AI and Operational Excellence

What Healthcare Needs to Learn from the Airline Industry, Waze, and the Ubers of the World

Key Takeaways:

• Healthcare can learn a great deal from how airlines, Waze, and Open Table manage supply and demand for operational excellence. These industries and platforms utilize AI/ML to match supply and demand, as well as link services across the customer journey. This could help healthcare reduce costs, increase profits, improve access to care, and reduce wait times. Scheduling and managing assets in healthcare currently are insufficient for improving access, cost structures, and revenue.
• EHR tools for scheduling will never be able to accomplish what focused, AI, or mathematical engines can provide for this challenge. Scheduling with EHR tools works well for conference rooms. But optimizing asset utilization–with regards to scheduling for operating rooms, infusion tables, and more–is an extremely complex challenge that requires sophisticated algorithms. Period.
• Healthcare organizations maintain a number of assumptions about the role of IT departments in innovation and digital transformation that block innovation and operational excellence. While HCOs tend to do a great job with clinical excellence, the role of IT in operational excellence and how they tend to think about innovation in siloed ways is often an obstacle. The collaboration of business leaders, clinical, and IT departments is necessary for operational excellence and maximizing ROI.

Introduction

Over the past five years we’ve all heard the refrain of “Where is the Uber for healthcare?” This often is used as shorthand for making healthcare more consumer friendly and app-driven. Data platforms could also be used to disrupt an industry all too often reliant on dated assumptions, paper-based transactions, and “consumer-driven” as a hollow marketing slogan. While this may be true, the discussion via this analogy of what is going on in these industries–and the potential improvement of healthcare for both providers and patients–merely scratches the surface. And is Uber the best point of comparison, or could it be the often-vilified airline industry? Or perhaps Waze, an app that many consumers use daily in normal times?

I am currently conducting the research for our upcoming report on AI and Operations, and recently had the opportunity to speak with Sanjeev Agarwal, the President and COO of LeanTaaS and co-author of Better Healthcare Through Math. His book is a must read for anyone interested in operational excellence, AIML, and the lessons we can learn from other industries, despite the incessant cries in some quarters that healthcare is different from every other industry. In our discussion, and in his book, we learn a great deal about how airlines are able to do scheduling in real-time while managing a quite broad range of services from baggage logistics to security, refueling, ticketing, and food, while managing shared resources such as runways.

When it comes to scheduling and optimizing utilization of expensive resources, one of the central operations issues in healthcare boils down to the entrenchment of traditional mindsets. These ways of thinking often prevent healthcare organizations from utilizing new tools, based in mathematics and AI, that can do a far better job of matching supply and demand. At the same time, these tools can also link resources with ancillary services connected to the patient journey (assets such as infusion tables, operating rooms, and hospital bed supply). For too long HCOs have invested in infrastructure like EHRs and not utilized more lightweight tools that can address these challenges. Agarwal and co-author Mohan Giridharadas joke that the aircraft carriers need the speedboats to win the war.

First Thesis: Systemic inefficiencies come down to a mismatch between supply and demand for every single appointment

Demand for appointments for utilization of assets such as operating rooms (ORs), infusion tables and beds tends to have an element of unpredictability. While there are obvious seasonal trends with influenza and patterns discernible, on any given day hospitals must deal with missed appointments and unexpected demand. Patients often need access to labs, out-patient facilities, pharmacies and a host of other assets that constitute the supply. This unpredictability and status quo methods of scheduling result in bottlenecks leading to customer dissatisfaction, physician burnout and lost profits for HCOs. 

The problem of asset supply in healthcare is also complex, and many assets such as MRIs are extremely expensive. Agarwal and Giridharadas note that even these expensive assets have inefficiencies and lost revenue potential, due to the typical Monday-Friday 9-5 scheduling practices that are dominant in healthcare. That means these assets are underutilized the majority of the time. Airlines, however, are loathe to let their assets go to waste, so they arrange scheduling to keep planes in the air and optimized for seating to buffer the dynamics of demand. 

If we take the example of infusion chairs for chemotherapy, and say an infusion center with 30 chairs doing 60-70 treatments per day with 4 different types of clusters of patients (different drugs require different amounts of time), statistically speaking, the number of possible calendar appointment permutations is an integer plus 100 zeroes behind it, they argue. This is not the work that an EHR block scheduling capability can handle. They not only lack the mathematics or AI engine to handle this challenge computationally but they lack the domain knowledge of operations and assets to optimize the scheduling. The result is many dashboards and reports and block scheduling that do not give insights on optimizing utilization of the infusion chairs, nor insights on why bottlenecks are occurring. The authors are very critical of the penchant for descriptive analytics associated with Tableau and Excel that fail to provide deeper analytics, which could reveal underlying causes and what can be done to address inefficiencies.

Beyond Block Scheduling

There is an important distinction between scheduling and optimization that healthcare tends to overlook. Scheduling is merely the act of putting an appointment on a calendar. In healthcare, this is the act of a patient calling the scheduling assistant, who checks the calendar for an open block. Done. The problem is that this results in the all too familiar long wait times and stumbling through ancillary services around the appointment that only aggravates patients and providers more. That is, it creates more barriers to access. 

Optimization is about utilizing the underlying intelligence–based on historical patterns and large datasets–composed of time stamps from appointments, and determining the best option for a particular type of patient, appointment and asset. Once again, the permutations of possibilities are immense. But this is about connecting volatile stochastic demand with stable assets and linking these dynamics with the other services, both shared and non-shared, to that particular appointment. Mathematics acts as the choreographer. 

This type of optimization is the field of yield management that airlines use to fill seats. Each service is a node in the network and patients flow from node to node. Optimization here is focused on understanding the highest utilization level for all nodes and avoiding gridlock. These services need to have buffers and operate below the maximum utilization so that flexibility and resilience are built into scheduling.  EHRs and their scheduling assistants simply cannot do this. The traditional “first come, first served” scheduling approach is appropriate for tennis courts and conference rooms, the authors argue, but not complex assets. They compare the traditional approach to having every pilot pre-book runway appointments in airports a week ahead of time. We know how that works, just as we often see in healthcare.

Sacred Cows Make the Best Burgers or How to BBQ Healthcare’s Sacred Cows

So why hasn’t healthcare followed the lead of other industries? We are now accustomed to the “healthcare is different” argument, and Agarwal and Giridharadas have a list of 9 sacred cows that healthcare needs to rid itself of to achieve operational excellence. In additions to the ones covered previously, the list goes on:

• Many organizations think that the IT team should take the lead on innovation or that because they invest in startups, the innovation challenges will take care of themselves. In reality, innovation in respect to operational excellence needs to be everyone’s job. Implementing AI solutions for operations, and many other functions, requires substantial change management to drive a successful digital transformation of healthcare.
• Investments in process management and Lean, Kaizen, Six-Sigma, etc. will improve operational excellence. The authors point out that consultants in this area are often skeptical of computational approaches. As we’ve seen above, if we look at scheduling from an optimization vantage point, the number of combinations of solutions is beyond the ability of the human mind to compute. Yes, there is a role for humans in AI, but we can’t overlook the work that AIML can do much more readily.
• Each project needs its own ROI attributed to that intervention. Often the causal connections are ambiguous or multi-factoral. The other issue is that ROI needs to go beyond just reducing costs and increasing revenue, but also extend to improved access to care, reduced wait times, and improved ability of independent physicians to work with a health system.

Throughout the last several months of interviewing leaders of vendors focused on AI and operations, I’ve been hearing similar insights as companies have had to confront the challenges of the pandemic and work to deploy these solutions under trying conditions. Prior to the pandemic, we heard a lot of talk about finding a single point to focus AI and digital transformation interventions and grow from this point. It is clear that the pandemic has forced many to think more broadly and look for ways to drive organizational change and AI adoption simultaneously in more ambitious ways; at least, in the operations area, where results can be seen more rapidly without the clinical risks faced by an area like clinical decision support. The insights that Agarwal and Giridharadas provide will be useful for those engaged in digital transformation of HCOs.

Conclusion

As mentioned in the introduction, we hear a lot about the need for learning from other industries in order to provide a better patient journey and increase the overall quality of healthcare. Less commonly discussed are the concrete steps and strategies to get there and what really is needed “under the hood” to transform the way healthcare works.  We also hear a lot of hubris about the magic that AI can bring to healthcare, which tends to feed the skepticism. But when we look at focused approaches to deploying AIML in use cases where results can be seen rather quickly, we begin to see the beginnings of what digital transformation can look like when AIML is used in smart ways. LeanTaaS is one of the companies we will be covering in our upcoming operations report, where we examine some of the leading vendors in this space, and how their solutions were deployed to address the challenges that the pandemic created for HCOs. We look forward to providing more insights from this research into how these solutions addressed the strains on systems, as well as bringing systems back to normality. We are excited to be sharing these insights over the summer as we release the report.