10decodersSix sentences we have now heard from healthcare operators more times than we can count.
If three of these sound like your last operations meeting, the rest of this page is written for you.
“Intake spends the first three hours of every day clearing the fax queue.”
Provider group · operations lead“We can't tell a complete referral from an incomplete one until a human reads it.”
DME supplier · intake manager“Denials come back for things we could have caught before submission.”
Billing organisation · RCM director“We bought an OCR tool. It reads the easy documents.”
Healthcare platform · VP product“Every payer wants the same information in a different shape.”
Provider group · revenue cycle“We're recruiting for a job nobody wants to keep doing.”
DME supplier · COOMost healthcare organisations we meet have already tried to automate this once.
These are the six reasons it stalled, in the order we encounter them.
01. OCR returned text, not decisions
Character recognition tells you what the page says. It does not tell you whether the order is signed, whether the diagnosis supports the code, or whether this packet is ready to submit. The reading was never the hard part.
02. The bot broke when the form changed
Screen-scraping automations are bound to pixel positions and field order. One payer portal redesign or one new referral template, and the queue silently backs up until somebody notices.
03. Nobody owned it after go-live
The pilot worked. Then the vendor left, the internal champion moved teams, and the exception queue grew until staff went back to doing it by hand. Automation without an operator becomes shelfware within two quarters.
04. The platform assumed you had a build team
The licence was signed on the promise of a drag-and-drop canvas. Building anything real on it needed engineers who understood both the tool and your payer rules. Mid-market operations teams rarely have either to spare.
05. Accuracy was reported as an average
A single accuracy percentage hides the documents that matter. What decides whether staff trust the system is how it behaves on the messy 8%: the third-generation fax, the handwritten amendment, the packet with two patients in it.
06. The workflow was never actually written down
Much of the rule set lives in the heads of two or three long-tenured staff. Any automation attempt that skips the step of capturing that knowledge is automating a guess.
Five digital workers, running the path end to end.
Each one has a defined job, a defined input, and a measurable output. You can start with one and add the rest, or run the full chain. We build them, deploy them into your environment, and keep them running.
Intake
Watches the channel
Monitors fax, shared inbox, SFTP, portal downloads and API feeds. Classifies each document by type, splits multi-document packets into their parts, de-duplicates re-sends, and routes by service line.
- Handles the packet, not the page
- Recognises re-faxes of a document already in process
- Applies your naming and folder conventions on the way in
Extraction
Pulls the fields that matter
Reads each document type against its own field map rather than a single generic schema. Returns structured output with a confidence value and a page-and-coordinate reference for every field, so a reviewer can see where the value came from.
- Patient demographics, payer and member ID, group number
- Ordering provider, NPI, signature presence and signature date
- ICD, CPT/HCPCS, quantities, dates of service, length of need
Validation
Checks before you submit
Runs the extracted packet against your payer and product rules while the referral is still in your hands. Returns a pass, or a specific list of what is missing and which rule requires it — in language a coordinator can act on without asking anyone.
- Completeness against payer-specific documentation requirements
- Internal consistency across documents in the same packet
- Eligibility and authorisation status lookups where an interface exists
Downstream
Writes into your systems
Creates and updates records in the EHR, billing system, CRM or order management platform. Integrates through whatever the system actually supports — REST, HL7 v2, FHIR, X12, database, or supervised UI automation where no interface exists.
- Idempotent writes, so a retry never creates a duplicate patient
- Full audit trail linking every written value back to its source page
- Status notifications to the queue, the referring office, or both
Exception
Hands off what it shouldn't decide
Anything below the confidence threshold, or anything a rule flags, goes to a human review queue with the document, the extracted values and the reason attached. Corrections are captured and fed back, so the same exception type shrinks over time.
- Reviewer sees the page and the field side by side, not a raw JSON blob
- Exception reasons are categorised, so you can see what is actually failing
- We staff and monitor this queue during the operate phase
Every platform in this category sells you a canvas. We deliver the worker.
This is the honest comparison. If you have a strong internal automation team and want to own the build, a platform is a reasonable purchase. Most mid-market healthcare organisations we meet do not, and that is where the licence quietly goes unused.
| Dimension | Buying a workflow platform | Working with 10decoders |
|---|---|---|
| What you receive | A licence, a canvas and a set of connectors | A working process, running on your data |
| Who builds it | Your team, or a partner you source separately | Our engineers, with your operations staff in the room |
| Who runs it at 2am | Your team | Ours, under an agreed support window |
| When a payer changes a form | A ticket in your backlog | Part of what we already operate |
| First production workflow | Dependent on internal capacity | Targeted inside the first engagement quarter |
| What you are billed for | Seats and platform tiers | Delivery, then an operating engagement |
| If it doesn't work | You still hold the licence | We stop at the pilot gate and tell you so |
The extraction and validation layer, already built for healthcare documents.
DocuFindr is our document intelligence product. It is where the reading, structuring and rule-checking happens, so an engagement does not start from an empty repository. It is HIPAA-compliant and deploys inside your environment.
- Document-type-specific field maps rather than one generic model
- Confidence scoring and source-page evidence on every extracted value
- Payer rule sets expressed as reviewable configuration, not buried code
- Pre-denial validation: the packet is checked before it leaves your building
- Deployed in your cloud tenancy or ours, with full audit logging
The same five workers, pointed at a different bottleneck.
Provider groups
Referral intake, prior authorisation packets, records requests and physician order tracking across multiple locations and one shared fax line.
DME / HME suppliers
Documentation completeness is the whole business. Orders, CMNs, face-to-face notes and proof of delivery, each with a payer-specific definition of complete.
Billing & RCM organisations
Charge entry from client documents, EOB and remittance processing, denial letter classification, and appeal packet assembly across many client formats at once.
Healthcare platforms & SaaS
You need document intelligence inside your own product, under your own brand, without building an ML team to maintain it.
Payers & TPAs
Inbound authorisation requests, claim attachments and appeal correspondence arriving in every format a provider office can produce.
Health systems, single department
One department with a defined document flow is a better starting point than an enterprise programme. We are comfortable starting small and proving it there.
Narrow first, in production, then widen.
We do not open with a twelve-month transformation programme. We pick one high-volume path, put a worker on it, and let the results decide what happens next.
Weeks 1–2
Document census
We take a real sample of your recent documents under an executed BAA, sit with your intake staff, and write down the rules that currently live in their heads.
Weeks 3–6
One worker live
A single path, in production, on real volume, with the exception queue visible from day one. There is a gate at the end: measured against the baseline, or we stop.
Ongoing
We operate it
Monitoring, exception handling, rule updates when a payer changes requirements, and an accuracy report you can take to your own leadership.
Quarter 2+
Widen the path
Additional document types, additional service lines, additional workers in the chain. Each one earns its place against the same gate as the first.
Five situations where we would tell you not to do this yet.
We would rather lose the meeting than start an engagement that cannot succeed.
Your volume is genuinely low
Below a few hundred documents a month on a given path, the engineering and operating cost is hard to justify against the labour it replaces. We will say so on the first call.
The documents are mostly handwritten
Handwriting-dominant sources — a wet-signature order block is fine, a fully handwritten intake form is not — need a different approach and a different conversation about expected review rates.
No two people describe the rule the same way
If three staff give three different answers about when a packet is complete, that disagreement has to be resolved before automation. We can help facilitate it, but it is not an engineering problem.
We cannot get a document sample
We do not quote extraction work on a description of your documents. If legal or IT cannot release a representative sample under a BAA, there is no honest way to scope it.
The bottleneck is the payer, not you
If your turnaround time is dominated by waiting for a payer response, faster intake will not move the number you care about. Worth checking before anyone signs anything.
You want the platform, not the outcome
If your intent is to own the build and staff it internally, buy a platform. We would be the wrong shape of partner, and you would feel it by month three.
A product engineering firm, not a reseller.
Engineers across four offices
Clients delivered for
Offices: Charlotte, Chennai, Madurai, Singapore
27001 and 9001 certified
Trusted by leading enterprises and healthcare teams
