survey if cost is a factor. It also quotes the ROI Institute's Patti Phillips saying that a negative ROI does not mean you stop the programme but does mean you improve it. I would treat a weak first reading the same way: as a design input.
Four signals that should converge
- Decision Quality: Evaluated against the reasoning and information available at the time, not just whether the outcome worked out.
- Pathway Progression: Measured by a decision category now genuinely owned, not by training hours attended.
- Multi-Point Feedback: Gathered at more than one moment, so no single strong or weak period dominates the picture.
- Pressure-Tested Behaviour: Observed under real difficulty, not in calm, low-stakes conditions where almost anyone looks capable.
Convergence is the distinguishing feature, not any single signal. A business moving toward evidence can still get this wrong by treating one impressive data point, a glowing 360 or a strong quarter, as proof. The discipline is to require the same conclusion across several independent sources before trusting it.
What I do in practice, from baseline to second reading
With one executive, the starting point was a 360 assessment plus interviews with the stakeholders around the executive, and a short list of observable behaviours we agreed would change. At 12 months, the 360 was run again. Most behaviours scored better among the executive's direct reports, and the CEO endorsed the shift in revenue uplift, growth and business performance.
I am careful about causation, and you should be too. Agree with the sponsor in advance what would count as progress, track the agreed measures each quarter alongside a short stakeholder check-in, and do not claim that the coaching or the programme alone caused any business result. That is the standard I would ask of any provider, including me: a baseline, behaviours named in advance, the same raters twice, and modest claims about cause. Geerts's review recommends the same rhythm of data at the outset, throughout, at the end and six to nine months later.
How strong is the evidence behind the tools you use?
Be as sceptical of the instruments as of the intuition. Cornell's 2023 benchmarking study of how companies measure leadership capabilities found that all 14 benchmarked companies use a leadership measurement tool, with a mix of 180 and 360 degree feedback, and that some use the output as an input to performance and promotion decisions while others use it purely for development. That is a choice with consequences, and it should be made on purpose.
The Center for Creative Leadership's Leadership Gap Indicator is a useful illustration of the other side. The CCL guide says the tool diagnoses development needs in aggregate and is not appropriate for selection, compensation or performance appraisal. A group instrument used on named individuals for a promotion decision is being asked to do something it was not built for. Check what each tool in your framework was designed to support before you let it carry a decision. The CCL guide is explicit about this limit, and so is the Cornell benchmark about how long scaling a measure takes: most companies took at least a year.
The risk of overcorrecting into false precision
There is a trap on the other side of this shift. Evidence-based does not mean reducing a person to a single composite score and treating it as true because it came from a system and not a conversation. A poorly built framework can produce false precision: a number that looks scientific but rests on a handful of weak or biased inputs added together without scrutiny. That can be worse than honest intuition, because the number's apparent objectivity makes it harder to challenge than a stated opinion. Genuine evidence-based development stays honest about the quality and convergence of its inputs.
- Require convergence, not a single strong signal: Trust a conclusion only once several independent signals point the same way over time.
- Evaluate decisions against reasoning, not just outcomes: A good decision with a bad outcome and a bad decision with a lucky outcome look identical if you only track results.
- Gather feedback at more than one point in time: A single 360 captures one moment. Development evidence needs several points so an unusual week does not distort the picture.
- Stay honest about evidence quality: A composite score built on thin inputs is more dangerous than an honest statement that you are not sure yet.
Where manager assessments fit
In an evidence framework, I would treat the manager's assessment as one input among several, not the dominant one. A manager's read is shaped by proximity, recency and the working relationship, which is real information but is also exposed to the hazards described earlier. A single manager relationship is a sample of one. That does not make manager input worthless. It asks the manager's judgement to earn its weight by converging with other signals, the same standard applied to everything else. Managers who have built genuine trust in their read of their people often find this uncomfortable, and it deserves a straight conversation, not a dismissal.
Build the framework without overbuilding it
A framework so elaborate that it collapses under its own weight produces analysis paralysis. Start narrow: one or two high-stakes decision categories, such as promotion into the executive layer or selection for a critical successor role, and not every people decision at once. Proving the approach on a bounded, high-value slice builds the practical muscle and the organisational trust to extend it. A narrow start also keeps the false-precision risk manageable, because it is far harder to aggregate thin inputs into a confident score when a small number of consequential decisions is being watched closely.
The framework does not need to be perfect before it starts changing decisions. Even a narrow version, tracking decision quality and multi-point feedback for one high-stakes role, produces a better decision than pure intuition, and it improves through use as gaps in what is tracked become visible in practice.
How promotion conversations change
The shift shows in how promotion decisions are discussed. An intuition-based conversation sounds like a character reference: strong presence, clearly ready, the room just knows. An evidence-based one is more specific. It might point to the pattern across the person's recent material decisions, describe how their feedback has moved across several measurement points, and cite a concrete example of how they handled a difficult situation under pressure. That specificity is more defensible, both to the candidate who wants to understand what drove the decision and to anyone reviewing it later.
Experienced intuition that can point to specific evidence when asked is exactly what this approach is trying to produce more of. Experienced intuition that cannot, however confidently delivered, is the pattern it is trying to move away from.
A test for your own decisions
Evidence-based leadership development is not about removing human judgement. It is about requiring judgement to point at converging, checkable signals, graded honestly for how strong they are. Some of your past confident calls may not survive being asked to show their evidence. That discomfort is a sign the approach is working. For where to begin, a leadership assessment gives you a baseline. CapabilityAI, the AI leadership companion, is structured around tracking those signals over time. The Architecture Accelerator is where I build the framework with a leadership team.
Evidence is a pattern across independent sources, graded for strength. A single confident voice, however experienced, is a hypothesis.