The 18 months of signals before the burnout cliff

The signals exist before the event

Founder and executive performance breakdown — the burnout cliff — is rarely a sudden event. It is the visible end of a period during which signals were already present, across physiology, behaviour and outcome, for many months before the cliff event registered with anyone in the room, including the operator at the centre of it.

The prospective evidence for the upstream period is now substantial. Vital exhaustion — Appels and Mulder's construct of unusual fatigue, irritability and demoralisation — predicts incident myocardial infarction at multi-year lead times in working populations. The PUMA Study cohort of 824 Danish human service workers found that burnout scores predicted both the duration and frequency of long-term sickness absence at three-year follow-up. The classical burnout developmental sequence — exhaustion, then cynicism, then reduced personal accomplishment — unfolds across 12 to 24 months when measured prospectively (Maslach & Leiter, 2016). Heart-rate variability, the most studied of the cardiovascular biomarkers, declines in step with rising burnout in cross-sectional and review evidence (Lennartsson et al., 2016; Immanuel et al., 2023).

The honest framing is a 12 to 18 month signal-to-failure envelope, with the strongest detectable cluster sitting at month minus twelve to month minus six. The signals are not uniform. They are a layered pattern across physiological, behavioural and outcome bands. The question, for the founders and boards reading them, is not whether the signals are there. The question is whether anyone is looking.

This is the upstream view of the pattern that the visible leader and the invisible cost compressed into a single image — the leader still performing while the cost accumulates beneath the surface. The territory this article works is the eighteen months that precede the moment that piece described — what we call shadow burnout in our practice: the mismatch between visible operating performance and the physiological cost accumulating beneath it.

The 18-month window in numbers

The strongest commercial argument for taking the window seriously is the size of what the cohorts produced. Toker and colleagues followed 8,838 employees over 3.4 years and found that elevated burnout was associated with a 40% increased risk of incident coronary heart disease; in the high-burnout subgroup, the increased risk reached 79%. The Appels and Mulder vital-exhaustion programme prospectively predicted myocardial infarction in 3,877 men at 4.2-year lead times; in women, the adjusted relative risk was 2.75. The PUMA Study found that burnout predicted both the duration and frequency of sickness-absence spells, with the stronger effect on duration — the people heading for the cliff lost more days when they did break, not just slightly more spells.

The 18-month framing is therefore not a single periodicity. It is a defensible operating envelope across multiple endpoints — cardiovascular, behavioural, occupational. Across the three endpoint classes, the strongest signals consistently cluster in the 12-to-6 month window before manifestation. The cliff is preceded by a corridor.

The operator cannot reliably detect themselves

If the signals are present, the natural question is why the cliff still arrives unannounced for so many founders. The answer is not effort or attention. The detection mechanism that would normally surface the signals is itself part of what depletion erodes.

Three things converge.

The first is interoceptive metacognition — the ability to accurately judge one's own internal physiological state. Sarah Garfinkel's three-axis model distinguishes interoceptive accuracy (objective performance) from interoceptive sensibility (self-reported confidence) and interoceptive awareness (the correspondence between the two). The recurring finding in this literature is that the people most confident about their internal-state self-detection tend to score lower on its actual accuracy. A subsequent misjudgment study of body-state self-detection found a Dunning-Kruger pattern: low-accuracy individuals were systematically more confident than high-accuracy ones.

The second is the executive Dunning-Kruger compounding. The corrective feedback that would normally calibrate self-perception thins as status rises. Direct reports defer; investors brief upward; the founder's environment selects for confirming signal. The decision-maker is not getting the data that would allow self-correction.

The third is identity. Concealable-stigma research has documented for decades that the disclosure cost of a hidden personal difficulty rises with status. For founders, the dynamic is more specific: the founder's identity is fused with the business — what we call the Founder's Trap — so admitting personal decline reads, to investors, to the board, to themselves, as admitting business decline. Disclosure becomes commercial risk. The empirical signature is unambiguous: 81% of founders hide their stress, fears and challenges from others; 77% do not seek qualified professional help; only 10% turn to investors when they are struggling (Startup Snapshot, 2023).

The combined effect is structural. A founder reporting "I know my limits, I know when I'm depleted" is providing precisely the signal that their objective detection is degraded.

The mechanism — error-monitoring failure plus compensatory neural recruitment — is the same one the architecture of the high-stakes decision examined in the context of single-day decision quality. Compounded across the eighteen months that precede a cliff event, the founder is not just making one degraded decision. They are accumulating physiological cost while the metacognitive instrument that would flag it is itself drifting out of calibration.

A methodological note worth making once: the founder concealment data sits across populations with a European weighting. Balderton's evidence base is UK and Europe; Startup Snapshot is largely Israel and Europe; the broader executive evidence (DDI's Global Leadership Forecast and the cardiovascular cohorts) is global. The convergence across these sources is robust, but the global generalisation should be made with the appropriate caveat.

The architectural answer — detection delegation

If the operator cannot reliably detect themselves, the structural answer is to delegate the detection. Not to abdicate it. To engineer it around the operator, drawing the architecture from professions that have already engineered fatigue-related decision failure out of their operating environments.

Aviation is the cleanest analogue. FAA 14 CFR Part 117 (2014) is built on the structural premise that crew fatigue cannot be left to operator self-report. The regulation combines mandatory fitness-for-duty self-affirmation, hard scheduling rules (flight duty periods capped at 9–14 hours depending on time of day and number of flight segments, built around the scientific finding that alertness degrades after roughly 16 hours of continuous wakefulness), Fatigue Risk Management Systems with biomathematical models of fatigue accumulation, and data-driven monitoring at the operator level. The premise is that detection is a system function, not an individual one.

Elite sport has formalised the same principle around physiological load. HRV-guided training programmes and the acute:chronic workload ratio — keeping training load in roughly the 0.8 to 1.3 range against a rolling four-week baseline — are used to anticipate injury and overload risk before it manifests. Coaches do not ask elite athletes whether they feel ready to train hard. They look at the data.

Applied to founder operating environments, the principle generates a four-layer architecture. The framing is structural; the layers do different work and operate at different cadences.

The point is not that any single layer is sufficient. It is that the four layers operate at different time horizons against different signal classes, and that none of them depends on the operator being the one to raise the flag.

Detection as governance

The reframe matters. A detection architecture engineered around the founder is not a wellness check. It is a structural performance discipline — the operating-environment counterpart to the corporate governance the same boards apply to financial control.

A founder approaching the cliff is a key-person risk. The boards and investors who underwrote the business have a duty of care to surface that risk before it becomes a discontinuity event. The Borritz, Toker and Appels cohorts are the empirical underwriting for what most boards already intuit: that the cost of a founder cliff is paid in dilution, key-customer loss, capital-event delay and the cost of replacement leadership. The cost arithmetic favours earlier detection.

The commercial framing is therefore defensible against an evidence base that most other advisory propositions cannot draw on. The engagement model that follows from it is structural, not motivational — operating-environment design rather than coaching.

The harder question

The architecture works. The signals are detectable. The cohort evidence is large, prospective and convergent across endpoints. The structural premise that the operator cannot reliably do the detecting is not contentious in the literature. It is contested only in the operating cultures where its acceptance would change the decision-rights of detection.

The question for founders, boards and investors is therefore not whether to believe the evidence. It is whether to engineer the operating environment around it.

The cliff is not the place to discover the answer. The eighteen months that precede it are.