Five principles of gen AI for healthcare

New gen AI tools could transform the healthcare industry. But leaders need to make smart choices.

The arrival of generative AI presents an enormous opportunity for the healthcare industry to achieve some of its longest-sought goals, everything from improving access to care to bringing down costs.

For individual organizations, though, harnessing the power of this new technology means staying mindful to the unique set of pitfalls and possibilities that gen AI brings to this most unique and sensitive of sectors. Below, we outline five key principles for implementing gen AI in the healthcare space—a comprehensive framework to ensure the success, scalability, and sustainability of gen AI-powered healthcare initiatives.

Figure 1

As this chart makes clear, gen AI’s impact is pervasive, not disruptive. In almost every area of healthcare it can help improve the customer/employee experience, cut costs, and drive growth and profitability. Which brings us to our first principle…

The framework

1.    Portfolio management

Don’t think of gen AI as a standalone initiative, but as a way to expand and enhance the value of existing initiatives. By using new gen AI tools alongside existing capabilities and investments, including current AI/ML models, documents, and data, organizations can achieve better results, faster, while also continuing to build maturity across the entire tech portfolio.

For example, one key issue facing healthcare providers is the need to reduce churn rates among patients. Using traditional methods, this activity gathering data—the patient’s health information, previous treatment dates and contact details—and assembling it in a warehouse. A customer service representative then conducts outreach using a pre-written script or a set of analytical dashboards.

With gen AI, however, companies can reach a far higher level of nuance and personalization in those interactions. An AI-enabled tool can produce a detailed snapshot of each person and provide insights to the customer service agent to help anticipate and respond to the customer’s specific concerns: in the context of their current patient or member journey factoring in their local healthcare infrastructure and nuances of their health insurance coverage.

2.    IT operating model and architecture

This “progressive” quality of gen AI—meaning that it builds upon the organization’s foundational elements, including data and analytics capabilities, supporting processes, and human skills—makes it important to begin implementation with use cases that reflect the organization’s current level of digital maturity and that will allow the organization to naturally build skills and capabilities incrementally over time.

As gen AI plays a bigger role within the enterprise, it is also likely to impact the broader IT operating model and IT architecture. Therefore, when selecting a use case for gen AI implementation, careful consideration should be given to its potential impact on the organization's IT capabilities and broader ecosystem.

3.    Training and process design

Much has been made of the “training” process that lets large language models (LLM) work their magic, but the arrival of gen AI doesn’t mean you can neglect the training of human workers—quite the opposite, in fact. While generative AI tools can automate tasks like data collection, analysis, and script writing, people must still play an integral role in these processes, and be specially trained in the quirks, limitations and inaccuracies that still crop up in the output of gen AI models. Not only that, but existing training programs in pre-gen-AI skills may need to be ramped up, to ensure that workers maintain the necessary knowledge and expertise to complete tasks elsewhere in the workflow, which themselves should be reimagined with the incorporation of gen AI enabled agents.

To revisit our customer churn example: imagine an agent attempting to reduce churn contacts a patient and follows the personalized script proposed by the AI tool. What happens if the conversation veers off-script because a patient asks a question that the AI model did not consider? How do companies ensure that agents still maintain a core grasp on the subject matter if an AI tool is completing some of the groundwork that they had previously been responsible for?

Top-notch training is paramount, not to only to use this new technology, but to develop and maintain the critical thinking skills needed to evaluate the tool’s output and fill in the gaps were needed. Ultimately, achieving a seamless integration of gen AI into business processes requires a significant effort to ensure workers can effectively interpret and respond to its recommendations.

4.    Risk and change management

Data is the lifeblood of gen AI, and any business leveraging this new technology needs to be extra-mindful of risk management. That includes not only protecting the private information of patients, where privacy and security are of the utmost sensitivity, but also broader areas of operational, data and legal risk.

All of these security, privacy and ethical issues can at least be mitigated by putting in place the right system, operational and legal controls. But this needs to happen at the outset of every gen AI program, not after the fact. Don’t proceed until you have a comprehensive strategy that identifies key risk sources, assesses their potential impact on the business, and makes plans to implement mitigation measures. Given the evolving nature of gen AI—as well as a regulatory landscape that in many ways has not yet caught up with the technology—companies will need to constantly reappraise and readdress their risk strategy and approach. Some may find they need a completely new governance structure to oversee the responsible use of gen AI.

5.    Financials and outcome management

Generative AI holds enormous potential for healthcare organizations – but how do companies know if they’re maximizing the return on their investment?

To ensure companies are using gen AI to its full potential, we recommend developing an outcome management framework, which outlines processes and practices to guide the overall program, while establishing metrics and benchmarks to track progress, measure results and calculate ROI.

One of the key elements of such a framework is consistent measurement and metric analysis. For example, companies can use A/B testing, which lets teams compare AI-enabled workflows side-by-side with traditional processes, to capture and calculate productivity and efficiency gains, and measure the impact of gen AI tools on other phases of the workflow.

With many AI models, for example increasing the precision of a model may increase the accuracy of its output. But it may also increase complexity for human workers—for instance by producing too many options to evaluate—which can end up lowering productivity. A properly designed outcome management system can flag such unintended consequences and point the way toward resolution.

What’s your next step with generative AI?

For healthcare organizations, generative AI holds extraordinary potential. To unlock the power of this technology, companies need to take clear and decisive steps. By addressing these five issues consistently across all initiatives, organizations can ensure their gen AI programs are successful, scalable and sustainable.

For more insights, visit the Generative AI section of our website.