Multimodal Clinical Triage Reliability Leader

The mandate

The clinical-AI reliability director left during a patient-safety review after the triage service assigned materially different urgency when the same presentation arrived as spoken narrative, typed symptoms or a wound image with partial history. A partner network paused two regional launches, and operating teams began applying local overrides that reduced consistency further while the review separated model behaviour from workflow and escalation failures.

The interim must start within three weeks and take the executive reliability seat for twelve months. Recruitment for a permanent director will open after the first two partner sites complete a stable ninety-day surveillance cycle in month six; the chosen successor receives six weeks of joint decision-making before the interim’s authority expires, whether or not wider commercial rollout has resumed.

Handover is complete when modality-specific safety cases define allowed and prohibited uses, triage calibration is evidenced across the agreed symptom and demographic cohorts, missing or contradictory inputs reliably force the correct human pathway, and all live sites use one incident taxonomy with weekly drift review. The successor must chair a simulated deterioration event, sign the residual-hazard register and accept the next two release decisions without support.

The interim may suspend a modality, narrow a pathway, mandate clinician review, quarantine a model version, appoint temporary investigation leads and deploy up to £6 million from the approved remediation reserve. Board approval is required to change clinical risk tolerance, resume a paused regional launch, notify outside parties beyond established protocols or exceed that reserve; permanent hiring and alteration of clinicians’ professional accountability are expressly withheld.

Replacement of the electronic health-record estate, redesign of clinician pay, consumer acquisition strategy and expansion into new countries are not part of the seat. The leader may require interfaces, staffing data or launch assumptions from those areas to test safety, but does not own diagnosis policy, medical-device classification strategy or the commercial roadmap unrelated to triage reliability.

Why this seat is open

The departure removed the only executive who connected model evaluation, clinical workflow and live incident surveillance across every modality. Keeping the role inside the existing product hierarchy would make the same launch pressure that contributed to fragmented overrides responsible for judging remediation. The board wants a time-bounded operator who can impose one evidential standard while a permanent appointment is made deliberately.

What you will own

• Establish a common hazard taxonomy linking modality failure, missing context, cohort effect, clinician interaction and patient consequence to one accountable disposition.
• Decide which symptom pathways may continue, require mandatory human confirmation or remain suspended, documenting the evidence and reversibility of each boundary.
• Rebuild test cohorts around clinically material presentations, skin tones, accents, languages, disabilities, device conditions and incomplete histories rather than headline accuracy alone.
• Calibrate urgency and escalation thresholds separately for text, speech and image inputs, then test discordant combinations that should trigger safe abstention.
• Institute live surveillance that joins model version, modality quality, clinician override, queue pressure, deterioration signal and eventual care outcome without concealing denominator shifts.
• Run partner-site readiness reviews in Manchester and Birmingham, closing local workarounds before authorising any controlled return to service.
• Transfer release records, open hazards, cohort baselines, incident simulations and authority maps through a successor-led committee decision observed by the board chair.

Candidate qualifications

• Held executive accountability for clinical safety or AI reliability in a regulated digital-health, urgent-care or diagnostic service used in live patient pathways.
• Personally paused or restricted an algorithmic workflow after detecting a safety signal and can evidence the investigation, containment decision and monitored restart.
• Evaluated systems combining language, speech and medical imagery, including contradictions and quality degradation between modalities rather than isolated benchmark performance.
• Built cohort-level calibration and surveillance that connected technical outputs to clinician action, escalation delay and clinically meaningful downstream outcomes.
• Directed incident review across product engineers, practising clinicians, human-factors specialists, quality leaders and external care partners with competing interpretations of causality.
• Handed a repaired safety operating model to a permanent leader after demonstrating that local sites could apply it without exceptional executive intervention.

Non-negotiables

• Available within three weeks for an exclusive twelve-month seat, based in London five working days a week and able to complete the stated Manchester and Birmingham travel.
• Holds current, verifiable clinical or clinical-safety credentials appropriate to challenging triage decisions and has governed deployed machine-learning systems directly.
• Will stop or narrow a pathway on evidence of material patient harm even when the decision delays contracted launches or revenue recognition.
• Has no undisclosed financial interest in the model vendor, clinical-assurance provider or partner network involved in the review.

1. 49 words maximum. State your notice position, earliest London start and ability to sustain the required regional travel for twelve months.
2. 49 words maximum. Describe a live clinical algorithm you paused or constrained, the signal that moved you to act and the evidence required for restart.
3. 49 words maximum. How did you test whether two input modalities changed triage safety for a clinically important cohort rather than merely changing aggregate accuracy?