Ingest
Fax, email, upload or a photograph from a phone. Every page is stored exactly as received and never modified.
Clinical document intake
Documents in, orders out,
into any system.
IntakeLinx reads the packet a practice actually sends you, the fax with the handwriting in the margin and the insurance card photographed at an angle, and turns it into a complete, validated order for the system you already run on.
The reason platforms in this category fail
They rebuild the product for every customer, then cannot afford to keep them.
Intake vendors win a deal, discover the customer's forms are different, and write custom code to handle them. Do that fifty times and you are not running a platform, you are running fifty products with one invoice. The margin goes, and then the smaller customers go with it.
IntakeLinx is built on one rule that is not negotiable internally: onboarding a customer, a document type or a destination must be pure configuration. If it needs a code change, the design is wrong. That constraint presents itself repeatedly as a reasonable one time exception. It is not one.
Onboarding a new customerConfiguration
Adding a document typeConfiguration
Adding a destination systemConfiguration
Adding a canonical fieldConfiguration
The pipeline
Fourteen stages, each one a row you can inspect.
Nothing is chained inside a single request. Every stage is a durable, retryable job, so a replay against a corrected crosswalk or a newer model costs nothing and loses nothing.
Normalize
Pages are rendered and a text layer is pulled where one exists. Native text beats re-reading pixels, so it is used alongside the image.
Reassemble
Transmission headers are read and stripped, and a packet that arrived as three separate faxes becomes one packet again.
Decode
Barcodes and machine readable marks are read where present. Where no reader is wired, the stage says so rather than claiming it looked.
Triage
Blank pages, duplicates and cover sheets are separated from the pages that actually carry intake value.
Split
One transmission becomes one packet per patient, using document boundaries rather than a fixed page count.
Classify
Each document is identified by type, form family and revision. An unrecognized revision extracts against its family and flags, rather than failing.
Extract
Fields are read against the canonical schema. The model is never asked for a derived value, so it cannot invent one.
Resolve
Conflicts across documents are settled by a precedence table you can read. Confidence never breaks a tie.
Derive
Codes that follow from other codes are computed by crosswalk. A procedure plus a confirmed laterality selects a kit, and the kit selects the HCPCS.
Validate
Payor rules, documentation requirements and code checks run as configuration. Each finding names the rule that produced it.
Readiness
The order is submittable or it is not, and the reason is a list of named gaps rather than a score.
Review
Anything the system declined to decide reaches a person with the page and region it came from, so the check takes seconds.
Deliver
The finished order goes to NetSuite, Brightree, NikoHealth, DaisyBill, a FHIR endpoint, an HL7 feed or a CSV drop. Built and under test, not yet run against a live account.
What makes it trustworthy
Every value knows where it came from.
An extracted field is not a string. It carries the characters that were on the page, what those normalized to, which document and which region of which page it was read from, how confident the read was, and whether a human has confirmed it. Those are separate facts and the system never collapses them into one.
Confidence is never precedence
A model that is 0.98 sure it read "left" does not outrank a surgical consent that says right. Precedence comes from a table naming which document type wins for which field, and you can read that table.
Absence is a finding
"Not on the document" and "illegible" are different from empty, and both are different from a value nobody has checked. An order is not ready because nothing errored. It is ready because the requirements were met.
Laterality never auto resolves
Left and right are platform locked. If the diagnosis code and the procedure text do not corroborate each other, a person confirms it. This is the field where being wrong costs a returned product and a repayment.
Identifiers keep their leading zeros
An MRN or a member ID is never stored in a numeric type. 000431 and 431 are different patients, and a database that cannot tell them apart is a database that will merge two charts.
Capture
A rep photographs the stack in the parking lot.
Open the app on a phone, shoot page after page, tap New patient between patients, and send once. Five patients and nineteen pages arrive as five orders, not nineteen records somebody has to reassemble by hand.
No patient name is typed on the phone. That is PHI entry on a device in a parking lot, it is slow, and it is a second place for a name to be wrong. The packet is "the third one in this run" until extraction reads who it belongs to.
See the workflows this replacesGet a walkthrough on your own documents.
Bring a packet type that has been hard to automate. Redact it first, or use ours.
Book a walkthrough