the job description is the wrong unit of analysis
job titles average away the variation that matters, and the decisions enterprises now face need a smaller unit than the job.
Ask a board a simple question about succession and watch what happens.
The question is this: what evidence do we have that this person is ready, and what would have shown us they were not?
Most succession packs cannot answer it. They contain names, boxes, colour codes and a readiness label. What they contain very little of is evidence. The Korn Ferry 2026 CEO and Board Survey put this to 250 board directors and chief executives at organisations from 1,000 to more than 20,000 employees. Half said succession planning started too late during their last CEO transition. Seventy per cent said they did not gather enough 360-degree feedback. Only 15 per cent said their organisation did a very strong job of preparing its first-time CEO.
Those are not planning failures. They are evidence failures. The plan existed. It was built on material that could not survive examination.
the nine-box produces confidence, not accuracy
The nine-box grid is close to universal and almost nobody defends it. Gallup's CHRO Roundtable surveys in 2023 found that 64 per cent of CHROs said their companies used the nine-box for succession planning. Only 9 per cent strongly agreed it was an effective approach for their organisation.
That gap between use and belief is worth sitting with. The grid persists because it produces a picture a room can agree on quickly, not because anyone thinks the picture is true.
The weakness is structural and it sits on the vertical axis. Performance has some observable anchor in outputs. Potential does not. It is a forecast, made by one or two people, about behaviour in a role the subject has never held.
Scullen, Mount and Goff examined what performance ratings measure, in the Journal of Applied Psychology in 2000. They took two large samples of managers, 2,350 and 2,142, each rated by multiple supervisors, peers and direct reports, and decomposed the variance. Idiosyncratic rater effects accounted for 62 per cent of rating variance in one sample and 53 per cent in the other. The ratee's actual performance accounted for 21 per cent and 25 per cent.
Read that as an engineering fact rather than a criticism of managers. When more than half the signal in a rating describes the rater, aggregating ratings into a grid does not average out the noise. It launders it into something that looks like a measurement.
potential ratings carry a documented bias
The problem is not only imprecision. Benson, Li and Shue analysed 29,809 management-track employees at a large retail company, published in the American Economic Review in 2026. Women received substantially lower potential ratings than men despite receiving higher performance ratings. Differences in potential ratings explained roughly half of the gender promotion gap.
The forecasts were also wrong. The women who received lower potential ratings went on to outperform their male colleagues, on average and at the margin of promotion. Even when a woman exceeded her own forecast, her next potential rating stayed low.
An unfalsifiable rating that is systematically biased and does not update on disconfirming evidence is not a talent signal. It is a preference, recorded in a grid.
what happens when slates are built this way
The downstream cost is measurable. Research by the Corporate Executive Board, reported by Jean Martin and Conrad Schmidt in Harvard Business Review in 2010, found that nearly 40 per cent of internal job moves made by people their companies had identified as high potentials ended in failure. The research base covered more than 20,000 designated emerging stars across more than 100 organisations over six years.
The same work found that only about 30 per cent of current high performers were, in fact, high potentials. Around 30 per cent had the ability and aspiration but lacked engagement, and had a 13 per cent chance of succeeding at the next level. A third had ability and engagement but did not aspire to the roles on offer, and had a 44 per cent chance.
Gallup's data on the adjacent decision is blunter. Organisations fail to select the candidate with the best talent fit for manager roles 82 per cent of the time. Only 3 per cent of large-company CHROs strongly agree their organisation excels at identifying and selecting manager candidates.
what a defensible readiness signal looks like
A readiness claim is defensible when it survives four tests. Apply them to any name before it reaches a slate.
- Scope. Has this person carried a decision at or near the scale of the target role? Name the decision, the budget or headcount or risk involved, and the date. Scale is the variable that most often breaks a promotion, and it is the easiest to check.
- Recency. Is the evidence from the last 24 months? Older evidence describes a person who no longer exists in a business that no longer exists.
- Independence. Do at least two sources support the claim, and do they report to different people? Given what the rater-variance research shows, two views from the same reporting line are close to one view.
- Falsifiability. What would have shown this person was not ready, and did that test occur? If the assignment could not have failed, it did not test anything.
The fourth test is the one that changes behaviour. Most development assignments are constructed to succeed. Martin and Schmidt found the strongest programmes did the opposite, placing people in what they called live-fire roles where new capability had to be acquired. Procter and Gamble identified the complex, high-impact positions in its Family Care division, named them crucible roles, and set out to fill 90 per cent of them with high potentials. More than 80 per cent of its high potentials became ready for critical leadership roles each year.
the readiness statement
Replace the box with one line per successor. This is the format that survives a board question, because it exposes its own weak points.
Name is ready in 0 to 12 months for role. Evidence: assignment, scope, date. Verified by: two independent sources. Open risk: the one capability not yet tested. Next test: assignment, by date.
Three rules keep it honest.
- Every ready-now name carries a written open risk. A slate with no open risks has not been examined. If nobody can name the untested capability, the assessment was a summary of the person's reputation.
- Three names per critical role, at least one from outside the incumbent's function. Pipelines built inside a function reproduce the incumbent. That is a design choice and it should be a conscious one.
- The potential score does not appear on the slate. Keep it in development conversations if it helps managers think. Do not let an unfalsifiable number carry a board decision.
rate the slate, not the person
Three slate-level measures tell a board more than any individual rating.
- Coverage. Share of critical roles with at least two names whose evidence passes all four tests. Not names on a list. Names with dossiers.
- Evidence age. The age of the oldest piece of evidence supporting any ready-now claim. This number quietly tells you when the last real assessment happened.
- Test rate. Share of successors who completed a named next test in the last twelve months. This is the only measure that shows whether the pipeline is moving or merely maintained.
The consequence of getting this wrong is well documented. PwC's Strategy& global CEO succession study found that companies forced to remove a CEO forgo an average of 1.8 billion dollars in shareholder value compared with those that plan the transition.
the practical obstacle
Nothing above requires a new assessment methodology. It requires evidence that is already scattered across project records, assignment histories, multi-rater feedback and performance data to be assembled per person, per claim, with dates attached. That assembly is the work, and it is the reason most organisations fall back on the grid. Building the dossier by hand for forty successors is a quarter of somebody's year.
That is the problem entomo's succession planning approach is built around: holding the evidence chain rather than the rating.
Boards are not asking for better ratings. They are asking a question ratings cannot answer. Build the slate so the answer is already written down.