Building a Leadership Team and Succession Bench

A fictional company runs its data platform through six directors who report to a VP of Data Platform. The director of platform reliability, the one who handles the cloud vendor, the security team, and every production-change exception, gives two weeks’ notice. The VP opens the succession file and finds a one-line entry for each director: “Ready in 1-2 years: a senior manager.” None of the named managers has led an incident that crossed team lines, approved a change exception, or defended a budget. The VP also notices something the departure did not cause. When the ingestion director missed a date that the ML platform director depended on, nobody told the ML director until the monthly operating review, and the VP resolved it alone, as she resolves every disagreement between directors. This invented case follows what she has to change: how directors are hired and developed, how they hold each other to commitments, how performance and succession are judged, and what evidence says that a person is ready for a larger role.

The post builds on leading through directors, which sets the decision rights, goals, and operating reviews that a VP uses to lead through them. Those arrangements assume a team that can fill them. Here the question is where that team comes from, how it works as a team and not as six separate reporting lines, and what happens when one of its members leaves.

Hiring a director is hiring for a role you have defined

The usual way to fill a director vacancy is to find the strongest person available and trust the conversation. The VP’s company did that the last time, when it hired an external director for ML platform after three well-liked interviews. The director left after eleven months. Nothing in the record shows a lack of talent. The company had never written down what the job required, so the interviews tested general impressiveness, and the director discovered the real job (arbitrating between platform teams, defending a shared budget to finance, being the named owner when a model-serving outage reached customers) only after starting.

A director’s job differs from a senior manager’s in specific ways. The scope changes from a team’s output to the performance of several teams through their managers, so the director’s results are largely other people’s results. Decisions cross team boundaries, which makes negotiation with peers part of the work. The director is the person executives hold answerable for a domain, and also the person who must hire, calibrate, and develop managers. A written role definition lists these duties and marks which ones the role cannot function without. It comes from the operating reviews and decision rights already in place, not from a generic job description, and it is the first input to hiring and later to the succession file. A career framework can supply common expectations by level, but someone still has to translate them into this domain’s version of the role.

With the duties written, the interview can test them. A structured interview asks every candidate the same prepared questions tied to the role, records detailed answers, and scores them against criteria that describe weak, borderline, solid, and outstanding responses. Google’s guide describes this approach and the training that makes interviewers use it consistently. A 2022 reanalysis of selection research by Sackett and colleagues, published in the Journal of Applied Psychology, ranked structured interviews as the strongest single predictor of job performance among the procedures it reviewed, while reducing earlier validity estimates; the authors also point to wide variation around the average. That is a reason to prefer the method and to keep judging it, not a guarantee for any one hire. For directors, add material drawn from the actual job: a written scenario based on a past operating review that the candidate must critique, or a conversation in which the candidate walks through a cross-team commitment that went wrong and what they changed afterward. Questions about past behavior need follow-ups that separate what the candidate decided from what the team did.

The panel matters as much as the questions. Peer directors belong on it because they will depend on the new hire’s commitments, and they can see whether the candidate would be an honest counterpart. They should score independently before they compare notes, so the first confident voice does not set everyone else’s score, and the VP should read disagreements as information about unclear criteria or missing evidence. Peer participation carries a cost: a panel of people who like the same style may screen out useful difference. Written criteria give the panel something to check that preference against.

Internal and external candidates present different risks. An internal candidate is known, but the VP is judging performance in a smaller job and may discount weaknesses she has already adapted around. An external candidate brings practices the team lacks, but arrives with no relationships, no context about unwritten commitments, and a first ninety days that must go to learning rather than leading. An external hire therefore needs an explicit start plan: which commitments transfer on day one, who is the interim contact for the vendor and security teams, which decisions the director may make alone from the first week, and when the VP will check the arrangement. Specifying those limits is itself a delegation level agreement, and it removes the ambiguity that new directors otherwise settle by guessing and learn about when a decision is reversed.

Developing the people who could become directors

Most directors are promoted from management, and the skills that earned the promotion are not the ones the new role requires. A strong senior manager owns outcomes through direct influence over a few teams. A director manages through other managers, spends time on commitments among peers, and handles budgets and executive communication. A person who has never practiced those things cannot be expected to perform them on the first day, and the move is where promotions fail.

Development therefore works best through real assignments with real stakes. The U.S. Government Accountability Office, writing about succession planning in the public sector, describes developmental or “stretch” assignments for high-potential employees as a complement to formal training. For this team, an assignment might be leading a cross-director initiative with a named executive sponsor, covering a director’s vacation with a written scope of authority, or presenting the platform’s investment case to finance with the director in the room but not in front. Each assignment needs a purpose stated in advance (which duty is being practiced), a person responsible for observing it, and feedback afterward. Without that, a stretch assignment is only extra work.

Directors also develop each other’s managers in a way a single VP cannot. A manager who spends a quarter embedded with another director’s organization learns how a neighboring domain makes trade-offs, which is exactly what a future director needs. This costs the host director capacity, so the VP has to account for it in the host director’s commitments rather than treat it as free. And development has to reflect what the person wants: a manager who prefers deep technical ownership to broader scope is better served by a path that rewards that, and pressing them toward a director role they did not choose produces a poor director and a lost specialist.

Directors who struggle need attention too. The underperformance guidance applies at this level with a twist: because a director’s results are mostly produced by others, a missed result might reflect the director’s choices, the structure the director inherited, or a priority the VP changed without saying so. Investigate which before concluding. The earlier posts on expectations and feedback and on underperformance and key-person dependency cover the conversation mechanics. What changes for directors is that a performance gap often shows up first as a pattern across the director’s managers: attrition on one team, a commitment missed twice, a manager who has stopped raising issues.

Making directors accountable to each other

In the fictional company, every disagreement between directors reached the VP, and she settled it. That arrangement feels like leadership, but it makes the VP the only route by which directors learn about each other’s problems. A director who misses a date has little reason to warn a peer in advance when the only consequence arrives through the VP at the next review. The VP’s own calendar then caps how quickly any cross-director problem can be resolved.

Peer accountability means that directors hold each other to commitments directly, and the VP intervenes when the peers cannot settle something, not before. It needs three things in place. First, commitments have to be explicit: who is promising what to whom, by when, and how the receiver will tell the promise was kept. A vague statement such as “ingestion will support the ML team’s launch” fails this test. Second, the commitments need to be visible to all directors, so that a miss is not a private matter between two of them. Third, directors need an agreed way to say that a commitment is at risk, early, without being penalized for the warning. Google’s research on its own teams treated psychological safety, the shared sense that it is safe to take interpersonal risks, as one of five dynamics of effective teams, alongside dependability, which it defines as members reliably completing quality work on time. A director team needs both. Psychological safety without dependable commitments produces pleasant meetings and missed dates. Dependability without safety produces hidden risk.

A commitments register is the simplest tool. The directors’ own forum reviews it, not just the VP’s operating review. Take a hypothetical quarter with 14 commitments among the six directors. Nine were delivered on time (9 of 14, or 64.3%). Five were late. Of the five late ones, three were flagged by the providing director before the due date, and two were discovered by the receiving director after the date passed. A lateness rate of 5 out of 14 says the team is unreliable. The split says something different: three of the five late items (60%) came with early warning, so the receivers could re-plan, while two did not. The team’s first problem is that two directors found out too late, so the register should track warning time alongside delivery. Measuring only lateness would punish the warnings and teach directors to stay quiet.

The VP’s role changes with this arrangement. She sets the decision rights that say what directors may settle between themselves (reordering a shared backlog, trading capacity for a quarter) and what they must bring to her (changes to a funded commitment, anything that alters risk the company has accepted). She asks, at the operating review, which commitments are at risk and who knew first, rather than which ones are late. When two directors deadlock, she asks them to bring options and a recommendation, not a dispute. And she models the behavior she wants: if her own commitment to a director slips, it goes on the register. A register that applies to everyone except the VP teaches the directors that it is for show.

Directors also need a collective responsibility that no one of them owns: the platform’s shared standards, the health of the leadership bench, the quality of promotion decisions across the organization. Calibration across the whole director team is the natural place to practice this. When directors compare how their managers are performing and who is ready for more, they learn to judge people outside their own organizations, and they become invested in the bench as the team’s responsibility. Agreement among directors does not prove the judgments are accurate. If they share a bias, such as undervaluing operational work, calibration repeats it with more confidence, so the criteria and the evidence need to be open to challenge.

Judging directors, and planning who comes next

A director’s performance has two parts that need separate evidence. Results are the outcomes the director’s domain was accountable for: reliability targets, delivery of funded commitments, cost against plan. Leadership is how those results were produced and whether they can be repeated: retention and growth among managers, quality of hiring, how commitments to peers were handled, how the director’s decisions held up. A director who hits every target by personally running every incident has achieved results that will disappear when she leaves. A rating that records only results rewards that pattern. The earlier post on performance management and who comes next explains how to build ratings from results, behaviors, and contribution, and the same structure applies here with higher stakes for the second part.

Succession planning starts from that assessment but asks a different question. The U.S. Office of Personnel Management describes workforce and succession planning as a systematic approach to determining staffing needs, analyzing the current workforce, addressing gaps in numbers and competencies, and building continuity and bench strength. GAO adds that leading organizations go beyond replacing individuals: planning should help an organization become what it needs to be, not recreate what it was. For the data platform that matters in concrete ways. If the platform expects its ML workload to grow and its ingestion work to be automated, the next ingestion director may need a different profile from the current one, and a succession file that simply names someone similar to the incumbent plans for the past.

GitLab’s public handbook offers a concrete vocabulary that shows how much a succession file can carry. For senior leaders it recommends succession planning after the performance and growth-potential assessment, and it notes that potential alone does not determine who will be a successor: strong candidates also need sustained exceeding performance and the aspiration for a larger role. It sorts readiness into four levels: an emergency successor who could cover roughly six months, someone ready now with about 90% of the competence and experience for the role, someone ready in one to two years, and someone ready in two to three years or more. Where no one could cover in an emergency, the plan should name another route, such as engaging a consultancy or splitting the team. GitLab also has leaders review succession plans with their peers and the people function. These are one company’s categories. The structure is portable even if the thresholds are not: each role has a named emergency cover, a candidate with a stated readiness level, a development plan with owners, and an answer to the question “if nobody here fits, what do we do?”.

The arithmetic of vacancies explains why a plan is worth having. Suppose, hypothetically, that each of the six directors has a 10% chance of leaving in a given year, independently of the others. The chance that at least one seat is vacated within a year is 1 minus 0.9 to the sixth power, which is 46.9%. Over two years, with twelve director-years at the same rate, it is 1 minus 0.9 to the twelfth power, 71.8%. The expected number of vacancies per year is 6 times 0.1, or 0.6. Real departures are not independent (a reorganization or a competitor’s hiring can move several at once) and the 10% is an assumption, not a benchmark. The point is that a six-person team should expect a vacancy more often than a manager’s intuition says, so a bench is routine capacity planning and not insurance against a rare event.

The cost of having no one ready is also countable, again with hypothetical inputs. If an external search takes five months and the new director needs six months before holding the role’s commitments independently, the seat runs at reduced capacity for about eleven months (5 + 6). An internal successor who is ready with a one-month handover shortens that to a fraction. The external route is not wrong; it can bring skills the team lacks. But the company should choose it deliberately, not because no alternative was prepared.

Key-person risk at the director level

The bus factor asks how many people’s absence would disrupt defined work. Applied to engineers, the dependency is mostly knowledge and access, and documentation plus a trained second operator reduce it. Directors concentrate additional kinds of dependency that documentation does not transfer. Relationships are one: the reliability director’s standing with the security team and the cloud vendor existed because of years of shared incidents, and nobody else had a comparable standing. Delegated authority is another: she could approve production-change exceptions because the VP had told security she could. A successor does not inherit that automatically; someone has to grant it again, and the people who relied on it have to be told. Unwritten commitments are a third. Promises made in hallway conversations disappear when the person who made them leaves, and the people who relied on them find out when the promises fail.

GitLab’s handbook defines a single point of failure as a person whose failure or absence would cause significant operational disruption, prevent critical work, or lose knowledge that others cannot readily cover, and it warns that being one does not make someone Key Talent. The reverse also holds, and the distinction changes decisions. A strong director who is a single point of failure needs both retention and a funded plan to reduce the dependency. A solid director who is a single point of failure still needs the dependency addressed, without any claim about their potential. A strong director who is not a single point of failure needs growth opportunities. Treating “dependency” as a compliment, or as a reason to deny someone a promotion because the team cannot spare them, damages trust in both directions.

For each director role, the VP can list the dependencies explicitly and assign an owner to each: a second named contact for each external relationship, a written record of delegated authorities with who granted them, a log of cross-team commitments, and a deputy who runs the director’s forum seat and decisions during planned absences. The check that matters is a test, not a document. A planned two-week absence, announced in advance, with the deputy making the decisions the director normally makes, shows quickly which dependencies are real. The absence should leave the escalation path clear for real emergencies. The result is a list of what failed and who fixes it, not a rating of the director.

What counts as evidence of readiness

“Ready in 1-2 years” in the VP’s file was an opinion, and nothing in it could be checked. Readiness evidence is a record of observed performance of the duties the target role requires, gathered in a way another person can examine. Four properties make evidence usable. It is relevant: the situation resembled the target role’s scope, not the candidate’s current one. It is recent, because skills and context change. It is observed by someone other than the candidate, and ideally by more than one person, including someone who is not the candidate’s champion. And it recurs, since one good incident does not show a pattern.

Return to the reliability director role. The VP’s written definition lists eight duties: leading an incident that crosses team lines, owning and defending a budget, hiring and calibrating managers, negotiating commitments with peer directors, presenting trade-offs to executives, managing the cloud vendor relationship, approving production-change exceptions, and coaching a manager through a performance problem. She marks two as gating for this role, incident leadership and change-exception authority, because a director without them cannot run reliability at all. She then records what each of two internal candidates has actually shown in the past twelve months, with the observer. The table gives the lookup.

Duty	Gating	Candidate A (incident response manager)	Candidate B (warehouse platform manager)
Lead a cross-team incident	Yes	Shown: three incidents, observed by two directors	Not shown
Own and defend a budget	No	Not shown	Shown: finance review, observed by the VP
Hire and calibrate managers	No	Shown: two hires, one calibration	Shown: three hires, two calibrations
Negotiate commitments with peers	No	Shown: two items on the register	Shown: four items on the register
Present trade-offs to executives	No	Shown: one session, with the VP present	Shown: two sessions
Manage the cloud vendor relationship	No	Not shown	Shown: renewal negotiation
Approve production-change exceptions	Yes	Not shown	Not shown
Coach a manager through a performance problem	No	Shown: one case	Shown: one case

Candidate A has shown five of the eight duties (5/8 = 62.5%). Candidate B has shown six (6/8 = 75%). A coverage percentage would rank B higher. The gates reverse that: A has shown one of the two gating duties and lacks the other, while B has shown neither. A’s gap is specific and closable. The VP can grant A supervised exception authority for a defined period, with a named reviewer and criteria for when the supervision ends. B’s gap is two duties in a field B has not worked in, which is a longer development path, though B’s budget and vendor experience may suit a different seat. Coverage counts are useful for finding gaps, and the gates decide readiness. A percentage threshold applied to the average, such as GitLab’s “ready now” description of about 90% of the competence and experience, still requires the role-specific judgment of which missing 10% matters.

Two cautions keep the evidence honest. Evidence gathered only when the candidate is being considered is shaped by the consideration: people behave differently when they know they are observed for promotion, and the observers know whom they are evaluating. Collect it over time, through the ordinary work of the operating reviews and the register, not through a special audition. And a plan that depends on a person staying is itself a risk, since a developed successor can leave. The VP therefore reviews the bench on a regular cadence and whenever a role changes, a candidate leaves, or the strategy shifts what the role needs.

The VP, a year later

In the fictional company, the VP’s immediate problem was the open reliability seat. She did not wait for the search. She asked candidate A to cover the seat for six months under a written delegation: incident leadership as the main duty, change-exception approval with a named reviewer for the first ninety days, the vendor and security relationships shared with a second contact on each side. In parallel she opened an external search, using the written role definition and a structured interview with a panel that included two directors, as a hedge in case A’s evidence did not hold. The decision on A was set to follow the first ninety days of observed exception decisions.

For the other five seats she asked each director to write a succession entry in the format the evidence table suggests: the role’s duties, the gates, up to two candidates with dated, observed evidence, an emergency cover, and a plan for each gap. She reviewed them in a session with all six directors, so every director had seen every other director’s bench. Two directors found that their bench was a single name. One discovered that his deputy for planned absences had never made a decision alone. Another realized that his best manager was a single point of failure for a pipeline, and that the manager’s departure would have damaged the platform more than his own.

She also moved the commitments register into the directors’ own forum, shifting the first agenda item from lateness to risk. Within two quarters the directors began bringing items to each other earlier and settling reorderings between themselves. The VP’s calendar held fewer disputes, but the useful change was in what the register showed: warnings had become the normal way a commitment was changed. Whether the new reliability director will succeed is not yet known. What the company has now is a way to find out sooner: a written role, evidence collected against it, and peers who can see whether the work is being done.

References

• GitLab Handbook, Talent Assessment (succession planning, readiness definitions, key talent and single point of failure).
• US Office of Personnel Management, What is Workforce and Succession Planning?
• US Government Accountability Office, Human Capital: Succession Planning and Management Is Critical (GAO-04-127T)
• Google re:Work, Understand team effectiveness
• Google re:Work, A guide to structured interviewing for better hiring practices
• HumRRO, Is cognitive ability the best predictor of job performance? (summary of Sackett, Zhang, Berry, and Lievens, Journal of Applied Psychology, 2022)