What if the answer was already in your data?

What if the answer was already in your data? | Databricks Blog Skip to main content

Summary

Health system leaders make high-stakes strategic decisions — payer mix, M&A, market expansion, referral leakage — on incomplete claims data, and the expertise to interpret it has historically been gated by budget.

Kythera Labs packages that expertise into AI agents built on Databricks (Agent Bricks, Genie, Unity Catalog, and Lakebase over 339 billion claims), so any leader can ask strategic questions in plain language and get governed, trustworthy answers in minutes.

In production, a Louisiana health system went live in 10 days and saw 150% more visibility into patient encounters, 22% less leakage, and $3.8M in estimated annualized value.

What If the Answer Was Already in Your Data? Kythera Labs is building an AI-native healthcare strategy platform on Databricks that gives any health system access to the expert intelligence they need and can trust. The meeting ends the way these meetings always end: with a question no one can answer quickly enough. A CEO, CIO, and CFO walk out of a planning session with a mandate: identify how much oncology revenue is leaving the system and where it's going. In a well-resourced health system, that question goes to a consulting firm and comes back six weeks and several hundred thousand dollars later. In most health systems, it goes to an analyst with a BI tool and comes back whenever it does, with whatever confidence the data allows. The gap between those two experiences is the problem Kythera Labs was founded to close. The Intelligence Gap Nobody Talks About Health system executives face a complex set of strategic decisions simultaneously: growing patient volume, optimizing payer contracts, evaluating M&A targets, identifying underserved markets for expansion, and reducing administrative overhead, all with incomplete data. These decisions align with value creation levers, which historically have required expert analytical capacity that correlates almost entirely with institutional budget. "The market has been served by BI tools sitting on top of claims data," says Jeff McDonald, CEO of Kythera Labs. "BI tools can do a good job of representing what's in the data. They don't do a good job of telling you what's not in the data. That's the antithesis of what the tool is designed to do." The analysts who can bridge that gap (who understand the missingness and bias in claims data, who can reconstruct a patient journey from fragmented billing records, who know the difference between what a claim says and what actually happened clinically) take years to develop. Large health systems hire them. Smaller organizations do without or spend millions on consulting firms to rent that expertise by the engagement. The strategic intelligence gap in American healthcare is not primarily a data problem. It's an expertise distribution problem. Kythera Labs is solving it with AI. The Data Has to Come First Before any agent can answer a strategic question reliably, the data it reasons over has to be reliable. That's a harder problem than it sounds. Claims data is billing exhaust, generated so providers can be reimbursed, not so executives can make market strategy decisions. Repurposing it requires resolving provider identities across dozens of competing sources, harmonizing procedure codes across 130 standardized medical vocabularies, correcting for systematic missingness, and reconstructing patient journeys as temporal sequences rather than collections of disconnected billing events. The answer is Kythera’s healthcare data tech, which takes 339 billion medical and prescription drug claims representing over 300 million patients, eight years of history, more than three petabytes of storage and creates something an agent can actually reason over — an event-based structure where a knee replacement isn't a billing code but a surgical event with a pre-operative history, a discharge, and a post-operative care trajectory. All built on Databricks. That translation is the work. It is also what makes the agent's answers trustworthy. Kythera's operational layer runs on Lakebase, Delta Lake, Delta Sharing, Unity Catalog, and serverless infrastructure — so the transactional data powering real-time workflows shares a single governed foundation with the analytical data the agents reason over. No ETL. No data movement. No seams between the question and the answer. The proof is in production. A health system in Louisiana signed a contract with Kythera in December 2024 and went live before Christmas. Ten days from contract to first insight with visibility into their patient population that they had never had before: 150% increased visibility into patient encounters,

12% more keepage,

22% less leakage, and

$3.8M in estimated annualized value from retained encounters.

That kind of time to value is only possible because the data foundation was already built. The Guided Analytics Experience With Kythera’s  Healthcare Strategy Agent  built on the Agent Bricks framework, deployed into the health system’s Databricks workspace, the Chief Strategy Officer opens a conversation and asks: “How many cancer patients are being referred to non-affiliated providers for services we offer?”

What follows is not a dashboard refresh. The agent works through the question the way a seasoned analyst would, surfacing 6,800 referred oncology patients, naming the competing providers capturing their volume, identifying the highest-leakage referring physicians by name, and putting $23.1 million in reimbursement opportunity on the table. Specific retention strategies follow. The whole session takes minutes. Every result in that response represents a query that a human analyst would have had to write, test, validate, and consolidate. The agent runs them in minutes. "A lot of what we're doing is packaging the expertise around how to work with these datasets into intelligent agents, so that capability isn't limited to a small group of specialists," says Ryan Leurck, Co-founder and Chief Analytics Officer. "Even when you have the right people, answering complex questions can take days or weeks. The idea is to make that expertise more accessible and help people get to answers much faster." The oncology scenario is one demonstration of a platform built to address more than a dozen strategic question types, from payer mix optimization and demand forecasting to...