Four week pilot
We run 2,000 of your real files, publish a measured accuracy report and you keep the results either way.
Fernpost pulls dates, codes, providers and amounts out of scanned faxes, handwritten forms and PDF soup, scores its own confidence and routes anything uncertain to a human.
Every engagement starts with a paid pilot on your real documents, not a canned demo set.
We run 2,000 of your real files, publish a measured accuracy report and you keep the results either way.
Committed annual volume, hosted by us, with the review queue and audit log included.
Runs inside your AWS or Azure account so documents never leave your boundary.
Fields specific to your forms, trained on your labeled examples with a measured before and after.
Low confidence extractions route to your reviewers with the source page highlighted next to the value.
REST and async batch endpoints, documented publicly, with sandbox keys available before you sign anything.
Fernpost was started in 2022 by two engineers from a claims platform and a former utilization review nurse who spent eleven years reading these documents by hand. That last part matters, because knowing which fields are worth being uncertain about is most of the problem. Every extraction carries a confidence score and a link to the exact page region it came from.
“Our reviewers went from reading everything to checking the eight percent the system flagged. Throughput roughly tripled and nobody was laid off, they just stopped drowning.”
Lauren M., VP Claims Operations
“The pilot report was refreshingly blunt about which field types it struggled with. That honesty is why we signed.”
Raj S., Director of Analytics, TPA
“Running it inside our own VPC was non negotiable for us. Deployment took eight days including our security review.”
Kevin D., CISO, regional health plan
It is field level exact match across our production customer base, measured against human reviewed ground truth, not a document level or character level figure which would look higher and mean less. Accuracy varies by field type and document quality, so the pilot measures it on your specific documents and we publish the breakdown by field.
Every extracted value gets a confidence score, and you set the threshold. Anything below it routes to the human review queue where a reviewer sees the proposed value next to the highlighted source region and confirms or corrects in a keystroke. Corrections feed back into your custom field training.
That is your choice. The hosted plan processes documents in our SOC 2 audited environment with encryption in transit and at rest and configurable retention down to zero. The in your cloud deployment runs entirely inside your own AWS or Azure account, and we never see the data. We sign BAAs for covered entities on both.
Because a real pilot means ingesting your document formats, tuning for your fields and producing an audited accuracy report, which is several weeks of engineering work. Charging for it means we do it properly, and you keep the report and the extracted data whether or not you go forward.
Send us your ugliest sample set, the faxes and the handwriting. That is the honest test and it is the one we would rather be judged on.