[[Audio 1]] [[DownloadsSidebar]]

As the healthcare industry continues to evolve, operations leaders face a complex set of challenges, including high administrative costs and employee attrition rates. Administrative accounts for about 25 percent of the more than $4 trillion spent on healthcare annually in the United States.[[footnote 1]] Simultaneously, in response to rising expectations, organizations face continued pressure to improve consumer experiences along the end-to-end healthcare journey.

Against this backdrop, advancements in AI, including generative AI (gen AI), could transform the healthcare industry, boosting operational efficiencies across internal and customer-facing operations at payers, care delivery organizations, and government entities such as the Centers for Medicare & Medicaid Services and public hospitals. In a 2023 survey of operations leaders in the customer care function, 45 percent cited deploying the latest technology, including AI, as a top priority, a 17-percentage-point increase from 2021.[[footnote 2]]

[[globalsidebar QB]]

In this article, we explore ways healthcare leaders could use AI to transform their service operations and also outline critical considerations that could help them succeed (see sidebar, “About QuantumBlack, AI by McKinsey”). Service operations encompass financial transactions (such as claims processing); industry-agnostic functions such as finance and human resources; industry-specific functions such as underwriting, enrollment, quality reporting, and accreditation; customer and patient services (the set of activities and processes that provide services to customers); and administrative clinical support functions (such as nursing administration and case management).

Why healthcare leaders struggle to realize digital investments’ full value potential

[[MostPopularArticles 5]]

Digital and AI solutions have substantial potential to optimize operations (for example, through automation) and enhance consumer experiences. Healthcare leaders are committed to investing in AI solutions such as chatbots, conversational AI, and virtual assistants to stay relevant and competitive.[[footnote 3]] However, technology transformation programs across industries, including healthcare, have historically failed to deliver value quickly enough to generate expected ROI; they typically realize less than a third of their expected value.[[footnote 4]] Moreover, only about 30 percent of large digital transformation efforts are successful.[[footnote 5]]

Specifically, operations leaders report difficulty in scaling AI and automation use cases from pilot to production; 25 percent of surveyed leaders indicated this is their biggest challenge.[[footnote 6]] For example, only 10 percent of respondents’ interactions with healthcare organizations’ conversational AI and chatbots fully resolved their queries without a subsequent need to interact with live agents.[[footnote 7]]

Mission-led road map. Leaders often lack a clear view of the potential value linked to business objectives as well as a road map to capture it. Perceptions of value may be overly optimistic, with leaders believing that AI solves all problems or provides a “quick win” compared with yearslong transformation programs. Conversely, leaders may underestimate the transformative potential of AI. Quantitatively and qualitatively measuring value (for example, using metrics for quality, safety, experience, and access) may be lacking, or the organization may not know which domains to prioritize. A detailed analysis of operations could uncover sizable delays in claims approval times, among other claims-related root causes of operational inefficiency, indicating that claims processing could be a high-priority domain for AI transformation.

Talent. Adopting AI requires a distinct set of skills and capabilities. Organizations may lack technical skills and a hiring plan to close gaps, and their upskilling and reskilling programs may lack investment and the right design. At the same time, however, many nontechnical employees in healthcare and other industries are already using gen AI as part of their everyday work and may have a clearer understanding of its value than employers realize.[[footnote 8]]

Leaders can use the adoption of gen AI as the impetus for assessing their overall talent strategies and what work employees consider meaningful and for crafting jobs that boost productivity while also putting people before tech.[[footnote 9]]

Agile delivery. To succeed with AI, organizations may need to accelerate their decision-making and delivery processes, which could entail shifts in funding and overcoming historical cultural norms and attitudes.

Technology and tools. Most healthcare organizations have legacy technology infrastructure, architectures, and tools that are difficult to scale to support AI solutions. For example, based on McKinsey benchmark analysis of healthcare companies, call center tools frequently lack the ability to tag reasons for calls with the level of detail needed to support AI and glean nuanced customer insights, with as much as 60 percent of calls untagged. Workforce management tools often lack advanced forecasting and scheduling capabilities, leading to mismatches between staffing and demand. Enhancing technology infrastructure is vital in addressing these limitations and improving overall efficiency, responsiveness, and service quality.

Data management. Healthcare data needed by AI solutions is heterogenous: unstructured, spread across multiple data sources, and stored in varying data structures. Organizations may lack data maturity—data completeness, data availability, capabilities to mitigate bias and risk, and data governance—needed to support AI solutions. After AI models have been launched, organizations may have difficulty adapting or integrating new data as it becomes available.

Additionally, ensuring compliance with strict privacy regulations—such as the Health Insurance Portability and Accountability Act in the United States and the General Data Protection Regulation in the European Union—is paramount. This effort includes effectively mitigating risks associated with handling sensitive data such as protected health information and personally identifiable information.

Consider a hypothetical health system with plans to implement AI-powered remote patient monitoring to predict and prevent potential health incidents. This use case relies on continuous data collection from various sensors and monitoring devices, such as wearables. The health system is challenged to ensure AI algorithms have access to sufficient data for effective learning and prediction while guaranteeing the anonymity of individual patients.

Changing the operating model. Successful leaders broadly consider AI’s impact on the operating model and on internal and external users. Workstreams dedicated to change management, communication, and training will likely be needed. Integrating AI insights into operations could entail changes to workflow, and leaders will need to ensure the AI algorithm’s outputs are transparent and able to be interpreted.

AI use cases in service operations

Private and public healthcare organizations are increasingly adopting AI to improve patient care and reduce costs (exhibit). Even so, to control risks associated with data privacy and security and ensure quality along with efficiency, most leaders prefer to use AI to augment human decision making rather than replace it entirely.

[[Exhibit 1a]] [[Exhibit 1b]]

Best practices for transforming service operations using AI

Leading organizations are using a set of best practices to transform their service operations with AI.

Prioritize key domains, and clarify their impact across use cases

Prioritizing service domains and defining clear AI use cases is a crucial early step. Some successful organizations have created a heat map to prioritize domains and use cases based on their potential impact (for example, increasing operational efficiency, enhancing customer experience, and supporting business objectives), feasibility to implement, and associated risks. Next, they design AI solutions to pursue high-priority use cases and identify any functional and technical needs and capabilities required to fill gaps. For example, an organization with the contact center as a priority domain could use machine learning algorithms to analyze data and identify drivers of incoming demand, customer sentiment, agent performance, and process breakpoints. Based on this analysis, an organization could identify a use case that would be best served by a customer-facing bot with real-time audio transcription and the ability to route to a human agent for any critical need. The features of the customer-facing bot can be mapped onto specific functional and architectural capabilities to ensure the AI system is designed to meet the specific needs of the use case and can deliver the desired outcomes.

Several AI use cases have proved especially relevant and effective in service operations and delivery. In all cases, minimizing the many risks of AI, including those associated with staff and patient concerns, is paramount to the effort.[[footnote 10]]

Hyperpersonalized customer touchpoints. About 75 percent of customers across industries now initially engage with an organization digitally and later go on to have an omnichannel experience.[[footnote 11]] This allows leading organizations to provide a hyperpersonalized experience. They are using AI to analyze customer data from multiple sources and generate personalized profiles of customers. This enables customers to engage through their channel of choice and increases the likelihood the organization can resolve a customer’s issues on the spot (without the help of a live agent).

For example, calls about claims and finding care account for about 50 to 70 percent of the total call volume at payer organizations, and a recent surge in billing errors now generates an additional 10 to 15 percent of calls to clarify explanation of benefits, according to McKinsey analysis. Using AI and voice analytics, payers could analyze millions of call recordings in real time, uncover detailed reasons for calls, and devise containment strategies such as more self-service options.

Conversational dialogue for resolving customer issues. Conversational AI bots could be used to resolve or more intelligently route issues of low to medium complexity. Supplementing transactional agent interactions with effective, empathetic virtual assistance AI bots, where appropriate, could enhance the customer experience, the quality of work and outcomes, and employee productivity. Additionally, skills-based analysis using AI tools and smart workflows could help route customers based on their needs to the most capable agent.

Consider the hypothetical case of a member who contacts a payer’s customer service department after being denied claims approval for a nonemergent ambulatory service, such as physical therapy. An AI bot could swiftly analyze claim details, patient history, and policy parameters and offer to send a preauthorization letter to the therapist. This empathetic and efficient virtual assistance enhances the customer experience and boosts employee productivity by managing routine inquiries, freeing human agents to focus on ensuring the quality of output and to devote their attention to more complex tasks.

Agent empowerment. Agent copilots (conversational interfaces that use large language models to support agents in real time) have the potential to help agents better understand customers and suggest responses based on their prior interactions. Gen AI could improve agent knowledge and adherence to processes by enabling them to access knowledge libraries with ease. Successful organizations use cutting-edge AI voice analytics to capture and summarize customer complaints and actions in real time.

In our analysis, about 30 to 40 percent of claims call handling time is dead air as agents search for information. Across industries, less experienced employees use twice as many knowledge resources, our analysis reveals, highlighting personal-coaching opportunities.[[footnote 12]] AI could be used to generate actionable insights into what helps or hinders the performance of the top- and bottom-quartile employees and offer personalized coaching for frontline agents (via AI-powered nudges) to improve performance. Furthermore, gen AI–powered virtual assistants could help agents handle inquiries more quickly and efficiently. These virtual assistants could help enhance agent responses by analyzing customer sentiment and providing suggestions or targeted prompts based on existing call transcripts and profile data. Virtual assistants could also improve customer sentiment by suggesting techniques to enhance their experience.

AI-enabled automation and planning processes. Healthcare organizations could pursue end-to-end smart-process automation for back-office processes and auditing of customer–agent interactions, as well as AI-enabled workforce management for agile forecasting and scheduling. For example, employees typically spend about 20 to 30 percent of their daily work hours on nonproductive activities such as administrative tasks and idle time, according to McKinsey analysis. AI-driven forecasting and schedule optimization embedded into existing tools could improve employee capacity management. The analysis further showed that by optimizing schedules through AI-enabled shift scheduling, organizations could increase occupancy rates[[footnote 13]] by 10 to 15 percent, improving overall efficiency, employee productivity, and job satisfaction.

Implement iterative test-and-learn approaches

Implementing AI requires an iterative test-and-learn approach. Using A/B testing for evaluating and refining the performance of different AI models and algorithms, organizations could quickly identify what works and what doesn’t and make necessary adjustments to improve the customer experience. This approach also helps minimize risks and optimize return on AI investments. For example, payers could employ A/B testing to evaluate different configurations of an AI-driven fraud detection model in the context of claims processing. Through systematic testing, they could quickly identify variations in model parameters to determine which algorithm is most adept at detecting fraudulent claims, allowing for timely interventions. This iterative approach not only strengthens the payer’s ability to safeguard against fraudulent activities but also helps streamline claims processing, thereby optimizing operational efficiency and minimizing financial risks.

Set up cross-functional teams

Evolving the operating model and setting up cross-functional teams (business, product, customer service, data and analytics, and IT) is critical for the successful implementation of AI use cases. These teams collaborate to understand and address customer care challenges and opportunities as well as the needs of the business. Cross-functional teams serve as early-adopter champions, shaping deployment of use cases, proving value to build momentum for change, and propelling adoption across the business. For example, in government organizations, a cross-functional team comprising policy makers, healthcare experts, IT specialists, and community representatives could collaborate to implement an AI-driven system for optimizing public health initiatives. This team would work together to address healthcare challenges, enhance patient services, and ensure that the AI solution aligns with the government’s healthcare objectives and citizen needs.

[[PullQuote 1]]

Design customer-backed experiences

Designing a customer-backed experience is essential for the successful implementation of AI in service operations. Leading payer organizations are using feedback and insights from customers to design AI solutions that can meet their needs and preferences. By using natural-language processing (NLP) and sentiment analysis, they could discern customer intent and emotions and provide personalized and contextualized responses. This approach helps improve customer satisfaction, loyalty, and advocacy, and it propels business growth and profitability. For example, a payer organization could use customer feedback and insights to implement AI-driven tools for claims processing. By employing NLP and sentiment analysis, the organization understands the concerns and sentiments expressed by customers in their inquiries. This allows for personalized responses, efficient issue resolution, and an overall improvement in customer satisfaction, contributing to increased loyalty and positive customer sentiment.

How service operations can get started on an AI road map

To initiate an AI-powered service operations transformation, organizations can consider taking the following actions today.

Conduct a rapid diagnostic assessment across operations

Using AI-powered tools such as process insights and speech analytics solutions, organizations could identify process inefficiencies and assess the potential value at stake of improvements (for example, across end-to-end customer care operations, including voice and nonvoice processes). This step can provide deep insights into current-state operations and customer needs and preferences, and it can help identify use cases to address gaps and opportunity areas. For instance, with advancements in AI technologies, depending upon the complexity of the use case, solutions can be implemented in weeks rather than months, allowing organizations to glean insights from day-to-day operations and customer interactions with speed.

Prioritize key domains, and develop initial proofs of concept for use cases

Next, the organization can develop an initial proof of concept for one or two use cases for the prioritized domains—such as gen AI–enabled copilot assistants or speech analytics to support claims queries in the call center—in which the pain is particularly acute and the potential for meaningful improvement is high. Doing so will help leaders understand the potential of these solutions and how they apply to customer care operations.

For example, healthcare payers frequently struggle to process claims with a high degree of complexity. AI-driven claims assistance solutions could help claims examiners make faster and more accurate claim adjudication decisions by suggesting appropriate payment actions and minimizing errors. These solutions could increase processing efficiency for complex claims by more than 30 percent and reduce penalties payers incur for failing to pay claims promptly.[[footnote 14]]

Scale additional use cases with an agile, iterative approach

Expanding the effort entails taking an agile, iterative approach to scale additional use cases and build a road map that clearly articulates solutions and priorities tied to specific owners and milestones. To help ensure successful scaling, organizations could form a cross-functional operating model (an AI task force) with representatives from the business, product, customer service, and IT teams and take steps to ensure it enables and supports continuous improvement.

Establish an effective governance framework

As companies move from use case pilots to mainstream adoption of AI and gen AI tools, they will need appropriate governance frameworks and comprehensive risk guidelines to maintain quality and manage risks at scale. These might include ongoing monitoring and auditing mechanisms to assess AI system behavior and ensure that it aligns with established ethical guidelines. Cross-functional teams of AI experts, ethicists, and legal advisers can evaluate AI models and applications for potential biases or ethical concerns. It is essential to identify, codify, and regularly review the ethical, legal, regulatory, and cyber risks of AI implementations. Defining explicit guidelines and policies to manage data, particularly sensitive information such as protected health information and personally identifiable information, can prevent unauthorized access and mishandling of data, including by third parties.

Align talent strategy with the AI and gen AI road map

Successful organizations will align their talent strategy with the use case road map and implementation approach, focusing on critical areas such as member services, claims processing, and policy development. They will set up cross-functional teams to incorporate new technologies and methodologies and ensure seamless integration of competencies, including advanced data analytics for risk assessment, machine learning for claims automation, and precision medicine for personalized policy offerings. To bridge skills gaps, successful organizations will implement targeted training programs and foster a culture of continuous learning and innovation.

The future of AI-enabled service operations for healthcare organizations such as payers, care delivery organizations, and the public sector is promising. Although many operations activities will still require a human touch, AI could help bolster efficiency by streamlining processes and supporting more convenient and personalized services for patients and customers. As the healthcare industry continues to evolve, the use of AI will become increasingly important to reimagining and continuously improving service operations.