Quick answer: Gastroenterology practices automate inbound fax triage by routing their existing fax line through an AI agent that reads and classifies each document, auto-files the routine majority directly into the EHR, and escalates only exceptions to staff. Done properly, that removes 70% to 90% of manual touches from the queue without changing the fax number referring offices already use and without hiring anyone. The work that remains is exception review, which is a fraction of a role rather than a whole one.
Why "just hire someone" stops working in a GI practice
The instinct when the fax queue backs up is to add a person. Sometimes that's right. In GI it usually isn't, and the reason is arithmetic rather than ideology.
Fax volume in a gastroenterology practice doesn't scale with staff — it scales with referral sources, scope volume, and the number of IBD patients on biologics. Each of those grows independently of your ability to hire. Add a provider and you add referring offices, scopes, pathology reports, and prior auth correspondence in the same quarter.
Hiring into that is a treadmill, and it's an expensive one. MGMA found practices reporting operating cost increases averaging about 11% year over year, driven primarily by labor. Meanwhile front-desk and entry-level administrative roles remain among the most common churn points in medical groups heading into 2026 — and document sorting is exactly the kind of work people leave over.
So you're hiring into a role with high turnover, to do work that grows faster than your headcount budget, in a labor market where demand for these roles is projected to grow 15% through 2033. The automation case isn't that people are expensive. It's that this particular job doesn't stay filled.
Baseline your queue before you buy anything
Every vendor has an ROI calculator tuned to produce a favorable answer. Build your own. It takes two weeks and it's the only version your partners will accept.
Count three things:
- Daily inbound volume, by category. Tally pathology, referral packets, prior auth correspondence, labs, outside procedure notes, and records requests separately. The mix matters as much as the total, because different categories automate at different rates.
- Handling time per document. Time twenty documents end to end — open, identify, patient lookup, chart location, upload, task creation. Most GI practices land between 90 seconds and four minutes, with multi-page referral packets running far longer.
- Where things currently break. How many documents sat more than 48 hours last month? How many referrals aged out before anyone called the patient? How many pathology reports arrived without a recall being set?
That third number is the one nobody has and everybody needs. It's also the number that makes the case, because it converts a labor argument into a revenue argument.
Route your existing line — don't replace it
The single biggest adoption barrier practices imagine is the one that doesn't exist: they assume automating fax triage means changing the fax number.
It doesn't. There are two paths, and both leave your published number intact.
Port the numbers. The triage platform takes ownership of your existing fax DIDs. Inbound documents land in the AI pipeline first, get processed, then file into the EHR. Nothing changes for the sender.
Sit behind your current service. If you're on a cloud fax provider you're happy with, the triage layer consumes documents from it through an API, a monitored inbox, or a virtual printer. Your fax service keeps doing transport; the AI layer does comprehension.
Either way, the 340 referring offices, three labs, and eleven payers who fax you don't do anything differently. That matters more than it sounds, because a referral source asked to update a fax number is a referral source with a reason to send the patient somewhere else.
Define GI document categories before turning anything on
Generic categories produce generic results. The routing rules are where a GI deployment either works or doesn't, and they should be built from your actual document mix rather than the vendor's default template.
At minimum, a gastroenterology practice needs these as distinct categories:
- Pathology reports — routed to the performing endoscopist for sign-off, with the diagnosis and recommended surveillance interval extracted and pushed to whatever drives your recall process.
- Referral packets — split into component documents (referral form, H&P, medication list, outside scope report, insurance card), each indexed separately, with a scheduling task created and incomplete packets flagged for callback rather than silently filed.
- Prior auth correspondence — approvals, denials, step-therapy notices, and peer-to-peer requests, each linked back to the original request and put on a clock.
- Lab results — filed to results with a review task for the ordering provider.
- Outside procedure notes and hospital records — filed to the chart, flagged when they change the clinical picture for an upcoming visit.
- Records requests — routed to whoever handles release of information, away from clinical queues entirely.
The point of separating these isn't tidiness. It's that each one has a different owner and a different urgency, and a single "incoming faxes" inbox flattens both.
Run shadow mode before you cut over
This is the step practices skip and the step that determines whether staff trust the system.
In shadow mode the AI processes every inbound document — classifies it, matches the patient, proposes a filing location and a task — but writes nothing to the EHR until a human confirms. Staff do roughly the work they did before, except now they're reviewing a proposal instead of starting from a blank PDF.
Three to four weeks is the right length. What you get out of it:
- Accuracy tuning on your real document mix, not a vendor demo set
- A measured straight-through rate you can use in the business case instead of a vendor estimate
- Identified problem sources — the referring office whose faxes are always upside down, the lab whose format the system keeps missing
- Staff who watched it work for a month before being asked to rely on it
That last one is the real product of shadow mode. One visible misfile in week one of a hard cutover sets adoption back further than a month of parallel running costs.
Honey Health's fax triage agent is designed to run this way — shadow first, with every classification and candidate patient match surfaced for confirmation, then progressive cutover category by category as confidence in each one is established. Start with labs and pathology from your regular sources, which stabilize fastest, and hold referral packets and handwritten documents for last.
Decide where the recovered hours go before you start
Automating the queue is the easy part. The part practices underestimate is what happens on the other side.
When four or five hours a day come off the fax queue, that time does not convert into value on its own. It diffuses. The practices that get the most out of this decide in advance, in writing, where the hours go.
For a GI group, the highest-return redeployments are usually:
- Referral conversion. Calling new referrals the day they arrive instead of three days later. This is where the revenue is, and it's measurable against your current time-from-referral-to-scheduled-appointment.
- Prep compliance outreach. Fewer cancellations and fewer inadequate preps, both of which cost you a procedure slot.
- Prior auth follow-up. Working the biologic auth queue actively rather than reactively.
- Surveillance recall. Chasing the patients whose interval came due and who never got called.
Be honest with your partners about which version you're proposing. "We eliminate a position" and "we redeploy this person onto referral conversion" are different business cases with different numbers and different internal politics. Picking one before the rollout separates a project that delivers from one that produces a tidier inbox and no measurable change.
What this costs against what a hire costs
The comparison practices actually run is automation versus an FTE, so it's worth putting both on the same line.
A front-office hire working the document queue costs base wage plus benefits plus payroll tax — typically 1.25 to 1.4 times base. At a $20 base wage that's roughly $52,000 to $58,000 fully loaded, before recruiting cost, onboarding time, and the productivity gap while they learn your referral sources and your EHR's filing conventions. Then factor the replacement cost when they leave, which for a frontline administrative role runs well into five figures once the vacancy period is counted.
Fax triage software prices per document or per seat, and for a mid-sized GI group typically lands in the low four figures monthly. Against one loaded FTE, that's usually a clear margin even before second-order effects.
The costs buyers forget, and should budget:
- Implementation hours from your own team — a practice administrator and whoever owns the EHR relationship, four to eight weeks
- The parallel-run month, where you pay for the software while still doing the work manually
- Integration effort, which ranges from an API connection to a custom interface build with a one-time fee
- Exception-queue staffing, which is real residual labor at 15% to 25% of prior volume
What tips the comparison isn't the labor line, though. It's that software capacity scales with volume at a flat marginal cost, while a hire caps out and then you're hiring again. A GI group adding a third endoscopist doesn't need a fourth document clerk if the queue is automated.
Frequently Asked Questions
How much of our fax volume can realistically be automated?
Model 75% to 85% straight-through processing at steady state for a typical GI document mix, not 100%. The residual is handwriting, multi-patient batch faxes, poor scans, and anything clinically urgent that should escalate by design. A vendor quoting above 95% across your entire mix is describing their best category rather than your queue.
Will we have to lay off staff?
Most practices don't, and the ones that do usually didn't need to. The more common outcome is attrition absorption — when the person working the fax queue leaves, you don't backfill, and the exception queue gets absorbed into an existing role. Given front-office turnover rates, that opportunity tends to arrive on its own within a year.
How long does implementation take?
Four to eight weeks for most multi-provider GI groups, with shadow mode accounting for three to four of those weeks. EHR integration depth is the biggest variable: a documented API connection lands at the short end, a custom interface build at the long end. Budget real internal hours from a practice administrator during setup.
What happens to urgent documents during automation?
They should escalate, not auto-file. Any triage system handling clinical documents needs a category for time-sensitive items — abnormal findings, urgent consult requests, prior auth denials with appeal deadlines — that routes to a human queue with priority flagging rather than filing silently into a chart. Confirm this behavior explicitly in a demo.
Does this work if we run multiple locations on one EHR?
Yes, and the return usually improves. Multi-site GI groups typically carry duplicated document-handling labor at each location, which consolidates once triage is centralized. Routing rules can still direct documents to the correct site's work queues. Implementation effort rises somewhat, but not proportionally to the number of locations.
