For key lines of business such as private passenger auto and workers compensation, more than half of all premiums and claims liability are for payments to medical providers. With 80,000 diagnosis codes and thousands of procedure codes governing tens of billions of dollars in payments to medical providers for casualty, enormous complexity clouds the detail behind medical bills. And time is of the essence, manifested in state level regulations like California’s Labor Code § 4603.4 that requires payment to workers compensation providers within 15 days.
There are three major stakeholder groups that need insight on this granularity to make better decisions: special investigations units (SIU), claims adjusters and actuaries. A siloed approach typically characterizes how these groups look at the data, which leads to inevitable overlapping of efforts.
This re-work is a result of users not being able to get the level of data granularity in the way the want it. Instead, each stakeholder group sources data the way they want due to lack of granularity and lack of trust in the data. What carriers fail to recognize is the common goal among these groups - to predict the future in a repeatable, efficient process.
The correct approach is to take the most granular data available – at the medical bill line level with corresponding diagnosis / procedure codes – and meet the shared objective of identifying claims development patterns taking shape in the portfolio. For the SIU, this means detection of anomalous billing patterns previously unnoticed and a greater proportion of referrals resulting in action. For claims adjusters, this means reduced claims adjusting expense, improved medical outcomes for claimants, and improved customer / business partner experience. And for actuarial, this means more homogeneous blocks of business, more precise solvency monitoring and augmented feedback to pricing and underwriting.
Beyond the inherent inefficiency of a siloed approach to complex medical claims, the derived insights often overlook nuanced provider abuse across the fraud spectrum while failing to connect operations-oriented, claims level predictions to actuarial reserving and pricing strategies. There is a better approach that weaves together the objectives of these disparate audiences by building a common foundation of data and insights.
As a harbinger of the future that artificial intelligence will enable,
McKinsey states that “claims processing in 2030 remains a primary function of carriers, but head count associated with claims is reduced by 70 to 90 percent compared with 2018 levels.” To achieve this level of transformative AI, carriers must upgrade the status quo of isolated point solutions to a unified vision of claims excellence that transcends stakeholder groups.
Consider a portfolio of workers compensation claimants, each with a unique clinical pathway defined by an underlying web of complex medical codes defining each encounter. A
single AI-driven anomaly detection algorithm answers the following questions asked by
three different stakeholders:
- SIU/Fraud: Which claims identify as outlier pathways indicative of fraud & abuse?
- Claims Adjusters/Clinical Intervention: Which claims align to common inlier pathways that typically lead to expensive surgeries?
- Actuarial: What are the key profile shifts between inlier and outlier pathways that suggest a fundamental movement of injuries and / or treatment patterns in the portfolio?
The era of the black box point solution is over and the demand for trust in managing medical claims is no exception. Filing any claim is “the moment of truth” from a customer experience standpoint. And given the tremendous human element inherent with medical claims, the stakes could not be higher. In other words, the answer to questions from policyholders, claimants or internal decision-makers cannot be “this is what the system told me.”
Trusted AI for casualty medical claims begins with a clear line of sight into the data. Given the inordinate number of possible diagnosis and procedure codes that define a single encounter, let alone over the life of an entire claim, numerous assumptions must be made to build and implement AI solutions. Visibility into such assumptions and the ability to modify where needed or desired, in conjunction with how structured coding data is blending with unstructured sources like physician notes, is paramount to instilling faith in the promise of AI.
Explainability is a requirement of any business case and is particularly crucial in communicating with external audiences in the context of medical claims. Confronting providers over suspected fraud and abuse, for example, requires tremendous care. Most are honest, well intentioned providers, and they are also crucial business partners and so the case must be clear and justified in raising objections to a bill.
The third pillar to transparency, fairness, is no less important. According to the
National Council on Compensation Insurance, approximately two-thirds of large workers compensation claims (defined as over $ 1 million) start off as “fairly routine,” so there is a compelling argument for utilizing AI to take a more proactive stance in management of projected losses. But as seen with existing efforts in recommending patterns of care (such as
Optum’s tool), algorithms are subject to bias leading to disparate treatment protocols across different slices of protected classes of individuals.
The bottom-line question is can you afford NOT to deploy AI for analyzing medical claims?
