In an era where healthcare systems globally are buckling under demographic pressures, workforce shortages, and spiraling costs, the promise of Artificial Intelligence (AI) often rings with a double-edged resonance. The specter of automation and job displacement looms, particularly in fields built on human compassion, like nursing. Yet, what if the most profound application of AI in healthcare isn’t about replacing the human touch, but fundamentally deepening it? What if technology could free clinicians from the shackles of administrative burden to reclaim the time and mental space for the very care they were trained to provide?

This is the core hypothesis behind a pioneering pilot at Sciensus, a European specialty healthcare provider. Moving beyond a simplistic narrative of efficiency-driven automation, Sciensus is trialing an ambient AI solution designed not to reduce headcount, but to radically enhance patient-centric care. The initiative, known as the CareTranscribe pilot, leverages Microsoft’s Dragon Copilot and Azure OpenAI services to securely capture ambient audio from clinician-patient home visits, transcribe conversations, and generate structured clinical notes. The goal is audacious: to capture the crucial 80% of patient experience—the fears, frustrations, family dynamics, and subtle sentiments—that is lost in mandatory checkbox forms, and use it to personalize support and improve long-term health outcomes.

To understand the philosophy and mechanics of this approach, Drug and Device World spoke with the minds spearheading the project – Christian Tucat, CEO of Sciensus, and Katie Duncalf, Director of Innovation and Market Strategy, and the project lead. They talk about their mission to integrate data and AI into the fabric of care delivery, how they are building adoption for the technology, and the long-term scalability of the project.

From Drug-Centric to Patient-Centric

The Sciensus pilot is not an isolated tech experiment; it is a tactical implementation of a broader strategic vision. As Christian Tucat outlines, the company sits at a critical “triangle between healthcare, technology, and the patients.” With a quarter of a million patients receiving complex treatments at home across Europe, Sciensus manages a vast, longitudinal dataset that extends far beyond prescription records. It encompasses the real-world experience of chronic illness managed in a personal environment.

“We are moving away from what some might call a drug-centric model to a patient-centric model and data-driven models of care,” explains Katie Duncalf. This evolution capitalizes on digital engagement and real-world evidence. The traditional model captures structured, clinical data—symptoms, vitals, and drug administration. The new model seeks to understand the holistic patient journey: the psychosocial barriers, the caregiver stress, the logistical hiccups, and the unvoiced anxieties that ultimately determine whether a patient adheres to a life-saving regimen.

Tucat frames the business and health imperative clearly: “If the patient stays longer on the treatments, they get better. The pharma company is able to sell more products. The healthcare system is able to keep the patients outside of the hospital… It’s a nice golden circle.” In this model, value is created not just through transactional drug delivery, but through demonstrably improving adherence and health outcomes. The ambient AI pilot is a key tool to unlock that value by revealing the ‘why’ behind adherence or dropout, enabling proactive, personalized intervention long before a crisis point.

The “Bottom-Up” Build

A critical differentiator of the Sciensus initiative, heavily emphasized by both leaders, is its development methodology. Rejecting a top-down, technology-first mandate, they adopted what Tucat terms a “bottom-up approach.” Before a single line of code was written, cross-functional teams immersed themselves in the organization’s daily realities.

“We went through the whole organization and sat down with the different teams,” says Tucat. They spoke to warehouse staff managing drug logistics, drivers facing delivery delays, and, most crucially, the nurses on the front line. The questions were simple yet profound: What are your biggest challenges? What do you spend too much time on? What do patients struggle with that we don’t currently capture?

This grassroots engagement served two vital purposes. First, it identified high-impact use cases where AI could genuinely alleviate pain points, such as handling routine patient queries or automating documentation. Second, and perhaps more importantly, it fostered a sense of ownership and buy-in from the very staff who would eventually use the technology. “This is their product today because they need to be comfortable with what’s going to happen,” Tucat states. This philosophy directly addresses a key failure mode in health tech: brilliant solutions that never get adopted because they were built in an ivory tower, alienating the end-user.

Duncalf confirms this, noting the pilot’s design is deeply integrated with clinical workflow. Weekly feedback sessions with nurses ensure the tool enhances, rather than impedes, their day-to-day responsibilities. This collaborative foundation is credited for the overwhelmingly positive initial reception from the nursing staff.

Capturing the “Glacier” of Unstructured Data

The technical heart of the pilot is its ability to convert unstructured conversation into structured, actionable insight. As Duncalf powerfully articulates, mandatory clinical forms capture merely “20% of all that patient information,” akin to the tip of an iceberg. “There’s still 80% of real patient insights… home environment barriers, family support dynamics, patient confidence levels, real concerns and anxieties.”

The AI’s role is to listen for these nuanced cues. Christian Tucat highlights patient and carer sentiment as a prime example. “Is that patient likely to be depressed or not? That is where the AI [is] able to capture… that’s very hard for a nurse speaking to a patient to pick that up unless it’s very obvious.” The technology can detect tonal shifts, hesitations, and concerns expressed not just by the patient, but by family members in the background—data points traditionally lost.

Duncalf provides a poignant fictitious example that brings this to life: a 54-year-old single mother who is unusually quiet and short during a visit. Through gentle probing, the nurse discovers the patient is hiding her diagnosis from her children. “That does not get captured in the mandatory clinical documentation,” Duncalf explains. “What that tells us is that there’s a way in which we can personalize care for this patient in providing more psychosocial support.” This depth of understanding allows Sciensus to move from reactive to proactive care, tailoring support services to the individual’s lived reality, not just their clinical diagnosis.

Navigating the Real-World Rollout

For all its technological sophistication, the pilot’s success hinges on human factors: nurse adoption and patient consent. Sciensus is implementing in Europe, “the highest regulated market in the world,” as Tucat points out, making GDPR-compliant consent paramount. The strategy has been transparency and clear communication, resulting in consent rates exceeding 90% in the pilot cohort.

Nurse adoption was cultivated from the start via the bottom-up design process. The value proposition to them is unambiguous: this tool reclaims time. “They didn’t train to fill up forms, they trained to look after patients,” says Tucat. The supplementary nurse testimonials validate this. Specialist Nurse Stewart Bower notes the technology “allows me to be more present with my patients… without unnecessary distractions.” Raquel Owens adds that it “will genuinely help our nurses spend more time being patient-focused.”

This human-centric messaging is vital. The AI is positioned as a “copilot” or “clinical support tool,” a concept reinforced in the technical background document. Nurses remain the ultimate authorities, validating all AI-generated notes. This creates a crucial “human firewall,” ensuring clinical responsibility stays with the practitioner while the AI handles administrative heavy lifting. Duncalf summarizes the ethos: “The future of AI in healthcare is about providing the opportunity for humans to do human activities.”

The Technical and Validation Framework

Trust in the system’s output is non-negotiable. The technical outline from the supplementary document details a multi-layered approach combining established and novel technologies. The pipeline begins with Microsoft Dragon Copilot, a medical-grade speech-to-text engine, ensuring an accurate initial transcription. This transcript is then processed by Azure OpenAI models, which are carefully prompted and “grounded” to perform specific tasks: topic extraction, contextual validation, and the generation of the structured Clinical Evaluation Form.

A critical design principle is that the AI models operate only on the supplied transcript. They are not generating inferred or fabricated information but are structuring what was actually said. This traceability back to the source audio is key for auditability and trust.

To validate accuracy, Sciensus has implemented a robust metrics framework. As Duncalf detailed, key performance indicators (KPIs) include:

• Data Capture & Transcription Accuracy: Nurses verify the transcript matches the actual visit.
• Technical Issue Frequency: Aiming for fewer than one issue per 50 visits.
• Usability & Adoption Metrics: Tracking consent rates and weekly nurse satisfaction surveys.

The plan includes “spot audits comparing transcript segments against generated outputs” to refine the AI’s performance continuously. This closed-loop system, where clinician feedback directly improves the model, ensures the tool evolves in step with real-world clinical need.

Beyond the Pilot

The potential ramifications of this pilot extend far beyond streamlined note-taking. Christian Tucat envisions a cascade of value across the healthcare ecosystem, ultimately influencing policy. “The longer-term vision is to… become standard practice,” he says, which “will completely change the way that patients who should be eligible for home care are treated.”

The value proposition shifts for each stakeholder:

• For Patients: More personalized, empathetic care and better health outcomes.
• For Nurses: Enhanced professional satisfaction and a return to core care duties.
• For Pharma Companies: Richer real-world evidence on adherence drivers, potential for regulatory-grade data to support value-based pricing, and more effective patient support programs.
• For Healthcare Systems (like the NHS): A powerful tool to alleviate systemic pressures, such as A&E surges and winter bed shortages, proving that technology-augmented home care can keep complex patients healthier at home is a compelling answer to the corridor care crisis. It transforms home care from an alternative into a superior, data-proven first-line strategy for chronic disease management.
• For Payers & Insurers: Reduced costs through fewer hospital readmissions and complications.

The pilot, developed in partnership with pharmaceutical sponsors, is already generating this kind of multi-stakeholder interest, according to Tucat, showing a value proposition for different stakeholders.

Long-term Plan

The Sciensus CareTranscribe pilot represents a nuanced, ethically grounded blueprint for AI in healthcare. It rejects the dystopian swap of human for machine and instead engineers a symbiotic partnership. By using ambient AI to listen deeply—to the patient’s words, tone, and unspoken context—it aims to fortify the human connection at the heart of healing.

The lessons are clear: successful health AI requires a grassroots, human-first development process; it must solve for clinician burnout as much as for cost; and its ultimate metric must be improved patient experience and outcome, not just algorithmic precision. As Katie Duncalf passionately concludes, this work gives patients a sense of being heard and gives nurses the gift of time. “Organizations that aren’t [integrating AI in this way] are falling behind,” she states.

The company’s longer-term ambition is to explore how this approach could be applied across different healthcare systems, recognizing that many face similar structural challenges. While the current pilot is focused on the UK, the learnings are intended to be transferable across markets over time. The pilot complies with applicable Class I medical-device standards. In later phases, Sciensus plans to explore extra features, such as leveraging Azure Machine Learning models to detect possible adverse events or adherence issues.