Health Plans: Your BI Tells You MLR Moved. Can Your AI Tell You Why?

Health Plans: Your BI Tells You MLR Moved. Can Your AI Tell You Why? | Databricks Blog Skip to main content

Summary

BI shows MLR moved; AI shows why. Conversational AI lets payer finance leaders decompose variances across claims, utilization, cost, and population risk in minutes, without waiting on analysts or another round of reports.

AI needs payer-specific context to earn trust. MLR is a stack of components (IBNR, rebates, risk adjustment, provider settlements) spread across systems with inconsistent definitions. Without that business logic baked in, AI amplifies confusion instead of resolving it.

Databricks + Abacus Insights = one reusable payer intelligence foundation. Databricks provides the governed data and AI platform; Abacus brings normalized health plan data and domain expertise, together powering answers across MLR, payment integrity, total cost of care, and more.

A health plan CFO closes the month after the usual round of extracts, spreadsheets, and manual reconciliation, and sees the financials are behind budget on a higher-than-expected medical loss ratio (MLR). A BI dashboard can surface the variance quickly. But “MLR is up” isn’t really the answer. It’s the beginning of the question. Which line of business or market is driving it?

Are claim costs up and, if so, due to utilization, unit cost, or service mix?

Was a group or product mispriced?

Did population morbidity change?

And what corrective action can be taken?

The real opportunity for AI in payer finance isn't identifying the variance. It's understanding what caused it and what to do about it. AI transformation in finance results in three things changing at once. Answers arrive in minutes instead of a reporting cycle. Leaders double-click on their own, without a canned report or waiting on an analyst. AI can look past claims, revenue, and membership to clinical information, quality measures, and risk scores, reading all of it at the same time to assess what is driving the variance. That work was never practical when the expertise for both the data and the subject matter was siloed. The challenge then becomes whether AI can understand enough about a health plan’s data and business for finance to trust and act on its insights. That’s where Databricks and Abacus come together. Databricks provides the data and AI capabilities that let organizations interact with governed enterprise data in new ways. Abacus brings the payer-specific data foundation, business context, and operational knowledge those capabilities need when the questions involve healthcare finance. From finding the insight to asking the next question For years, business intelligence worked by anticipating the questions someone would want to ask. Teams built reports. Analysts created dashboards. Executives reviewed the metrics someone had decided belonged on the page. When a number moved, figuring out why meant leaving the dashboard, finding the right analyst or team to investigate, pulling another report or reconciling multiple systems of record. And every answer left something behind: another report to run, another dashboard to build and maintain. Each one was already stale by the time the next question arrived. More work for the team and diminishing returns. Conversational AI changes that interaction model. Instead of stopping at what a dashboard was designed to show, a finance leader can ask a question in plain language, get an answer, and follow it with another question based on what they just learned. The experience starts to look less like navigating reports and more like investigating the business. What really breaks down is the barrier between two kinds of expertise. The executive knows the business and owns the decision-making but cannot query the data; the analyst knows the data, where it lives, how it is structured, how to manipulate it, but not always which question matters. AI collapses that handoff: the person who needs the answer can now ask for it directly. But the quality of that conversation depends on what the AI understands underneath it. AI needs more than access to the data For years, the data modernization question was largely about access: can we bring claims, membership, provider, contract, clinical, and financial data together so people can use it? That remains essential,  but AI adds another requirement . It needs to understand what that data means. Consider the metric of MLR, the share of premium revenue spent on medical care for a plan’s members. Easy to define, hard to calculate. Claims and premiums come on different structures and different timelines, and “medical” expense is really a stack of components, including medical claims, pharmacy claims and rebates, incurred but not reported (IBNR), payment integrity recoveries, risk adjustment transfers, provider settlements, and more. Then it has to be cut by line of business, market, or product, and defining those cohorts takes business logic that never comes out of the box. Without that context, an AI system can have access to enormous amounts of data and still not produce adequate insights. Much of this business context has traditionally lived in different places: data models, reporting logic, documentation, spreadsheets, institutional knowledge, and the heads of the people who built the reports.

For AI to answer finance questions reliably, more of that context has to become explicit, consistent, and reusable. That is why a payer-specific data foundation matters. Health plan data does not arrive neatly organized around the questions executives want to ask. Claims, eligibility, provider, contract, clinical, and financial information come from different systems, in different formats, with different relationships and timing. Before AI can reason across it effectively, the data has to be connected and organized in a way that reflects how a health plan operates. Abacus provides that payer-specific foundation: bringing healthcare data together in a consistent structure and adding the business context needed to understand the relationships among members, claims, providers, contracts, clinical information, and financial measures. Even on the same underlying data, the “same” metric is often defined differently depending on who is calculating it. One version lands in the CFO’s monthly reporting package, while another shows up in an ad hoc analytics request, and eventually someone has to reconcile them. AI does not solve that problem on its own. Without...