What is driving the increase in Medicaid spending?
A representative output from Unsupervised's current analytics engine, showing the report quality Finder is being packaged to support: quantified drivers, caveats, and open questions for follow-up analysis.
One prompt, a long-running workflow, and a report artifact.
This example came from an agent-led run over the HHS Medicaid data. The important point is not just the final answer; it is the process that kept the work scoped, checked, and source-backed.
One sentence started the run.
The agent was asked to run the full analysis workflow on the HHS data in the project directory: a CMS Medicaid provider-spending CSV with 227M rows.
The agent selected the analysis procedure.
It read the project context, identified the dataset, validated row scale, scoped the work, and launched parallel workstreams with process checks and quality gates.
Browser work added external context.
The run used browser-accessible BI and public federal data to validate findings and bring in context beyond the local CSV.
Finder handled the combinatorial pass.
The pattern engine explored multi-feature interactions directly, including trillions of candidate combinations that would be impractical for an LLM to test one query at a time.
The final artifact was interactive HTML.
The agent generated a shareable report with narrative structure, charts, caveats, open questions, and source-backed findings.
The full HTML report is embedded below.
This is the original report artifact from the run, served as a standalone HTML page and embedded here so the example is inspectable rather than summarized away.
increase in Medicaid FFS spending from 2018 to 2023
total spending growth over the period
growth explained by volume rather than price
growth attributed to HCBS service category expansion
The spend increase was mostly a volume story.
Medicaid fee-for-service spending increased $90.1B between 2018 and 2023. The output identifies volume as the primary driver, HCBS as the most important service category, and provider billing structure as the strongest pattern signal.
Utilization and volume
$55.2BPrice and mix
$35.0BHCBS category contribution
$38.8BNew provider participation
168K providersA report should show what changed and why the agent believes it.
Growth was broad-based, but not evenly explained.
Fee-for-service spending grew from $108.7B to $198.8B. Pattern exploration separates volume, price, service category, provider structure, and open policy questions instead of reducing the result to one trend line.
HCBS was the standout service category.
Home and community-based services accounted for roughly 43% to 44% of total growth, making it the clearest category-level driver for follow-up investigation.
Billing structure was the strongest pattern signal.
The strongest discovered segment involved organizational billing complexity: servicing-to-billing ratios above 1.31 and agency-based care delivery.
The useful artifact is not just a chart.
Finder should help an agent create a report that survives review: concrete metrics, traceable assumptions, limitations, and the next questions a human analyst would ask.
Quantified drivers
Volume, price, service category, and provider effects are separated instead of blended.
Narrative synthesis
The report explains the result in plain language while keeping the numbers visible.
Caveats
Limitations and normalization questions stay attached to the conclusion.
Open questions
The agent leaves a path for the next run instead of pretending the analysis is done.
Good analysis ends with better questions.
Which state waiver or eligibility changes explain the remaining variance?
How much of provider growth reflects access expansion versus billing fragmentation?
Where do policy changes, service mix, and provider structure reinforce each other?
Which patterns remain stable when claims are normalized by beneficiary population?
Put a pattern-finding run inside your agent workflow.
Start with the developer preview, then bring the same pattern-discovery workflow to your own data.