How a Leading Healthcare Company Pioneered the Future of AI Innovation

By harnessing cutting-edge AI technologies, the company has not only revolutionised patient care but also set a new standard for innovation in the healthcare industry.

Glenn Burgess

Blue Cross, Blue Shield of Michigan (BCBSM) transformed from being the least efficient in using technology within the Blue Cross system to becoming the most efficient, based on technology costs per employee.

They have adopted many new technologies, including cloud computing, security, mobile support, over 20 AI applications & now generative AI.

Ask any organisation about responsible AI, and you’ll probably hear about frameworks, guidelines, and principles. There are many high-level reports, but they often lack practical advice from top companies.

This leaves many businesses wondering: How is the work really done? And does responsible AI actually help organisations move faster, as many claim?

To find answers, it’s helpful to look at how one leading company took on a multi-year journey to upgrade their core systems. They did this to handle new challenges and customer needs, while also becoming more agile.

The Situation

Blue Cross, Blue Shield of Michigan (BCBSM) is a $36 billion healthcare provider serving over 5 million members with a network of 37,000 providers. Healthcare was always complicated, but it became more so in recent years.

The Affordable Care Act (ACA) increased healthcare access in the U.S. from 83% in 2010 to 93% in 2023. Along with an aging population, this put tremendous pressure on BCBSM.

Additionally, major changes in policies and services added new layers of complexity. For example, in 2022, McKinsey predicted that up to $265 billion in healthcare spending could shift from medical settings to homes by 2025.

Medicare Advantage, a fast-growing insurance product, now offers services like meal delivery, gym memberships, and access to healthy food through Uber Eats, all aimed at improving health outcomes and reducing costs.

Seven years ago, BCBSM’s leadership realised they needed to overhaul their outdated, rigid technology systems to manage this growing complexity without drastically increasing costs. As a nonprofit, BCBSM operates with very thin margins of 1% to 1.5%, unlike for-profit companies that aim for around 4%.

This meant that changes had to be made gradually and had to clearly add business value to justify the expense. With over 4 petabytes of mostly unstructured data, BCBSM saw opportunities for efficiency but couldn’t unlock that value with their old systems.

Moreover, much of this data contained sensitive patient information, subject to strict regulations & any mistakes could lead to severe fines or prison time.

Throughout this journey, BCBSM transformed from the least efficient technology user in the Blue Cross system to the most efficient, as measured by technology cost per employee.

They adopted new technologies like cloud computing, advanced security, mobile support, over 20 AI applications, and now generative AI. Today, they use three generative AI applications and plan to sell them to other companies, all while complying with more than 700 regulations.

In interviews with BCBSM’s management team, they highlighted seven key principles that guided their successful transformation.

1. Secure Board Support

Healthcare payor boards are very focused on following regulations and avoiding any mistakes, but they often aren’t fully updated on the latest AI developments. While many organisations have informed their boards about the potential and risks of generative AI, few offer regular educational sessions on this rapidly evolving field.

EVP/CIO Bill Fandrich frequently updates the BCBSM board on AI initiatives and ensures they receive ongoing education about AI, linking each small project’s results to the company’s larger business goals. Since AI is a broad and complex topic, it takes time for boards to fully grasp its potential & impact.

Fandrich and his team approach their work not just as a technology project, but as a capability-building effort.

For instance, in a meeting last year, they presented the guiding principles, policies, and controls they had established for AI use, and detailed the communication strategy they had implemented across the organisation.

In another meeting, they invited a guest speaker from a leading AI and cloud provider’s healthcare division to share insights & experiences. With the board now regularly engaged, Fandrich & his team have found it easier to provide updates, leading to quicker approvals for projects that need board consent.

2. Build Real and Regular Team Collaboration

Working together across different departments is crucial for using AI in the right way. Many companies plan to involve leaders from various areas to ensure responsible AI use, but this often doesn’t happen. Usually, IT and analytics end up leading the effort after the initial discussions.

Some companies focus AI projects on cutting costs, which often means reducing staff. However, BCBSM took a different approach. They worked on improving how different parts of their organisation worked together and aimed for ongoing innovation, not just short-term savings.

BCBSM’s team included members from compliance, analytics, IT, legal, business, and cybersecurity. They met weekly to oversee AI projects and make sure they followed guidelines and regulations. For example, they reviewed legal and compliance aspects for every important AI model.

This teamwork not only reduced AI risks but also helped BCBSM deliver AI solutions quickly. During the Covid-19 pandemic, for instance, they used this collaboration to create tools for predicting claims costs, patient risk, and disease spread in Michigan counties, all within two weeks. This allowed them to reach out to vulnerable planholders much faster than before.

3. Make Sure Access is Scalable and Secure

For healthcare payors, keeping data secure and compliant is crucial. BCBSM uses encryption, authentication, and role-based access controls to protect data. Since generative AI tools are new and lack mature access controls, BCBSM had to create its own security solutions instead of waiting for vendors.

To address this, BCBSM’s IT and analytics team collaborated with NASCO and Quantum Gears to develop SecureGPT. With the trust and resources invested over two years, this technology was built and widely used within BCBSM.

SecureGPT offers role-based access to large language models (LLMs), so, for example, contract managers only access LLMs related to contracts.

It also ensures conversations with LLMs stay on topic, prevents unethical content, logs all interactions for audits, and removes private information before queries reach public LLMs. This approach meets strict data privacy regulations like HIPAA and the Affordable Care Act.

4. Understand That Architecture is Very Important

To handle new services, clients, and processes, BCBSM had to go beyond automating old systems. They needed to create flexible, modular services to meet emerging needs and access data from large, siloed databases for their new AI models.

They developed a cloud-based data solution that integrates data from different business units through APIs and other methods. This setup allows for efficient, secure data management and supports quick AI deployment across various areas. This change was driven by their need to adapt to increasing market complexity and member demands.

Generative AI is enhancing this platform further. For example, information needed for benefits administration is spread across different units like provider network and claims processing. BCBSM used early AI models to understand connections between unstructured data.

For instance, AI could link “physical therapy” with “knee surgery” as related in post-operative care or find all documents related to “physical therapy,” even if they use different terms like “rehabilitation services.”

Newer AI models make it even easier to establish these connections, eliminating the need for multiple data management systems. The system learns relationships on its own, reducing costs, effort, and time needed to update and improve BCBSM’s solutions. This approach can benefit any organisation with important operational data spread across different systems.

5. Keep Educating Employees Regularly

Fandrich and his team saw their AI work as building an organisational capability, not just a technology project. Because AI is complex and rapidly evolving, they focused on ongoing employee education and training on responsible AI practices, as needs and threats change daily.

To be effective, they used a cross-functional AI leadership team to appoint representatives for employee education. They emphasised that responsible AI requires the same level of commitment and teamwork as cyberattack prevention. The leadership team consistently reinforces that adopting and managing AI responsibly is an ongoing effort, not a one-time task.

6. Tackle Bias Effectively

Insurers, especially health insurers, are experienced in handling bias in their products and services. In AI, bias can come from training data and model design. For instance, a model might misinterpret data about an ethnic group’s heart attack risk if that group has less access to care, not if they are less at risk.

To address this, BCBSM’s cross-functional AI team developed algorithms and procedures for testing and managing model bias. They regularly test for bias to ensure their AI systems are fair, equitable, and meet regulatory standards.

BCBSM also keeps an updated inventory of its AI systems, including model documentation, bias test results, change logs, and compliance records. This inventory helps carry forward lessons learned to new applications, speeding up innovation. The same AI team manages these processes.

7. Create a Safe Space for Innovation

In a company where decisions can impact life and death, like BCBSM, careful consideration is needed for where and how to innovate. Given their tight margins and operational demands, Fandrich and his team used the NASCO subsidiary to invest in new technology and test models on smaller datasets.

As Fandrich pointed out, other large insurers have faced high costs from cloud providers because they didn’t first test and optimise their models in a controlled environment like NASCO before full deployment.

BCBSM’s close relationship with NASCO helps them innovate quickly. By having this nearby technology hub, BCBSM ensures that applications are compliant from the start, avoiding costly retrofits that slow development.

This setup allows BCBSM to experiment and create new solutions without disrupting daily operations and provides funding for ongoing innovation. NASCO benefits by reselling these co-developed solutions, which also helps BCBSM save costs.

These practices have helped BCBSM shift from a traditional organisation to a nimble platform company that can handle the increasing complexity of healthcare and dynamic member needs. They have already used this innovative approach to successfully implement two generative AI applications in core business areas early in the technology’s development:

BenefitsGPT: This internal tool helps benefits analysts quickly search through unstructured data (like plan details, covered benefits, exclusions, and drug coverage) from various business functions. It aims to speed up response times and improve customer service for BCBSM’s 5.2 million planholders, reducing call times and increasing satisfaction. It should also cut down the time needed to set up benefits systems.

ContractsGPT: BCBSM uses thousands of contracts with healthcare and wellness providers. ContractsGPT allows analysts to search these contracts using natural language, making it easy to find terms, pricing, and other key details. It also helps identify efficiencies, such as finding overlaps in 30% of IT contracts, and suggests updates to contract passages as needed.

These tools not only benefit BCBSM but also other Blue Cross organisations, which can adopt them with confidence in their compliance.

Conclusion

Today, BCBSM stands out as both the most efficient tech user among its peers and a leader in innovative AI use. Their journey shows that by focusing on security, compliance, architecture & proactive governance, organisations can successfully innovate with generative AI while staying within regulatory limits.

BCBSM envisions a future with improved service and outcomes for patients through their increasingly intelligent AI-powered platform for services and payments. While the full scalability and future benefits of this model remain to be seen, other firms can learn how to responsibly implement AI in regulated environments from BCBSM’s approach.