FAQ
Questions worth a straight answer.
Including the ones where the answer is that we do not know yet.
How accurate is the extraction?
We do not have a number yet, and we are not going to invent one. The pipeline records the token count, latency and stop reason of every model call and every field carries its own confidence, so accuracy on your documents is measurable from the first packet rather than estimated. Any vendor quoting you a single percentage across all document types is quoting a marketing number, because accuracy on a clean typed form and accuracy on a third generation fax with handwriting are not the same problem.
How long does onboarding take?
Onboarding is configuration: your document types, your payor rules, your destination field mappings, your crosswalks. There is no code written for your account, which is the whole design. What takes the time is the decisions, not the build, and the honest answer for a first customer is weeks rather than days because we will be finding gaps in the configuration surface together.
What happens when a practice changes a form?
A revision the system has not seen extracts against the form family it belongs to and is flagged as an unrecognized revision. It does not fail, and it does not silently pretend it recognized the layout. Somebody reviews the first few and, if the new revision needs different handling, that is a configuration change.
What happens when the model gets something wrong?
Every value carries the page and region it was read from, so a reviewer confirms or corrects it in seconds rather than reading the whole fax. Corrections are recorded against the case, and the field then carries the fact that a human confirmed it, which is a different state from a high confidence machine read.
Do you use our documents to train models?
No. Model calls run under a business associate agreement with zero retention enabled, so document content is not retained by the model provider after the response, and it is never used for training by us or by them.
Where does the data live?
In a dedicated database and private object storage, encrypted at rest, with tenant isolation enforced by row level security in the database rather than by a filter in application code. Original documents are stored exactly as received and are never modified.
What if we already have an intake team?
You will still have one. The work changes from typing to confirming. The interesting number is not headcount, it is how many orders one coordinator can put through in a day and how many of them come back.
Which systems can you deliver into?
NetSuite first, then Brightree, NikoHealth, DaisyBill and DrChrono, plus FHIR, HL7 and SFTP CSV for everything else. A destination is a field catalog and a mapping profile, both configuration. Adding one does not change the pipeline.
What is not built yet?
The review queue is minimal today, the mapping studio is being built, barcode decoding is not wired, and SOC 2 and third party penetration testing are not complete. We would rather you read that here than discover it in month two.
What does it cost?
Per packet, with the price depending on document mix and volume. We are not quoting a number in public before we have measured our own cost per packet on real document types, because a price set by guessing gets corrected later, and that conversation is worse than this one.
Our current vendor is exiting. How fast can you move?
That is the situation we built the timeline around. Tell us your cutover date and what your current vendor does for you today, and we will tell you honestly whether we can be ready and what we would need from you to get there.