Faculty Lead
Gianluca Cesare
Group Chief Information Officer, Humanitas; Adjunct Professor, Humanitas University; participant in the AI-Driven Business Models (AIDBM) Executive Master, Università Cattolica del Sacro Cuore.
About this AI Insight
Humanitas Research Hospital set out to deploy an ambient AI Scribe across roughly one hundred Operating Units: microphones capture the physician–patient conversation and the system drafts the clinical record, so that the physician’s attention returns to the patient and the resulting structured documentation enriches secondary use of data for research.
A six-month pilot in onco-haematology – deliberately the hardest linguistic domain – cut visit time by 20% (from 30 to 24 minutes), reached a 100% patient-satisfaction score and measurably enriched the clinical record. Yet the enterprise rollout to some two thousand physicians is stalling, with many quietly reverting to the keyboard.
This AI Insight names the stall precisely and argues that the root cause is cultural rather than technical: the programme was governed as a technology deployment when it is, in substance, a redesign of the clinical method itself. The keyboard pause is not clerical waste but cognitive scaffolding – the reflective moment in which conversation becomes clinical reasoning – and scaling borrowed against an adaptive capacity that had never been repaid.
The prescribed remedy transfers ownership of the rollout from the Information Systems function to the clinical line: a Scribe Clinical Lead in every Operating Unit, a hybrid review-and-sign protocol, and three leading indicators – weekly active use, persistence of the time saving and silent reversion rate – that surface trouble while a wave is still recoverable. The case is a reminder that AI does not transform an organisation: leaders do.
Inside this Insight
A pilot that worked, a rollout that stalled
Humanitas Research Hospital set out to deploy an ambient AI Scribe across roughly one hundred Operating Units: microphones capture the physician–patient conversation and the system drafts the clinical record, reasoned and classified according to clinical standards. A six-month MVP in onco-haematology – deliberately the hardest linguistic domain – involved five physicians and ten nurses and delivered a 20% reduction in visit time, from 30 to 24 minutes, a 100% patient-satisfaction score in the post-visit survey, and a measurable enrichment of the clinical record for secondary research use. On the strength of those KPIs the hospital launched an enterprise rollout to about two thousand physicians within six months. That rollout is visibly stalling: adoption is slow and, in many units, physicians are abandoning the tool and reverting to the keyboard.
Naming the stall precisely
Cesare insists on precision because four of the five causes initially identified have already been addressed: sponsorship and budget were reconfirmed, leadership and project management strengthened, the copy-and-paste step eliminated by integrating the Scribe with the Electronic Health Record, and the medical libraries extended beyond onco-haematology. The technology works, the integration works, the governance is in place – and value has stopped accruing. The stall point is therefore this: the pilot cannot be scaled because it rests on a capability the operating model cannot absorb. The MVP measured the tool in the hands of five self-selected early adopters; the rollout demands a new clinical method – the narrated visit, in which the physician edits and signs rather than authors at the keyboard – from two thousand physicians who did not build it.
Why the root cause is cultural, not technical
The primary root cause, in the author’s diagnosis, is a change-management and adoption failure: the transformation was governed as a technology deployment when it is, in substance, a redesign of the clinical method itself. The moment a physician pauses at the computer to write the Objective Examination, the Anamnesis, the Diagnostic Assessment, the Therapy and the Prescription is not clerical waste to be automated away – it is cognitive scaffolding, the reflective pause in which conversation is converted into clinical reasoning. The Scribe automates the artefact, the note, but silently removes the scaffold. Seen this way, reversion is not resistance to technology but rational self-protection by professionals who experience the loss of their thinking time as a risk to clinical quality, for which they – not the algorithm – remain accountable. The paper sharpens this with the concept of transformation debt: the MVP borrowed against the adaptive capability of five volunteers, and scaling called that debt due at once, across two thousand physicians, inside a six-month blanket timetable that left no room to repay it.
The intervention: moving ownership into the clinic
Cesare prescribes a single organisational move – transferring ownership of the rollout from Information Systems to the clinical line by appointing in every Operating Unit a Scribe Clinical Lead, a practising, respected physician with explicit decision rights over the local visit protocol, coordinated in a federated Clinical Adoption Board chaired by the Medical Directorate. Three design elements make it concrete: decision rights, with each Lead licensed to sanction a hybrid protocol in which the Scribe drafts and the physician retains a structured two-to-three-minute review-and-sign pause, deliberately re-inserting the reflective moment inside the new method; incentives and accountability, with adoption on Unit Directors’ scorecards and the six saved minutes formally returned to physicians as research and teaching time rather than harvested as scheduling productivity; and operating rhythm, replacing the blanket six-month rollout with readiness-sequenced waves in which units are onboarded peer-to-peer by colleagues who already master the narrated visit.
Three indicators that reveal a silent stall
Because reversion to typing is silent – no physician files a complaint, they simply stop narrating – the intervention must be instrumented. Each Clinical Lead and the federated Board watch three leading indicators weekly, per Operating Unit rather than as a hospital-wide average that would mask a failing unit inside a healthy mean: weekly active use, the share of a unit’s physicians actually running a narrated visit; persistence of the time saving, whether the 30-to-24-minute reduction holds as the cohort widens beyond early adopters; and silent reversion rate, the proportion of visits that revert to keyboard-authored notes after a physician has started using the Scribe. A threshold logic makes these operational: a unit below an agreed floor on active use, or above an agreed ceiling on reversion, is flagged to its Clinical Lead for a peer conversation before the next wave. The paper also sets aside three plausible alternatives – a top-down mandate, a large centrally-run training programme, and extending the timeline – arguing each would defer or inflate the transformation debt rather than repay it. The closing claim: AI does not transform an organisation, leaders do; here, leadership has to be relocated into the clinic itself.
About the Author
Gianluca Cesare (Chrt Eng, MSc) is Group Chief Information Officer and Adjunct Professor at Humanitas University, where he leads digital transformation and the health information systems of one of Italy’s leading research and clinical hospital groups.
He has extensive experience in Digital Health, having led for over a decade one of the most relevant public projects in the field – the Regione Lombardia Electronic Health Record and the Patient Relationship Management services. He has collaborated with the European Commission and holds a chair on the EC Healthcare Cybersecurity board, and has worked with CONSIP and Accenture across pharma and life sciences, public government and durable goods. He holds an MSc in Logistics and Supply Chain Management from Cranfield School of Management and an engineering degree from the University of Catania.
Download Full Insight
Download the complete AIDBM Insight to see how a pilot with excellent KPIs can still stall at enterprise scale, why the reversion to the keyboard is rational rather than resistant, and which single organisational move – plus the three indicators that prove it is working – can unstick a real AI transformation.
Download Full Insight