Abstract
Healthcare organizations face persistent tension between efficiency and human connection. Digital systems often fragment attention and unintentionally distance staff from patients. At Mayo Clinic, we developed a conversational artificial intelligence (AI) agent embedded in Microsoft Teams to address this challenge. Connecting and Communicating (CoCo) provides staff with communication guidance, drawing on institutional resources aligned with the Mayo Model of Communication (MMOC). Built in Microsoft Copilot Studio and shaped by clinicians and operational stakeholders, CoCo was designed to be values-aligned and usable within existing workflows. CoCo is an active, user-initiated tool used primarily before or after challenging interactions to support preparation, reflection, and communication planning. Staff valued CoCo as a “just-in-time coach” that reinforced empathy while reducing stress. From August 2025 through February 2026, CoCo supported 1,903 conversation sessions, with early descriptive analytics suggesting favorable satisfaction and response-quality signals. We share insights from development and deployment, along with practical recommendations, to support other organizations considering bespoke, values-aligned AI tools to enhance patient and staff experience. We distill practice-based implementation lessons around governance, human factors, and change management, and offer practical recommendations to inform similar efforts elsewhere.
Introduction
In healthcare, human connection is a clinical necessity, with research linking empathic communication with improved patient outcomes, greater trust, reduced anxiety, and better adherence to care plans.1-3 Communicating with clarity and compassion supports teamwork, well-being, and job satisfaction. Yet despite its importance, interpersonal communication receives less real-time reinforcement than technical or procedural skills. Most communication training occurs in structured environments, far from the urgency, emotionality, and unpredictability of daily clinical practice.
Modern health systems are increasingly fast-paced and resource-constrained. In this environment, artificial intelligence (AI) has been rapidly adopted to streamline operations, boost efficiency, support diagnosis, and even automate note-writing. These advances have brought benefits, but they can also unintentionally distance patients from the personal connection that defines healing.4-6
As organizations strive to optimize operations, technology risks making care feel transactional and impersonal, leaving patients feeling alienated from their care team. Time constraints, emotional demands, and varying levels of experience can also make it difficult to maintain the warmth and purposefulness that patients and colleagues expect. Consequently, staff may struggle to recall best practices for patient-centered communication. Healthcare organizations have made significant investments in communication training but staff also need on-demand tools that reinforce these skills in practice. 7 , 8
At Mayo Clinic, we recognized this challenge. We wondered if AI could be used, not as a substitute for human care, but as a tool to nurture genuine connection. This question inspired CoCo, an AI-powered virtual assistant designed to reinforce evidence-based communication practices near the moment of need. Rather than adding to the proliferation of chatbots for patients, CoCo acts as a digital coach for staff and learners, offering language tips, reflective prompts, and conversation frameworks grounded in the Mayo Model of Communication.
This manuscript relates how CoCo was developed, how it works, and what we have learned from early beta testing. It is presented as an illustrative example offering practical, transferable lessons for organizations implementing AI-enabled support tools in healthcare settings.
Our Solution: CoCo
The Experience Training, Education and Coaching (XTEC) team in the Office of Patient Experience created CoCo to provide healthcare staff and learners with quick, helpful guidance on challenging communication moments. The idea was not to replace training programs or coaching but to augment learned skills with an easy-to-reach, just-in-time tool grounded in communication science. The name CoCo is a blend of the words Connecting and Communicating, chosen to reflect the assistant’s core purpose.
CoCo is a digital assistant focused on interpersonal communication, trained on the values and communication principles of our organization, and embedded directly into daily workflows. It is designed around the Mayo Model of Communication (MMOC), an evidence-based framework that emphasizes empathy, presence, and partnership in healthcare interactions. CoCo was trained on MMOC content to ensure its responses were consistent with Mayo Clinic’s evidence-based communication practices. 7 Selected peer-reviewed materials further enriched the knowledge base, ensuring accepted best practices in healthcare communication.5,6
Packard et al describe how MMOC was taught to frontline staff at Mayo Clinic using real-world scenarios, reflective listening, role-play, and large-group discussion helping staff recognize social cues and respond with congruent, compassionate communication, with reported improvements in empathy and patient experience scores. 7 CoCo draws on these same elements to guide its responses and recommendations.
Unlike traditional static guides, CoCo functions as a virtual agent offering contextual support tailored to the user’s immediate context. Unlike many AI tools in healthcare that focus on clinical data, administrative efficiency, or modelling predictions, CoCo centers on human experience. It supports refinement of the skills that shape how patients feel and how staff experience their day.
CoCo was embedded in Microsoft Teams (Microsoft Corporation, Redmond, WA), a cloud-based communication and collaboration platform already in daily use across Mayo Clinic. This allows staff to use it in a familiar environment without opening or managing a separate application. It was designed to be optional, private, and user-initiated, thereby minimizing disruption and encouraging reflective practice as an integrated part of the daily routine.
CoCo is intentionally active rather than passive. It does not listen to live conversations, monitor patient encounters, or provide unsolicited prompts. Staff initiate use by entering a prompt or selecting a scenario when they want communication support. Most use occurs outside live patient encounters, especially before anticipated difficult conversations or after emotionally challenging interactions. Thus, “just-in-time” refers to support that is available close to the moment of need, not real-time direction during patient care. CoCo does not make clinical decisions or replace professional judgment; it supports preparation, reflection, and adaptation to the user’s specific role and context.
Implementation: Building and Deploying CoCo
CoCo was created using Copilot Studio, a low-code platform that allowed rapid development while keeping design and content decisions close to operational stakeholders, rather than requiring a large technical build. 9 A base large language model is connected to a custom knowledge base, which in CoCo’s case consists of a document library containing curated XTEC content. By grounding responses in curated internal documents, CoCo can be updated through content governance rather than model retraining.
Several design principles shaped implementation: human-in-the-loop use, preparation and reflection rather than live patient-care direction, psychological safety in scenario-based coaching, grounding in institutionally approved communication standards, and expectation-setting that CoCo is a “thinking partner,” not an answer engine. Users are encouraged to adapt AI-generated suggestions rather than copy them verbatim. Selected training instructions are summarized in Appendix A.
A central implementation decision was to embed CoCo within tools already used daily. We connected CoCo to a secure SharePoint-based document library and deployed it as an application in Microsoft Teams. This approach reduced workflow disruption, improved accessibility, and simplified technical support by avoiding separate infrastructure, logins, or new training.
To make CoCo easy to adopt and use in day-to-day work, we created practical enablement resources focused on real-world use: • A self-paced online learning module • Quick-reference guides with tips for composing prompts. • In-application FAQs and guided support. • Teams support channel for questions, troubleshooting, and feedback.
How CoCo Fits Into Real-World Work
CoCo supports staff as they prepare for, reflect on, and improve communication with patients, families, and colleagues. Its most common use is before an anticipated interaction. For example, a nurse preparing for a challenging discharge conversation might ask CoCo for language that balances empathy with clear expectations. A clinician might request help responding to anger or frustration while maintaining boundaries. An administrative leader might use CoCo before a meeting to plan language that promotes psychological safety and respectful participation.
CoCo also supports scenario-based practice. Users can identify their role, select or describe a realistic communication challenge, and receive a simulated scenario with adjustable emotional intensity. They may then write a response and receive feedback aligned with the Mayo Model of Communication, including strengths, areas for improvement, and examples of more effective language. This allows staff to practice difficult conversations in a low-risk environment before encountering similar situations.
During live patient care, use is intentionally limited. CoCo is not designed to direct clinicians during patient encounters, and it does not passively listen, transcribe, or analyze live conversations. Some use may occur during operational or team-based work, such as drafting communication standards, establishing meeting norms, or facilitating inclusive discussion in a virtual meeting. After challenging interactions, staff may use CoCo for reflection: to consider what went well, identify what could have gone differently, connect the experience back to communication principles, or create a short personal reference for future situations. In all cases, use is optional, brief, and user-initiated, with the goal of reducing rather than adding to workload burden.
CoCo entered beta testing with voluntary users from the Interpersonal Communication Center, an audience already engaged in communication improvement and therefore well positioned to provide early feedback. We intentionally avoided a formal promotional campaign during the initial testing phase to observe organic uptake and identify usability and trust barriers. Despite limited outreach, early engagement was encouraging, and informal feedback suggested users felt comfortable accessing and using the assistant.
To support iterative improvement, we monitored descriptive usage and response-quality data through embedded Copilot Studio analytics. From August 2025 through February 2026, CoCo supported 1,903 conversation sessions. Sessions represent instances of use rather than unique users. During this period, the average satisfaction rating was 4.5 out of 5, although ratings were limited (n = 15). In a sampled quality review of 690 evaluated responses, more than 70% met predefined standards for relevance and usefulness. Lower-quality responses most often reflected gaps in the early organizational knowledge base during the evaluation period. At this stage, evaluation focuses on adoption, user satisfaction, response quality, and qualitative feedback rather than patient-level outcomes; future evaluation will assess perceived usefulness, changes in staff confidence, and potential downstream effects on communication quality and patient or staff experience. These early data should be interpreted as descriptive implementation signals rather than evidence of clinical or patient-experience outcomes. Evaluation methods continue to evolve as analytics tools mature, and future assessment will incorporate more customized measures of adoption, quality, and perceived usefulness.
Governance, Safety, and Privacy
CoCo’s content governance is organized around institutional communication standards. Knowledge sources are curated, reviewed, and updated by the XTEC team to maintain alignment with approved communication principles and organizational values. Because CoCo is grounded in an internal document library, content can be refined through knowledge-base updates rather than model retraining. This allows operational stakeholders to maintain oversight of the guidance CoCo provides.
CoCo was also developed with attention to responsible AI use. Training and presentations emphasize that CoCo is a support for thinking, not an authority. Staff remain responsible for reviewing AI-generated content for accuracy, appropriateness, tone, and potential bias before using it in patient-facing or colleague-facing communication. CoCo is not intended to provide clinical advice, make decisions, or replace professional judgment. Rather, it helps users prepare, reflect, and generate communication options for adaptation to context.
Privacy and confidentiality were central to the implementation approach. CoCo operates within Mayo Clinic’s enterprise Microsoft 365 environment under an institutional agreement with Microsoft. It is not designed to collect or store patient data, and patient information is not required for intended use. Users are instructed not to enter protected health information or other confidential information and to follow institutional privacy, security, and data-governance requirements. Review processes, including periodic review of available session logs, are being explored to support quality improvement, identify unintended or innovative use cases, and monitor for safety concerns.
Actionable Insights
Early experience with CoCo’s design, deployment, and adoption revealed several lessons. These insights can guide broader integration of AI-enabled communication support in healthcare settings. • • • • • • • • •
Practical Recommendations
For organizations considering similar AI-enabled communication tools, the following practices can increase success: 1. 2. 3. 4. 5. 6. 7.
Limitations and Risks
Several limitations should be considered. First, CoCo was developed within a large healthcare organization with an existing communication model, enterprise Microsoft 365 infrastructure, and a culture already invested in communication training. Findings may not generalize to smaller organizations, settings without comparable digital infrastructure, or environments where communication frameworks are less established. Second, early usage data are descriptive and should not be interpreted as evidence that CoCo improves patient outcomes, communication quality, or workforce well-being. Satisfaction ratings were favorable but limited by a small number of submitted ratings, and response-quality review focused on relevance and usefulness rather than downstream behavioral or patient-experience outcomes.
Third, AI-supported communication tools carry risks. Users may become over-reliant on suggested language, or communication may feel scripted if AI-generated wording is used without adaptation. Responses may vary as the knowledge base is expanded, prompts differ, or the underlying platform evolves. Although content governance and response review can mitigate these risks, they cannot eliminate the need for human judgment. CoCo is a support for reflection and communication planning, not as a substitute for empathy, relationship-building, or professional accountability.
Conclusion
CoCo illustrates how AI can reinforce, rather than replace, the human connection at the heart of healthcare. By embedding evidence-based communication practices into a familiar workflow, staff gained just-in-time support for the most challenging, but potentially rewarding, moments in their workdays. Early descriptive experience suggests that when AI tools are values-aligned, grounded in familiar frameworks, actively user-initiated, and designed with attention to workflow fit, they may support staff preparation, reflection, and confidence in challenging communication situations.
Our intent is not to suggest a one-size-fits-all solution; CoCo was developed for our local context. Our hope is that the experience offers a useful case example for examining common challenges such as workflow integration and change management, that could inform similar efforts elsewhere. Our implementation is still evolving, but the lessons learned point to a broader opportunity: healthcare organizations can build AI tools that reflect their own culture, priorities, and values. Importantly, CoCo is presented here as an illustrative case example to provide practical, transferable implementation lessons. Our experience suggests that the most consequential work was organizational rather than technical: clarifying scope and intended use, defining content, integrating into existing workflows, supporting adoption, establishing governance and responsible-use expectations, and iterative refinement based on early feedback. In doing so, technology can strengthen trust, empathy, and connection—the timeless foundations of healing.
Footnotes
Acknowledgments
The authors thank the Experience Training, Education and Coaching (XTEC) team for their contributions to the development of CoCo.
Ethical Considerations
The Mayo Clinic Institutional Review Board (IRB) waived IRB review of this project. (ID #24-002195).
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Use of Artificial Intelligence
Generative AI (ChatGPT, OpenAI) was used to assist in phrasing and clarity during manuscript preparation. This use was distinct from the CoCo system described in the article. All substantive content, analysis, and interpretation were generated and verified by the authors.