Closing the Action Gap

The modern practice of employee engagement rests on a genuine intellectual achievement. Kahn’s (1990) foundational account of personal engagement established that people bring, or withhold, their full selves at work moment to moment, as a function of psychological meaningfulness, safety, and availability. The measurement industry that followed operationalized this insight at scale, and gave HR a shared vocabulary and a defensible number for something that had previously been managed by intuition.

That paradigm has a structural ceiling, and this audience knows exactly where it sits. Three properties define it. It is aggregate: it reports the health of a department or an organization, not the trajectory of a named individual. It is state-based: it captures how people feel under present conditions, which is precisely the signal most sensitive to a good week or a bad one. And it is periodic: the canonical instrument is administered annually, occasionally quarterly, which means the data describing a person’s engagement is frequently months old by the time anyone reads it.

None of this is a flaw in the science. It is a flaw in the unit of analysis and the clock. An annual, aggregate, state-based measurement can tell you that engagement in a division fell six points. It cannot tell you which of the forty-three people in that division is most likely to resign in the next ninety days, why that particular person is at risk in terms specific to how they are built, or what their direct manager should say on Tuesday. The paradigm was designed to describe a population. Retention decisions are made one person at a time.

The reason the ceiling is frustrating is that the literature has already supplied everything needed to break through it. The knowledge is not speculative or emerging. It is decades old, replicated, and taught in every graduate program represented in this room.

Personality is a leading indicator, not a footnote. Where state engagement fluctuates, trait-level disposition is stable, and it predicts behavior that engagement scores alone do not. In a meta-analytic path model, Zimmerman (2008) found that personality traits exert direct effects on turnover intentions and actual turnover that are not fully mediated by job satisfaction or performance. Emotional stability was the strongest negative predictor of the intention to quit; conscientiousness and agreeableness best predicted whether an employee actually left. Barrick and Mount (1991) established the broader predictive validity of the five-factor structure for work outcomes a generation ago. The practical implication is direct: a person’s standing on the Big Five is information about their retention risk that no engagement survey captures, and it is available before the risk materializes rather than after.

Self-determination theory explains the mechanism. Ryan and Deci (2000) demonstrated that motivation and well-being are governed by three basic psychological needs: autonomy, competence, and relatedness. When those needs are met, people internalize their work and persist. When they are thwarted, motivation degrades in patterned, diagnosable ways. This gives a why behind a risk score. Two employees can share an identical engagement number while one is starved of autonomy and the other of relatedness, and the correct intervention for each is different, sometimes opposite.

Job crafting and idiosyncratic deals supply the individual levers. Wrzesniewski and Dutton (2001) reframed employees as active crafters of their own roles, able to reshape task, relational, and cognitive boundaries to restore fit. Rousseau (2005) documented idiosyncratic deals, the personalized arrangements individuals negotiate that keep them without redefining policy for everyone. Together these establish that retention interventions do not have to be either generic or expensive. They can be specific, negotiated, and local to one person and one manager.

The job demands-resources model explains depletion. Demerouti, Bakker, Nachreiner, and Schaufeli (2001) showed that burnout arises from a specific imbalance: high demands draining energy faster than available resources replenish it. Bakker and Demerouti (2007) extended this into a general model of how strain and motivation coexist. Burnout, in this frame, is not a mood. It is a measurable equation, and its inputs are known.

Psychological safety governs whether any of this can be acted on. Edmondson (1999) demonstrated that the willingness to speak candidly and take interpersonal risk is a property of teams, and that it determines whether problems surface in time to be solved. This is the precondition that binds all of the above. A perfectly correct diagnosis is inert if the employee will not answer honestly, or if the manager cannot raise the subject without the conversation feeling like surveillance.

Set these findings side by side and a single conclusion is unavoidable. The field can identify who is at risk, at the individual level, using stable traits available in advance. It can explain why, in the mechanistic language of needs and demands. It can prescribe what to do, using levers that are individual and low-cost. And it knows the conditions under which the intervention will take.

What the field has not done, at scale and in the ordinary course of business, is assemble these pieces for one employee and put the result in the hands of the one person positioned to act, fast enough for it to matter.

That is the action gap. It is worth being precise about its anatomy, because naming it correctly is what makes it solvable:

• A synthesis gap. State data and trait data live in different instruments, owned by different functions, and are almost never fused into a single read of a single person.
• A routing gap. When a warning sign does surface, it typically travels to HR, is queued, generalized, and returned weeks later, if at all. The direct manager, who owns the relationship and the daily context, is the last to receive a plan rather than the first.
• A specificity gap. In the absence of an individual diagnosis, organizations default to generic remedies. The blanket retention bonus and the all-hands morale initiative are expensive precisely because they are aimed at everyone and therefore at no one.
• A time gap. Every week between signal and intervention is a week in which the decision to leave hardens. Retention is a perishable problem, and the annual clock guarantees the data is stale.

The cost of leaving this gap open is not abstract. SHRM places the cost of replacing an employee at roughly one-half to two times their annual salary once recruitment, lost productivity, and ramp time are counted. The Work Institute’s retention research estimates that on the order of three in four voluntary departures are preventable, and that voluntary turnover costs employers in the United States over 700 billion dollars a year. The failure, in other words, is not that we cannot see it coming. It is that seeing is not the same as intervening, and the machinery to convert one into the other has been missing.

If the gap is one of operationalization rather than knowledge, then closing it is an engineering problem with requirements that fall directly out of the science above. A system that closes the action gap must satisfy all of the following, not some.

1. Fuse state and trait at the level of the individual. Combine present-condition signals with stable dispositional data to produce a read of a named person, not a department average.
2. Be trustworthy enough to act on. Individual-level conclusions carry individual-level consequences. The analysis must be consistent and reproducible, not a different answer each time the same inputs are entered. A diagnosis that wobbles cannot support a decision about a person’s career.
3. Deliver to the manager, in the manager’s language. The output is not a dashboard for an analyst. It is a plan and the exact words for a conversation, placed with the direct supervisor who has the relationship and the authority to act.
4. Protect the honesty of the input. People disclose candidly only when they trust that raw disclosures will not be used against them. Confidentiality is not a compliance afterthought here. It is what makes the data valid in the first place, exactly as Edmondson’s account of safety predicts.
5. Keep a human in the loop, by design. The instrument informs a conversation between two people. It does not make the employment decision. This is both an ethical requirement and a practical one: the intervention that retains a person is relational, and it happens between a manager and an employee, not between an employee and a model.
6. Operate on a short clock. The value of the analysis decays with time. It must run in the ordinary rhythm of work and surface risk while there is still room to change the outcome.

Stated plainly, the field has spent decades describing this system in the negative, as the thing our surveys are not. The requirements are not novel. What has been missing is a system built to meet all six at once.

Satisfying those six requirements as a single working system does not call for a new theory of retention. It calls for a commercial embodiment of the existing one, assembled deliberately against the specification above.

Fusing state and trait. During onboarding, each employee completes a one-time personality assessment using an instrument such as the IPIP-60, a public-domain five-factor measure (Goldberg et al., 2006). On an ongoing cycle, the employee and the manager each contribute structured input. An analysis engine fuses these sources into a per-person read: a flight-risk window, a retention probability, the primary risk driver, a burnout and latent-burnout assessment grounded in the demands-resources logic, and the specific unmet need behind the number. This is the synthesis the paradigm never performed, carried out for one named person.

Trustworthy by construction. A system of this kind must separate a deterministic analytical core, which computes every number, flag, and ranking, from a generative layer, which writes the narrative and the conversation scripts. The consequence is that the analysis of identical inputs is reproducible: the same employee data yields the same verdict, every time. For a system whose outputs inform decisions about individuals, this reproducibility is not a technical nicety. It is the difference between a diagnosis a manager can stand behind and a number that drifts.

A plan, delivered to the manager. The output is not an organizational dashboard. For each at-risk employee, the direct manager receives a written brief in plain language: the risk, the reason, a recommended intervention drawn from an evidence-based library, the single next best action with a date attached, a thirty-day plan, and six ready conversation scripts. The interventions are the individual, low-cost levers the literature identifies, including job-crafting moves and idiosyncratic accommodations, rather than blanket programs. This is the routing gap closed by design: the person with the relationship gets the plan first, not last.

Individual, not generic. Because the analysis is per-person and need-specific, it replaces the expensive blanket remedy with the targeted one. The system’s operating principle is deliberately narrow: one employee, one analysis, one plan. The autonomy-starved employee and the relatedness-starved employee receive different briefs, because the science says their situations are different.

For a system that handles personality data and retention risk, governance is not a section at the end. It is load-bearing, and this audience is right to interrogate it first.

Confidentiality that protects the validity of the data. The requirement follows. An employee’s raw survey answers and personality scores must remain with the engine and reach no one. The manager never sees the underlying disclosures or the personality profile. They receive only the finished, actionable plan. This is designed against Edmondson’s principle: people answer honestly when honesty is safe, and the analysis is only as good as the candor of its inputs. Protecting the raw data is therefore not merely respectful. It is what makes the instrument work.

Non-punitive by design. The individual briefs are built to help a manager retain a person, and they are not repurposed to grade the manager or surfaced to corporate leadership as a performance verdict. Leadership sees aggregate exposure and trend, never the private contents of an individual’s brief or a manager’s intervention notes. The intent is to keep the tool on the side of the employee and the manager rather than turning it into an instrument of evaluation, which would corrode the trust the whole approach depends on.

A human decision, always. The design philosophy is informed by the direction of regulation in this area, including the principle in frameworks such as the GDPR and the emerging EU AI Act that consequential decisions about people should not be fully automated. The engine analyzes and recommends. It does not decide. The retention itself happens where it always has, in a candid conversation between a manager and an employee. The engine’s job is to make that conversation better informed, better timed, and more likely to happen at all.

The organizational sciences represented at this symposium have, over four decades, produced a nearly complete account of individual retention. They have identified the traits that anticipate departure, the needs whose frustration drives it, the levers that reverse it, and the conditions under which those levers can be pulled. The one thing the field has not routinely done is assemble that account for a single employee and deliver it to the single person who can act, before the window closes.

That is the action gap, and it has always been a gap in operationalization rather than in understanding. Closing it does not require a new theory of retention. It requires four things the field has largely treated as separate: a continuous read rather than an annual one, an anonymity architecture strong enough that honesty carries no risk, per-person synthesis rather than aggregate reporting, and delivery into the hands of the one person positioned to act. The dots were already on the page. The work is connecting them, in time to matter, one person at a time.