One patient, four records
The CRM, phone system, scheduling tool and billing platform each hold a slightly different version of the same person. Duplicates and mismatched statements follow, and the front desk lead merges records by hand.
Data & Integrations
Stop re-keying between systems, and give every tool only the access its job needs
We connect practice management, EHR, CRM, billing, phone and web forms, and turn faxed referral packets into reviewable work. Every connection has scoped permissions, a named owner and an audit log.
Short answer
What are data and integration services for healthcare businesses?
They connect the systems a healthcare business already runs, such as the EHR or practice management system, CRM, billing, phone and web forms, so information moves without re-keying. The work covers APIs and FHIR where available, document extraction for faxes, patient matching, scoped permissions and audit logs. Integration was the most-cited barrier to scaling AI in one 2026 McKinsey survey.
The problem
Problems this service solves
Referral packets that live as images
Faxed referral letters, scanned insurance cards and PDF results need a person to read and retype them. Urgency written on page three is easy to miss.
AI tools with the keys to everything
A vendor asks for broad chart access to answer a scheduling question, and an administrator accepts. Nobody can later say what the tool read.
Integrations nobody owns
A connection built once stops working after a vendor update, and the failure surfaces as a missing report or a patient complaint. Without an owner and an alert, repair waits.
Context
Why this is harder in healthcare, and where it opens up
What makes it hard
- In a McKinsey survey of 150 US healthcare leaders reported in April 2026, integration was the most-cited barrier to scaling AI. A pilot that works in one tool can stall at the boundary between systems.
- FHIR is the foundation for API-based exchange, but it does not promise complete or correctly matched records. Support differs by vendor and profile, so each connection is tested, not assumed.
- Identity matching is where errors hurt. A supplier reported that a fax tool used by 427 practices matched patients for 75% to 85% of incoming faxes, which is why every pathway needs a review queue for the rest.
- Payer data promised by regulation arrives unevenly. CMS prior authorization API requirements generally begin 1 January 2027, with details varying by payer.
Where the opportunity is
- Fax and referral intake is one of the strongest places to start: it is bounded, measurable and needs no autonomous decisions.
- Where FHIR or other APIs exist, narrow read-only connections replace screen copying and keep access visible.
- One agreed definition of lead, booked, attended and collected lets marketing, the front desk and billing see the same funnel.
- Clean, matched data is the precondition for later work such as AI agents and automated reporting.
How it works
The mechanics, in order
- 01
Inventory systems, flows and owners
We list every system holding patient or customer data, the fields it holds, how it exposes them (API, FHIR, export or nothing) and who owns it. We note where PHI sits, whether vendor agreements exist, and the minimum fields each job needs.
- 02
Connect and match
Connections use APIs, FHIR or a scheduled interface and start read-only. Matching rules carry confidence thresholds, uncertain matches go to a review queue, and the source document stays linked to every record it creates.
- 03
Extract documents and check them
AI classifies scanned pages and reads fields. Identity, urgency and key values are checked before any write-back, and the original image is preserved.
- 04
Log, alert and review access
Every read and write is logged, failed syncs raise alerts, and permissions are reviewed on a schedule so access does not drift.
Human + AI
What AI does, and what people own
AI reads and sorts documents and helps map messy fields between systems. People decide what data may move, who may see it and when a match is certain enough to write back.
AI handles
- Classifies scanned pages as referral letters, insurance cards, lab reports or consent forms.
- Extracts patient name, date of birth, referring provider and reason for referral, each linked to its source page.
- Proposes patient matches and flags probable duplicates for a person to confirm.
- Checks data quality, such as missing phone numbers, impossible dates and inconsistent payer names.
People decide and review
- The intake coordinator confirms identity and urgency on unmatched or low-confidence items.
- The privacy or compliance officer approves data flows, vendor terms and anything that leaves the organization.
- The IT or security lead grants, reviews and revokes scoped permissions and keys.
- The practice manager defines what lead, booked and attended mean for reporting.
What we build
What is included
01
System connections
EHR or practice management, CRM, phone, forms, billing and reporting tools, connected by the method each vendor supports.
02
Referral and fax document intelligence
Classification, field extraction, patient matching and routing for faxes, scanned referrals and PDFs, with the source image kept and a review queue for uncertain items.
03
Identity matching and data quality
Matching rules, duplicate detection and routine quality checks so records stay trustworthy as volume grows.
04
Scoped connectors for AI tools
Authenticated, permission-limited connections, using the Model Context Protocol where it fits, with read-only defaults, per-tool allow lists and logging.
05
Audit logs and access reviews
A record of what each tool, agent and user read or wrote, plus scheduled permission reviews and a fast way to revoke access.
06
Shared reporting layer
Common definitions and one source for lead response time, no-show rate, referral conversion and days in accounts receivable.
Use cases
Examples from healthcare operations
Illustrations of how this work looks in practice. They describe workflows, not client results.
Fax referrals for a service line
Referral packets from many practices are classified, matched and queued for the intake team. One large US health system announced similar work in September 2026, and smaller groups can start narrower.
Hospitals & SystemsLead-to-visit reporting for a multi-site group
Web forms, the call platform and the scheduling system share one patient and lead identity, so leadership sees which referral sources produce attended visits at each site.
MSOs & GroupsClient onboarding for a billing company
Each new practice client runs a different system. A repeatable connection pattern and data dictionary make onboarding and reporting consistent.
Medical Billing & RCMCRM data for a health-tech vendor
Product activity and customer success notes join the CRM without patient data, unless a business associate agreement covers it.
Digital HealthRead-only assistant for a clinic
An internal assistant can look up appointment status and referral progress but cannot open clinical notes, and each lookup is logged.
Medical ClinicsGrowth link
One accurate funnel from first contact to collected revenue
Growth decisions are only as good as the numbers behind them. When marketing, the front desk and billing share patient identity and definitions, leaders can see which sources become attended visits and paid claims, and where leads leak.
- Referral source can be tied to attended visits, so investment follows the referrers whose patients actually arrive.
- Lead response time becomes measurable when form, phone and CRM timestamps line up, which feeds revenue operations reporting.
- Intake data that reaches billing intact should mean fewer front-end corrections, which you can track as first-pass claim acceptance.
- B2B sellers can answer security reviews with a clear account of what each connection reads, who approved it and where the log is.
Our approach
How an engagement runs
Phase 1
Map data and access
Discovery with operations, IT and compliance covers systems, flows, owners, agreements and where PHI sits.
Phase 2
Choose one pathway
We pick a workflow such as fax referral intake, set success measures and decide the fields it needs, building on AI automation where work must move.
Phase 3
Build read-only and test on real documents
Connections start read-only in a test environment using historical documents, with match thresholds set and reviewed before anything writes back.
Phase 4
Enable write-back with approvals
Writes begin with human approval and widen only when sampled results support it. You receive a data dictionary, an owner per connection and a runbook.
Outcomes
What usually improves
Less re-keying
Fewer fields are typed twice between systems.
Referrals become reviewable work
A faxed packet becomes a queued item with its source page attached.
One set of definitions
Marketing, the front desk and billing report from the same meaning of lead, booked and attended.
Narrower access, visible audit trail
Each tool sees only what its job needs, and you can show who read or wrote what.
What this depends on
Results depend on what your vendors expose, how clean your records are today and how consistently staff use the new queue. Matching rates reported by suppliers describe their customers, not your fax mix. Payer APIs will arrive unevenly, so we design for the data available now.
Industries
Where it fits best
01Hospitals & SystemsGaps between systems are daily friction for hospitals and health systems, and we complement EHR-embedded tools by connecting what they leave out. English trusts also need the Data Security and Protection Toolkit.02Medical ClinicsReferral letters, portal messages and phone notes can reach one queue, with a scoped assistant for lookups.03MSOs & GroupsAcquired sites arrive with different systems, so a common identity and reporting layer is the first standardization step.04Medical Billing & RCMPractice clients use many EHRs. Repeatable connection patterns and claim-data checks reduce rework.05Health InsuranceBroker leads, member contacts and claims intake data must move between systems while CMS-0057-F dates apply from 2026 and 2027.06Digital HealthSecurity questionnaires and partner marketplaces depend on describing exactly what each integration reads and who approved it.
Limits
What to watch out for
- HIPAA: the existing Security Rule is in force. The proposal to overhaul it (published 6 January 2025) is not final, and the July 2027 target in the 2026 regulatory agenda is not a legal deadline. Each AI use of PHI still needs a permitted basis, a risk analysis and a business associate agreement where a vendor processes PHI.
- CMS-0057-F: payer operational duties generally began 1 January 2026 and API requirements generally begin 1 January 2027, varying by payer. The claims attachment final rule (20 March 2026, effective 26 May 2026) has a compliance deadline of 26 May 2028, so faxes have not disappeared.
- TEFCA is operational, and from 1 January 2026 newly created or captured data sent under its requirements must conform to USCDI v3, but FHIR operations are staged. In England, NHS guidance points new APIs to FHIR R4 and UK Core, and shared care records are not one national record.
- Texas SB 1188 requires covered EHRs, including cloud-held records, to be kept physically in the US or a US territory, with storage provisions applying from 1 January 2026. Check with qualified counsel where integration platforms store copies.
- The Model Context Protocol (introduced 25 November 2024) is established, but healthcare adoption is emerging, and a pilot or available server does not show production use. A connector is not permission to expose an entire chart.
Questions
Frequently asked questions
Do we need FHIR to integrate our systems?
No. FHIR helps where a vendor supports it, but many systems still rely on proprietary APIs, scheduled exports or interface engines. We choose the method each system supports and test it with real records.
What is a scoped connector, and why does it matter for AI tools?
It is an authenticated connection limited to the data and actions a job requires, such as reading appointment status. A tool should not see the whole chart to answer a scheduling question, and you need a log of what it did.
Can AI read faxed referrals reliably?
For defined document types it is established, but not perfect. Supplier-reported patient matching for incoming faxes ran from 75% to 85%, so we use confidence thresholds, keep the source image and send uncertain items to a person.
Will payer APIs under CMS-0057-F give us data in 2027?
Some, not all. The API requirements generally begin 1 January 2027 and details vary by payer, so plan around what is live and verified rather than what is promised.
Can you work with our EHR or practice management system?
Usually, depending on the access your vendor allows and any terms attached. We check early, and where there is no API we say so and propose the safest alternative.
How do you keep PHI out of tools that do not need it?
We define the minimum fields per job, withhold the rest, keep PHI out of ad platforms and any tool without an agreement to hold it, and log every access. Our AI governance work covers vendor terms and risk assessments.
Find the connection that is costing you the most
Book a discovery call and we will map one pathway from first contact to record and name the access each tool should have.