How routing works

Decide what each page needs before you pay for it.

Most inbound fax pages don’t need an expensive model pass. InfoNotData’s router makes a confidence decision on every page in milliseconds and sends it down one of three lanes: Easy, Medium, or Hard.

  • Routing in milliseconds per page
  • Expensive passes only where needed
  • Hard pages go to a person
Illustration of an inbound queue: a cover sheet and a typed lab result routed Easy, an echo report routed Medium for targeted extraction, and a handwritten note routed Hard to a person.Illustration · synthetic queue, not patient data

01 / Why route first

Treating every page the same is the expensive mistake.

High-volume queues are mostly pages that are easy to handle, a smaller share that need a closer read, and a few that no machine should guess at. Routing puts each page where it belongs.

Every page, full price

Paying for pages that don’t need it

A full model pass over cover sheets, duplicates, and typed results costs the same as one over the page that matters.

Slower

Latency that adds up

Running heavy processing on every page of an 846-page packet takes time the clinician doesn’t have.

False confidence

Guessing on bad scans

A pipeline that treats a blurry handwritten page like a typed one will guess. Guesses erode clinical trust.

02 / The three lanes

Easy. Medium. Hard.

Every page lands in exactly one lane, and the lane decides what happens next.

Easy · auto-processed

High confidence, fast path

Pages the router is confident about are processed automatically, without a heavy model pass.

Medium · extraction-only

A targeted, cited pass

Pages that need a closer read get a targeted extraction pass. It pulls values only, each with its page and line.

Hard · human review

A person decides

Low-confidence pages, like handwriting or a poor-quality fax, go to a person on your team. Never a guess.

Routing illustration · inbound fax queue
Easy · auto-processed 7,214 pages in this batch

High-confidence pages are processed automatically on the fast path.

Medium · extraction-only 1,906 pages in this batch

Pages that need a closer read get a targeted extraction pass. Values only, each with its source.

Hard · human review 312 pages in this batch

Low-confidence pages, like handwriting or a poor-quality fax, go to a person. Never a guess.

03 / What it saves

Filter before you spend.

Design target
Up to 80%

lower OCR overhead, by design

Advanced pre-processing and a model selector decide which pages need which pass, so expensive processing runs only where it adds something.

Speed
≤ 5 s

retrieval, by design

InfoNotData is engineered for retrieval in five seconds or less, so a clinician gets the value while the patient is still in front of them.

Trust
Hard lane

uncertainty goes to people

Knowing what it doesn’t know is a feature. The Hard lane is how InfoNotData avoids false confidence.

Filter before you spend

Size it for your own queue.

Put in your volume and your rates. Every assumption is on the screen, so you can take the math to your CFO.

InfoNotData’s router decides page by page which pass each page needs. The share you set here is your estimate, not a promise from us.

Example values. Replace them with yours. Nothing you enter leaves this page.

60%
Full-pass spend today
$20,000 / month
Spent on pages that didn’t need it
$12,000 / month
Staff time spent searching
1,000 hours / month
Hours back at the projected 25–35% reduction
250–350 hours / month
Projected · proposed single-site VHA pilot
Check these numbers with an engineer

Questions about routing

What teams ask about the router.

What does the router decide?

How confident InfoNotData can be about each page, and therefore which lane it belongs in: Easy for automatic processing, Medium for a targeted extraction pass, or Hard for a person.

How fast is routing?

Milliseconds per page. The decision happens before any expensive model pass runs.

Can we change what goes to a person?

The Hard lane exists so your team reviews what the system isn’t sure about. We’ll go through how review works for your queue in the walkthrough.

Is “up to 80% lower OCR overhead” a measured result?

It’s a design target based on pre-processing and model selection, not a measured benchmark on your queue. Use the calculator above with your own numbers.

06 / Next step

See the router on your kind of packet.

A 30-minute walkthrough with an engineer. We’ll route a synthetic or de-identified packet and show you which pages land in each lane, and why.

What happens next

  1. An engineer replies to set a time that works for your team.
  2. A 30-minute walkthrough on a synthetic or de-identified packet.
  3. If it fits, we scope a pilot on your own queue.

An engineer, not a sales script. We’ll show you what it doesn’t do, too.

Tampa, Florida
Where do your faxes land?
Organization type
Pages a month. A rough estimate is fine.
Who should we talk to?
Anything we should prepare?
For example: your specialty, your EHR, or your security review timeline.

Business contact details only. Please don’t include any patient information.