Turning public healthcare data into proprietary investment signals
How an approximately $500 million long/short fund transformed reimbursement and physician-payment data into auditable diligence signals in under an hour, within its existing cloud environment.
- Client
- $500M long/short fund
- Industry
- Investment management
- Engagement
- Healthcare diligence
- Published
The research workflow
From fragmented public records to source-linked investment signals in under an hour.
Challenge
The signal was public. Turning it into evidence was not.
BTCP partnered with an approximately $500 million SEC-registered long/short fund running a concentrated portfolio of 20 to 30 U.S. small-cap equities. The engagement focused on two healthcare positions the portfolio manager covered closely, including the fund’s second-largest holding.
The investment theses turned on signals buried in Medicare, reimbursement, and physician-payment data: procedure adoption, equipped-site growth, physician behavior, and the quality of commercial engagement. The portfolio manager wanted to pressure-test management’s reported numbers against underlying activity and identify demand changes before they appeared in results.
The source data was public, but access was not the problem. Years of records were spread across large, inconsistently structured datasets that required extensive cleaning, entity resolution, and interpretation before they could answer an investment question.
Doing that work manually would consume weeks of analyst time and still be difficult to reproduce. The fund needed a governed way to turn raw records into evidence without building a new internal data team or moving its research workflow outside its existing environment.
Is installed-base adoption accelerating before it appears in reported revenue?
Do paid physician relationships translate into actual procedure activity?
Which physicians form the commercial core, and is that core strengthening or thinning?
How much reported engagement growth reflects meaningful investment rather than low-value activity?
Solution
A governed research engine inside the fund’s environment
Public data becomes a proprietary advantage when a firm can reconcile it at operating depth, connect it to an investment thesis, and reproduce every conclusion from source evidence.
BTCP deployed an agentic research layer through the fund’s approved model environment and existing cloud infrastructure. Claude provided the research interface, while data preparation, workflow state, governed source materials, and the resulting research artifacts remained within the firm’s controlled environment.
The pipeline ingests years of public reimbursement and physician-payment records, resolves inconsistent entities to the individual physician, removes duplicates, and reconciles activity across datasets. It then tests each finding against company disclosures and retains the source trail behind every material conclusion.
The analyst can run the workflow without manual data preparation or engineering support. A research effort equivalent to three or four weeks of work now runs end to end in under an hour and can be repeated as new data becomes available.
Validation controls also caught two ingestion errors before the results reached the portfolio manager. Each would have overstated a key growth metric by approximately 5%; the affected records were corrected and revalidated against the source data.
The system does not replace investment judgment. It gives the portfolio manager a faster, more inspectable evidence base for distinguishing reported growth from the operating activity underneath it.
Findings
What the underlying activity revealed
01
An installed-base ramp before the financials showed it
The number of equipped sites roughly doubled year over year, indicating a sharp increase in capital-equipment placement.
The company’s next report showed capital sales up 132.5% and platform revenue up 64.5%. The stock rose approximately 12% following the report.
02
A widening gap between paid and active physicians
The company paid 4,528 physicians while approximately 3,337 were actually operating—a 36% gap, up from 15%. Paid physicians grew about 26%; operating physicians grew about 6%.
The divergence exposed a demand-quality warning that the reported active-user figure obscured and preceded softer operating results.
03
Commercial depth was eroding beneath a stable headline
Average investment behind each meaningful physician relationship fell from approximately $14,500 to $9,700, a 33% year-over-year decline.
The high-volume physician core was thinning even as total contacts appeared stable, weakening the part of the network most closely tied to procedures and revenue.
04
Most headline engagement growth was noise
Sub-$100 interactions accounted for 94% of payment growth, while fixed legacy royalties represented 68% of more substantial spending.
Removing those effects reduced the apparent growth story to an approximately $3 million discretionary commercial core with little evidence of new investment behind it.
Impact
Weeks of diligence compressed into an auditable research run
- <1 hour
- to complete work equivalent to three to four weeks, or more than 120 analyst-hours
- 40,000+
- physician-payment records reconciled across five years and resolved to individual physicians
- 15% → 36%
- widening gap between paid physicians and those actually performing procedures
- 130%+
- capital-sales growth in the next company report after the data signaled an adoption ramp, followed by an approximately 12% one-day stock move
Figures were reconstructed from public physician-payment and reimbursement datasets, deduplicated to the individual physician, and cross-checked against source records and company disclosures. Subsequent company and market results are presented as observations, not claims of causation.
Beyond this workflow
One workflow, built as a reusable capability
This engagement began with healthcare diligence, but the underlying pattern is broader. BTCP implements AI capabilities where a firm’s governed data and workflows already live, using approved models as the interface and preserving the evidence required for review.
Any repeatable, judgment-heavy research process can become a scalable engine: one that absorbs more data than an analyst could process manually, retains the source trail behind its conclusions, and becomes more useful each time the firm runs it.