AI Symptom Triage and Digital Transformation at Harborview Clinics
Harborview Clinics is a network of 12 outpatient clinics with 300 clinicians. Average wait for a routine appointment is 45 minutes, and the call centre handles 4,000 calls a day. The executive team wants to deploy a vendor's AI symptom checker on the public website, claiming 92% triage accuracy in a vendor study. The tool recommends a care level but does not explain its reasoning. Clinicians are worried about bias, missed urgent cases, and liability; the data protection officer wants human oversight and clear consent. The regulator has asked for evidence of ongoing monitoring. The vendor offers a three-month pilot at low cost, and the marketing director wants to announce it as 'AI-powered healthcare transformation'. There is no agreed process for logging or reviewing safety incidents.
1. Distinguish digitization from digital transformation in this case. What would make the symptom checker genuinely transformative rather than a new front end?
Read the worked answer
Digitization would put the existing paper or telephone triage form online. Digital transformation would rethink how patients access care, with technology enabling new outcomes and roles. The symptom checker is transformative only if it safely changes triage and access in a measurable way-for example, directing patients to the right care level faster, reducing unnecessary calls, and freeing clinicians for higher-risk work-not because it is labelled AI. That requires integration with booking and clinical records, a human escalation path, consent, clinician review, and ongoing safety monitoring. It also changes the call-centre role and patient expectations. Without those changes, it is a website feature bolted onto the existing service. The executive team should define the intended outcome and risk appetite before treating the vendor pilot as transformation.
2. Identify the main risks and explain how you would adapt the VeriSM stages and governance to manage them.
Read the worked answer
Risks include unvalidated vendor accuracy, bias against underrepresented groups, lack of explainability, missed urgent cases, consent and data protection, clinician liability, incident blindness, and reputational damage from overclaiming. Adapt Define to agree outcomes, risk appetite, regulatory constraints, and stop criteria. Adapt Produce to require independent validation, bias testing, a clinical safety case, integration with booking, and a human-escalation path. Adapt Provide to cover consent, patient support, clinician review, and incident logging. Adapt Respond to monitor safety events, override rates, complaints, and model drift. Governance must include clinical governance, data protection, ethics, and vendor risk, not just IT. Start with assisted triage under clinician oversight rather than autonomous recommendations, and do not announce transformation before the pilot has evidence.
3. Propose a measurement and monitoring approach for the three-month pilot. What would cause you to stop or change the pilot?
Read the worked answer
Agree stop criteria before launch: any serious safety event, an unexplained accuracy drop, inability to escalate urgent cases, or evidence of discriminatory performance. Metrics should include appropriate referral rate against clinician review, urgent cases missed, clinician override rate, time to care, call-centre contacts avoided, patient understanding and trust, complaint themes, and safety incidents. Track equity across demographic groups and monitor data drift. A named clinical owner should review incidents on a fixed cadence, with records of decisions for the regulator. The service desk and risk functions need a route for safety concerns. If adoption is low but safety is sound, investigate usability and communication; if safety signals appear, pause. Marketing should describe a supervised pilot, not completed transformation. The pilot should end with a documented decision to scale, change, or stop.