You already have
Years of operational signal
Experts, workflows, case histories, edge conditions, decision rules, and outcomes generated through ordinary work.
Healthcare operations are an AI asset
Your RCM expertise can train the next generation of AI.
Pulaski Data partners with healthcare revenue cycle companies—not to sell them another dashboard, but to turn hard-won operational knowledge into protected AI training infrastructure. We capture how expert work actually gets done, recreate it in controlled environments, and structure partnerships so the operator can potentially share in the value while benefiting from the automation we uncover.
- Potential new revenuepaid or shared-value structures
- Protected by designapproved, transformed workflow packages
- Automation comes backleverage for your own operation
The business model
A partnership, not another software subscription
Your operation may be holding an AI asset.
AI systems already know how to write an email or summarize a policy. What they lack are examples of how experienced people resolve difficult cases across payer rules, portals, documents, deadlines, and exceptions.
That knowledge lives inside revenue cycle companies.
We build
The technical bridge
Instrumentation, transformation, de-identification, simulated tools, realistic tasks, evaluations, and automated checks.
Together we create
Training infrastructure + internal leverage
A reusable environment for AI training and evaluation—and a clearer path to improve or automate the work inside your company.
You do not need an AI strategy. Show us one difficult workflow your team performs repeatedly.
Show us a workflowWhy healthcare revenue cycle
Complex · multi-system · economically measurable
RCM contains exactly the work AI cannot learn from a textbook.
Revenue cycle work is rules-heavy, time-sensitive, and full of incomplete or conflicting information. Success depends on specialized judgment—and the result often has a clear economic outcome.
01Claim
02Denial
03Payer policy
04Analyst investigation
05Portal + record
06Decision
07Appeal
08Payment
Candidate workflows
How it works
From ordinary work to a repeatable learning system
We capture what expertise looks like—not just what the database contains.
The work begins beside your experts. The end product is a controlled representation that preserves what makes the job difficult without exposing the live operation.
- 01Map
Learn the workflow
Work alongside experienced team members on a narrow, difficult, high-value process.
- 02Signal
Capture the expertise
Identify what the expert sees, which tools they use, what they decide, what they do, and what success means.
- 03Control
Make it safe and reproducible
Remove or transform sensitive information and rebuild only the systems, records, and state the task requires.
- 04Environment
Turn it into training problems
Create realistic starting states, tasks, simulated tools, outcome checks, and expert rubrics that can be run repeatedly.
- 05Revenue + leverage
Share the value
Explore permitted training opportunities and use what we learn to improve automation inside the operating company.
- 01Real work
- 02Expert workflow
- 03Protected environment
- 04Better AI
- 05Revenue + automation
- 06More feedback
What is a training environment?
Think of it as a flight simulator for AI workers
An AI agent should not learn claims work on a live claim.
A pilot does not learn by immediately flying a passenger jet.
We recreate the important pieces of a real workflow in a controlled environment. An AI system can investigate a denied claim, inspect documents, use simulated tools, make decisions, draft an appeal, and attempt to resolve the case. The environment can then determine whether it succeeded.
Once the idea is clear: AI researchers often call these reinforcement-learning environments, agent environments, or evaluation environments.
Known starting stateDenied claim + objective
A specific case, constraint set, and target result.
Documents + data835 · policy · record
De-identified or transformed artifacts that preserve the task.
Simulated toolsPortal · PM · EHR · APIs
Safe replicas of the interfaces needed to do the work.
Agent actionsInvestigate · decide · appeal
Every step and state change can be captured and replayed.
Verified final stateOutcome + expert rubric
System checks measure the result; experts judge what cannot be reduced to a rule.
Security and governance
Patient information is not the product
Your production data does not become somebody else’s dataset.
The goal is to preserve the logic of the work while removing what an outside training system does not need. Raw source access and production credentials are not the model.
Minimize access
Begin with the smallest useful workflow, expert group, and data surface.
De-identify before approved external use
Remove or transform patient, customer, and employee identity; use synthetic or transformed artifacts where appropriate.
Isolate production systems
Create controlled representations instead of exposing source systems or credentials to a model builder.
Approve what may leave
Define packages, recipients, rights, retention, and permitted uses in writing, with an auditable release path.
Handle PHI deliberately
When PHI is involved, establish appropriate authorization, agreements, infrastructure, and secure exchange before access begins.
Keep the operator in control
Partners review the workflow representation and approve any permitted data package before release.
Current-stage note: These are partnership design requirements, not claims that Pulaski Data currently holds SOC 2 certification, “HIPAA certification,” or a blanket compliance status. Exact controls and agreements must be established for each engagement before protected information moves. Ordinary email and this website are not approved channels for patient or claim-level data.
Why now
Public market evidence · checked September 2026
The market for real workflows and expert data already exists.
These examples validate the thesis. They do not imply that Pulaski Data works with any company named below, and company revenue figures are not prices for an individual healthcare environment.
Scale AI · Operational data partnerships
Ordinary business workflows can become a paid AI asset.
Scale says it turns de-identified operational workflows into frontier-grade training data, lists healthcare and life sciences as an industry of interest, and seeks established businesses with real workflows.
$10K–$1M+
Scale’s published illustrative value per data partnership, scaling with cadence. This is Scale’s figure—not a Pulaski Data estimate or guarantee.
Read Scale’s data partnership page ↗Scale AI · Training environments
Frontier training is moving toward realistic, performed work.
Scale describes realistic applications, APIs and MCP tools, files, known state, expert tasks, and automated verifiers. In February 2026, it said nearly half of its new data-training projects involved RL environments.
Scale is also hiring a Data Acquisition Lead to find companies whose operating data can support sandboxed replicas of real knowledge work.
AfterQuery + Micro1 · Expert data
Specialized human expertise is supporting large AI-data businesses.
Forbes reported in April 2026 that AfterQuery had surpassed a $100M revenue run rate. TechCrunch reported in August 2026 that Micro1 had reached a $500M gross annual run rate and described contract experts including doctors, lawyers, and scientists.
These are company-wide run rates—not transaction prices. We have not found a reliable public disclosure of the price paid for an individual healthcare environment.
Design partners
Start with one workflow, not a data dump
We are looking for organizations that already do difficult RCM work at meaningful volume.
A strong first partner has experienced operators, repeatable work, useful historical signal, and a leader who can approve a narrow experiment. The first conversation requires no PHI and no commitment.
A partner may contribute
- Subject-matter expert time
- Representative workflows and edge cases
- Historical records where appropriate
- Definitions of correct process and successful outcomes
- Review and approval of tasks and packages
Pulaski Data may contribute
- Workflow instrumentation
- Data engineering and de-identification
- Simulated applications and agent tools
- Evaluation tasks and automated verifiers
- Model experiments and internal automation
- Development of downstream training opportunities
Are you asking us to sell patient claims to an AI lab?
No. The goal is to capture the logic of expert work and create a specifically permitted, de-identified or transformed representation. Raw production access, credentials, and unrestricted reuse are not the proposition.
Do you need PHI to begin?
No. A first conversation and workflow map should remain at the company and process level. If later work genuinely requires PHI, it begins only after scope, authorization, agreements, infrastructure, and a secure exchange are in place.
Is payment guaranteed?
No. We are developing the supply and downstream sides of this market. Economics depend on the workflow’s scarcity, quality, reproducibility, rights, cadence, and actual buyer demand.
How does this help our own operation?
Mapping the workflow, defining success, structuring the data, and building repeatable tools exposes where agents can safely assist your team. Internal automation may be part of the partnership or a separately scoped project.
Explore an RCM partnership
Show us one difficult workflow your team performs repeatedly.
You do not need an AI strategy. Tell us what the workflow is, who performs it, why it is difficult, and roughly how often it happens.
hello@pulaskidata.comOrdinary email is only for company- and process-level details. Do not include patient names, claim IDs, dates of service, documents, credentials, or other protected information.