Healthcare AI Consulting: What It Takes to Make AI Work

Challenges of Healthcare AI Consultation

Quality healthcare AI consultation does not simply prescribe measures; it tackles the relevant challenges and opportunities facing your organization. To this end, consider the following questions: Can your organization’s data enable predictive analytics, agentic workflows, or ambient documentation? Do your organization’s staff members know who will, and who will not, control the governance of AI policy? Where do your organization’s clinical leaders find common ground, regarding the use of AI? Which AI application has the lowest risk and reduces burden on staff the most? How do you shift ideas into practice, while maneuvering the organization’s holy policies, the operational policies, and the day-to-day practices?

These questions are important because most organizations do not have the luxury of starting with a clean slate. Most times, organizations are burdened with outdated technologies, data silos, definitional inconsistencies, and competing priorities. These factors generate policy and technological gridlock. In these contexts, the lack of AI is most often due to a lack of preparedness, and not a lack of aspiration.

Healthcare AI consulting will only succeed when the integration of technology, strategy, analytics, clinical feasibility, and the management of change within the organization is achieved. The absence of any of these factors, even the most technologically advanced solution, is more likely to fail than succeed.

Why Healthcare AI Consultation Values Readiness

AI health initiatives commonly stall during the pilot stage. Rapid AI adoption, without fully understanding the implications of data, workflows, and decision rights that would accommodate integration, leads to practically the same result. After extensive consideration of the tools, the integration of AI tools is most disrupted when there are issues related to the acceptance of the AI by clinicians and the quality of data, along with policy/system integration and the associated complexities.

Readiness denotes a more tangible state and encompasses fortified AI pilots that have been integrated with the many systems that comprise healthcare operations. Prompting AI integration and assuming a lack of readiness is merely a result of bureaucratic delays demonstrates a fundamental lack of understanding of the adoption of technology in healthcare operations.

At Hutchins, AI in healthcare consulting begins with an unembellished review of the people, processes, platforms, and policies in consideration of the conditions that allow the use of AI in healthcare to be both safe and effective. This includes the fidelity of the source systems, the clarity and definition of data, oversight, and the active commitment of the leadership in a given direction, as well as reliable workflows post integration.

Readiness involves more than just technology.

Many assume that the most critical component of AI readiness in healthcare is technology. While it most certainly is a dimension of readiness, there are many more challenges that arise within silos of teamwork. Models that are well conceptualized can fail to fulfill their potential if the oversight and policy teams do not have clarity of purpose regarding monitoring, do not have ownership regarding operational escalations, and do not communicate model trust to clinicians. The readiness of an organization is comprised of data trust, balance of oversight, a high degree of cross-sectional engagement, and the structuring of daily operations.

Building trust is about the integration of operations.

In the context of healthcare AI, trust is built when people have an understanding of how and where data is used, how human judgment is incorporated, and at what point potential adverse issues are brought to the attention of care and operations. Vague statements and broad descriptions of how things are done do not build trust. Therefore, impact does not build trust, but the integration of oversight within operations does.

Assessing AI Readiness: The Five Questions

The five questions articulated in plain and actionable terms, and without variance from one quarter to the next, represent the point of no return regarding AI readiness. The AI readiness tool was built around them as each question reveals a specific blockage preventing the adoption of AI post go live.

Can you stop a live model in less than twenty-four hours? The answer combines authority, procedure, and political reality. Most teams realize what they are up against when they find out for themselves what they have to do to consign the duty of stopping a live model to a series of approvals. This includes legal, IT operations, and the vendor contract, with no single role empowered to take action. The Emergency Care Research Institute, known as ECRI, has listed that inability among the many reasons that position AI at the top of the health technology concerns on their list.

Do you have validation data beyond what the vendor used in model training? Without local, held-out data, a health system cannot determine if a model serves the patients it was designed to serve. The Agency for Healthcare Research and Quality has justified such a separation as a baseline for clinical use. The HTI-1 rule of the Office of the National Coordinator mandates a description of the training data, allowing for local validation.

Does each model have a designated senior owner? If something goes wrong, the ownership of the committee comes to no ownership. The absence of a senior representative is cited by the Coalition for Health AI as the most frequent reason that AI committees fail to develop genuine authority in their initial year.

Has the board established a consistent quarterly rhythm for model performance evaluations? If reporting occurs only after a near miss, identifying a one-off occurrence versus a sustained pattern is impossible. The American Hospital Association 2024 Trustees Toolkit considers that reporting structure in the context of decisions that affect the extent of senior AI oversight.

Is there a workforce upskilling budget that is separate from the AI program budget? That budget line is a litmus test for whether an organization views AI as a tool to be implemented or as an innovation to be integrated.

A healthcare organization that can answer all five of these questions has likely done the foundational work needed to support the more advanced phases of AI. A healthcare organization that can answer two or fewer of these questions will likely experience the same problems from their next AI investment as they have from previous ones.

Who Owns the Model After Go-Live

An organizational structure designed for the age of the electronic health record is inappropriate for AI. The EHR is designed once, built to run. A model changes as the population, the documentation, the upstream data, and the vendor change. The chief information officer owns the infrastructure, the chief medical information officer owns the clinical informatics, the chief data officer owns the data; each of these roles intersects with parts of the model lifecycle and yet none of these roles owns the model lifecycle in its entirety. That is the space in which a model, once changed, continues to operate without anyone being able or willing to move, change, or reset it.

What is emerging is a single answerable seat, sometimes referred to as a Chief AI Officer, which takes ownership of portfolio strategy, pre-deployment validation on local data, drift monitoring and control, pause and failure terms, vendor workforce upskilling, and board presentation. The American Hospital Association identified that the largest remaining gap in health system AI oversight is senior AI ownership. Named ownership is where data governance in healthcare transforms from policy to operations. Our healthcare AI consulting work aims to articulate that ownership proactively; systems that define ownership in advance recover faster than systems that do so after harm is caused.

Healthcare AI consulting for health providers, payers and health innovators

All healthcare organizations must grapple with different dimensions of the AI challenge, but there are some commonalities. The need to develop integration and collaboration approaches is acute.

Providers and health systems

In the healthcare AI consulting space focused on providers, operational and clinical burdens are the main concerns. There are blocked flows in the emergency department, avoidable readmissions pose financial and clinical risks, and numerous clinical support tools that are negatively burdensome and poorly integrated are in place. The burden of documentation is increasing, resulting in a loss of time to care.

AI holds the potential to be revolutionary only when developed with a target audience in mind. Building pipelines that enable predictive analytics is one example. For ambient AI, this calls for developing intelligent systems for use in clinician-centered workflows to aid the processes of documentation and communication. The goal of AI in the workplace is to enhance the quality of workplace signals for time-sensitive decisions, thereby minimizing the burden of the process on the workplace, and not adding to the complexity of an overtaxed system.

Payers and Medicare Advantage organizations

Within the realm of healthcare AI consulting for payers, primary areas of concern include risk scoring and the management of utilization and of prior authorizations, providers, and populations. Currently, data systems that support modeling and scalable, robust, and justifiable predictions do not provide adequate governance and clear pathways for the models.

An AI strategy that is limited to the realm of analytics is insufficient. It must also include medical management and network strategy, and the integration of policy and audit requirements, and value-based care activities.

Health-tech companies

When these companies work on healthcare AI, they are enthusiastic about product visions though they may lack preparation for the actual healthcare market. While a solution may be highly valuable, provider and payer customers may have difficult questions, such as how the solution integrates with the real world, what type of oversight the customer should expect, and what data assumptions are built into the solution and are therefore likely to remain unobserved.

In this case, what they are seeking is not a consulting service that deals only with product strategy. There is the need to validate clinical flow, integrate with the market, implement responsible AI, and be operationally ready to move beyond pilots.

Investors and private equity teams

In the context of healthcare AI, investors may be seeking something different. In the case of an acquisition, investors want to be sure the AI capabilities will be strong and controllable. They want to see that companies have the technology and the supporting structures to allow the companies to thrive and grow beyond the acquisition.

What truly matters in healthcare AI consulting

Healthcare AI consulting should not be built around fads. It should be focused on frameworks that will help healthcare AI in the long run.

AI readiness benchmarking

To make adjustable investments, businesses and national structures require a detailed measurement of the readiness of their cadre, procedures, platforms, and policy frameworks. Self-assessment allows organizations to identify strengths and gaps and outline the necessary logical steps for the future.

Training predictive models

There are many opportunities to develop predictive modeling in analytics around capacity planning, readmission reduction, sepsis, care gap closure, etc. Predictive models are but one component in a system. Predictive analytics consulting addresses the entirety of the system and focuses on the pathway, input, and model validation and outlines operational guardrails and system integration.

Ambient AI scoping

There’s good reason for the growing popularity of Ambient AI. When correctly incorporated into a workflow, Ambient AI can ease the documentation burden. When incorrectly integrated, the Ambient AI model can, at best, add uncertainty to the oversight, quality, and responsibility of the decisions. AI consulting can help organizations determine the correct workflow integration for Ambient AI.

Responsible AI governance

For most, the success or failure of the strategy hinges on the governance and deployment of responsible AI. This goes beyond an expression of policy. It becomes ingrained in the organization. Who will have the power to approve use cases? What are the parameters of an approved use case that will instigate a governance review? Who will have the power to maintain oversight? Equity and safety checks of the model and the maintenance of oversight are crucial to governance. How oversight of the model is maintained will be a challenge to many organizations, and this will help determine the focus of the healthcare AI consulting most usefully to them.

The linkage of the data strategy and healthcare AI consulting

The failure to integrate data and AI strategies is a frequent cause of failed AI initiatives. The failure is particularly evident in healthcare.

The model will intrinsically reflect the data on which it is built. If data lineage is unclear, the trust in the data will be lost. If data governance is poor, the stewardship of the data will be reflected in the quality of the data. Ultimately, for AI to be successful, there has to be a sound data strategy, and this especially the case in healthcare AI consulting.

Organizations gain from defining AI as a horizontal enabling layer across the entire business. A vertical view focuses on separate systems within a department. Managers can use a horizontal view to create a foundational system for standardization, governance, integration of task management, and evaluation across the entire business.

What outcomes strong healthcare AI consulting can produce

The most productive healthcare AI consulting reduces the time and effort spent on decision making, oversight, and identifying and sequencing urgent actions. Clients are less likely to be distracted by the latest, interesting, and immature AI use cases. There will be a noticeable improvement in the quality of underlying data for AI and analytics. The gap will lessen between the output of AI and the subsequent action on that output.

The focus is on enhancing the speed of the enterprise, the agility of the AI, the resilience of the strategy, and the risk acceptance and management capability.

What to look for in a healthcare AI consulting partner

A viable healthcare AI consulting partner needs to be firmly grounded in the operational reality of healthcare. This means that the partner is able to make sense of highly complex and fragmented clinic pathways, healthcare, and control systems, as well as understand the innovative tools that tend to fail after a pilot.

A viable partner also needs to be able to tell a client that they are not yet ready for the next step. Responsible rapid prototyping lacks the same symptoms of the opposition to AI.

Our firm stands out for combining the unique depth of our 25 years of experience in health technology and systems transformation with a solid data strategy and AI frameworks for oversight, clinical integration, implementation, and policy planning and practice.

Healthcare AI Consulting Must Build the Basics

Before jumping into predictive analytics, ambient AI, oversight design, or AI road mapping, one must know the basics and not just market the latest or loudest tool. This entails knowing your data and workflows as well as your oversight frameworks, decision rights, and your ability and readiness to make the construct operational.

AI consulting in healthcare delivers the most value when grounded in the experience and knowledge of healthcare delivery and administration. This is the kind of support that Hutchins Data Strategy Consultants is enthusiastic to provide. We connect with a variety of health system stakeholders, especially providers, payers, health-tech companies, and funders, to solve AI and data readiness issues as well as develop data strategy and oversight, operational clinical integration and risk management, and strategy execution.

Authoritative sources

• NIST AI Risk Management Framework
• WHO: Ethics and governance of AI for health
• NEJM AI
• HIMSS