Medicaid is a lifeline for millions and yet is operationally complex for states to administer. Frequently changing eligibility rules, outreach and communication challenges, aging systems, processing backlogs, and overwhelmed call centers all create significant operational hurdles. And more change is coming: the 2025 reconciliation law (commonly referred to as H.R.1) introduces new Medicaid work requirements and more frequent eligibility redeterminations—changes that will compound existing beneficiary and operational challenges.
Driving good outcomes in this environment depends on having consistent, reliable, and timely data to measure and identify opportunities for improvement. Most agencies find that the data they need isn’t available when they need it, it takes too long to turn into something usable, or it doesn’t actually answer the question leadership was asking.
The result is that agencies end up flying blind at exactly the moments they can least afford to. Without data, or the right data, leaders may make decisions that don’t move the needle, or worse, move it in the wrong direction. The result is administrative cost growth, loss of eligible beneficiaries, or growing backlogs. The fix isn’t a bigger reporting team or a new dashboard tool, but something simpler: bringing the data analyst to the table—into the room where questions are asked and decisions get made—so they can provide needed insight with the full context of the desired program outcome their data is meant to inform.
The challenge of the disconnected data analyst
Most state Medicaid agencies have some form of data or reporting team, but its primary mission is often to track and report on metrics mandated by the Center for Medicare and Medicaid Services (CMS), which oversees Medicaid federally. These functions are at capacity with their existing reporting workloads, meaning data analysis to support answering program leadership’s fast-moving questions take a back seat.
When leadership wants to know what opportunities exist to alleviate operational challenges, such as why backlogs are climbing or why rates of procedural termination are increasing, the request usually gets routed to a systems vendor or IT team, where it lands in a queue behind other backlogged priorities. By the time an answer comes back, the window to act on it has often closed.
The pandemic-era public health emergency unwinding period is another example of this. During the unwinding, states had to renew coverage for the entire beneficiary population in short order. Coverage loss and operational load were significant concerns. As part of the US Digital Service team that helped states increase rates of ex parte, or automatic, renewals during this time, I saw first hand that data that leaders needed to identify bottlenecks and sources of errors were not available and were not straightforward to pull. In the meantime, hundreds of thousands of beneficiaries were being disenrolled each month for procedural reasons that could have been avoided.
There’s a deeper problem underneath delays, though. These requests are typically made without much context about the decision they’re meant to inform. As part of our Medicaid Delivery Innovation Initiative (MDII), we partnered with a state to understand whether beneficiaries are sufficiently aware they can submit renewals online. To start, we asked for data about what percentage of Medicaid renewals were submitted online. Surprisingly, we got back a number that was higher than we expected. After some discussion with the data analyst who provided the data, we found that it included renewal that navigators submitted online on behalf of beneficiaries. The data provided was technically correct—it included all renewals submitted online—but the context of what we were trying to understand was lost: how many beneficiaries submitted their renewals online. Had we not found that misalignment, we could have drawn the incorrect conclusion that beneficiaries were very aware of their ability to renew online. If instead, the data analyst was in the room while we were discussing the outcome we were trying to achieve—improving utilization of online renewal functionality—we would have been provided the right data the first time.
Why embedded analysts are different
Agencies don’t need to invest in more data resources to get timely, actionable analysis; they need to build a different relationship between analysts and the people making decisions.
When a data analyst sits with program leadership and policy staff regularly, something shifts. They stop being a request-fulfillment function and start understanding the actual outcome—and the context of the outcome—that a leader is trying to drive toward. That context changes how they approach the work: instead of answering the literal question asked, they can help figure out what will actually provide meaningful insight to determine leadership action.
This allows for a closed feedback loop: when a program need surfaces, the embedded analyst helps translate it into a data insight that is used to make a meaningful programmatic decision, and that decision generates the next research question. Each cycle refines the next one, which is only possible because the data analyst was in the room for all of it.
In the example of online renewal rates, a data analyst embedded in program operations meetings would have provided accurate data about what portion of renewals are submitted online, but then also been a part of the conversation in which further clarifying questions are asked: do people try to log in but get stuck? How many of our beneficiaries even have accounts that would let them go online? Are our beneficiaries using mobile devices or desktop computers? Using data available to them, the analyst could provide stakeholders with not only data, but actionable insights that can drive the team’s focus.
How to get started
Introducing new ways of working can be challenging, and it may require adaptation from program staff and data analysts to operate in an integrated model. We recommend starting small.
First, choose a specific priority outcome or workstream with which to try this model. Examples include reducing call center wait times, or updating notices and systems so residents can verify they meet work requirements.
Next, assign a data analyst to become part of the ongoing work related to this initiative. Have a frank, clear conversation with the data analyst about any changes to their role expectations and what success looks like. An embedded role like this requires proactive initiative, willingness to ask questions and offer ideas, and the ability to look for creative, efficient solutions. Include them in meetings, email updates, and group chats, including the ones where program leadership actually discusses what’s going wrong and what to prioritize. This alone changes what analysts understand about the work, even before any formal role changes, and it’s the fastest way to prove the value of the model internally. Leverage the data analyst to provide decision critical insight as the work continues. For workstreams that are introducing something new (such as a new policy implementation, or new system features), ask the data analyst to propose how the desired outcome should be measured and ensure that the systems team is tracking that data.
While existing data analyst staff have familiarity with systems and data schemas, moving to an embedded model may require additional—or different—staff. Existing reporting responsibilities will continue to need servicing, and data/reporting team members may find they aren’t as comfortable operating in an embedded model where they need to translate program needs into measurement approaches and collaborate directly with leadership. Piloting the approach allows for learning, adapting, and iterating.
If successful, this approach can be scaled by building outcome-focused teams of cross-functional experts—operations, policy, IT, and a data analyst—to drive improvement of key program priorities.
Now is the right time
The outcomes states will be tracking under H.R.1, from work requirement compliance and beneficiary retention to managing more frequent redeterminations, will demand exactly the kind of data fluency this model builds. Agencies that wait until those requirements take effect to build this capability will be building it under pressure, in the middle of a crisis, instead of being prepared ahead of one. Getting there starts with bringing existing data analysis talent to the table.