TL;DR: The ROI of fax triage software for a mid-to-large cardiology practice comes primarily from labor hours reclaimed on document intake — daily fax volume multiplied by minutes saved per document multiplied by loaded staff cost — with secondary returns from faster referral conversion, fewer scheduling delays on imaging orders, and fewer denials caused by documents filed late or to the wrong chart. Build the model on your own five inputs rather than a vendor calculator, haircut year one, and decide in advance what the recovered hours will be used for. Hours nobody reallocates show up as zero on the P&L.
The five inputs any honest model needs
If a savings calculator doesn't ask you for all five of these, it isn't modeling your practice. It's modeling a brochure.
- V — daily inbound volume, split by document type. Not one blended number. Echo reports, device transmissions, referrals, cath reports, and payer correspondence behave differently at every stage.
- T — current minutes of staff touch per document, by type. From your own measurement, not an industry average.
- A — automation rate. The share of documents in each type that reach the correct chart with no human touch, at the confidence threshold you set.
- R — residual review minutes. What the exceptions still cost after automation. Modeling this as zero is the single most common way these business cases get inflated.
- C — loaded hourly cost of the staff working the queue.
The arithmetic is simple once you have the inputs:
- Hours per day today = Σ (V × T) ÷ 60, summed across document types
- Hours per day after = Σ (V × (1 − A) × R) ÷ 60, plus QA sampling time
- Annual savings = (hours today − hours after) × operating days × C
Run it per document type and then sum. A blended average overstates savings on your hard documents and understates them on your easy ones, and the error compounds in whichever direction is least convenient.
Get your baseline from your own practice, not a benchmark
Spend one week having the people who work your fax queue log a start and stop timestamp per document, tagged by type and by who handled it. That's the whole exercise, and it's worth more than every published benchmark combined — because it's the number your partners will accept.
You'll see wide variation by category. A clean echo report for an established patient might be a two-minute touch. A 30-page hospital discharge packet for a patient with no chart in your system, arriving as a crooked scan, can eat fifteen minutes before anyone reads the clinical content.
Device transmissions deserve their own line. A time-and-motion study in CJC Open measured mean staff time per remote transmission at 9.4 to 13.5 minutes for therapeutic devices and 11.3 to 12.9 minutes for insertable cardiac monitors, with estimated annual management time of 1.6 to 2.4 hours per therapeutic-device patient and 7.7 to 9.3 hours per ICM patient. If you have a device panel, that's a substantial standing cost most practices have never priced.
Also count the junk. Marketing blasts, misdirected faxes, and blank cover pages typically run 10% to 20% of inbound volume, and every one of those is a document a person picked up, looked at, and discarded. Automating the discard is real time saved even though nothing gets filed.
How do you calculate loaded hourly cost without fooling yourself?
Base wage is the starting point, not the answer.
Take the base hourly wage of whoever actually works each document type, then gross it up for benefits. BLS Employer Costs for Employee Compensation put wages and salaries at 70.3% of total employer compensation cost for private industry workers in March 2025, with benefits making up the other 29.7%. Dividing a base wage by 0.703 gets you a defensible loaded figure.
Two adjustments people skip:
Use productive hours, not paid hours. PTO, training, and meetings mean a full-time employee delivers closer to 1,800 productive hours a year than 2,080. If you're converting saved hours into FTE equivalents, use the smaller denominator or you'll overstate the headcount equivalent.
Use the right person's wage per document type. If your device transmissions are worked by an allied health professional and your referrals by an RN, the front-desk wage understates the cost badly. Model each document type at the cost of whoever actually touches it — in cardiology this matters more than in most specialties, because the clinically-skilled staff carry a real share of the intake queue.
A worked example — illustrative inputs, not measured results
The numbers below are chosen to show how the model behaves. Replace every one of them with your own.
Take a mid-to-large cardiology practice receiving 140 inbound faxes a day:
- 40 echo, stress, and imaging reports at 5 minutes each
- 30 remote device transmissions at 11 minutes each
- 25 referrals with attachments at 13 minutes each
- 20 payer and prior auth documents at 9 minutes each
- 25 records requests, correspondence, and junk at 4 minutes each
Baseline: (200 + 330 + 325 + 180 + 100) ÷ 60 = 18.9 staff hours per day.
Now apply automation rates of 85% on imaging reports, 65% on device transmissions, 50% on referrals, 60% on payer documents, and 80% on records and junk — with residual review of 2, 5, 6, 5, and 2 minutes on what doesn't clear, plus a 5% QA audit of auto-cleared documents at one minute each.
After: about 190 minutes of exception review plus roughly 5 minutes of QA — call it 3.2 staff hours per day.
That's 15.7 hours a day recovered, roughly 83% of baseline. Across 250 operating days at a loaded cost that blends front-desk and allied-health rates, that's on the order of 3,900 hours a year — meaningful capacity in any practice, and a number you can defend line by line because you built it.
Treat 83% as the optimistic end. It assumes the automation rates above hold across your real senders, which they won't in month one.
If you're presenting this to partners, haircut year one by 20% to 30%. Let the upside be a pleasant surprise rather than a missed forecast.
The revenue effects operators leave out
Labor hours are the easiest thing to count and often not the most valuable thing you gain. Three second-order effects deserve their own lines.
Referral conversion. Measure hours from fax arrival to first patient contact attempt, before and after. A referral sitting three days is a patient who books somewhere else. Multiply the change in conversion rate by your average revenue per new-patient episode and the number frequently exceeds the labor savings on its own.
Imaging and device authorizations that miss their window. Cardiology carries one of the highest prior authorization densities of any specialty. A determination sitting unrouted for four days is a denial risk, and unlike a records request, nobody notices it's missing until the claim comes back. Estimate this from your own appeal rate and average claim value.
Documents that never get worked at all. Pull the age distribution of your current queue. Anything older than 72 hours is leakage you aren't currently pricing. Automated routing doesn't just move documents faster — it removes the failure mode where a document is never assigned to anyone.
Worth noting separately: an automated intake layer works overnight and on weekends. Monday morning starting at zero rather than at a Friday backlog is a scheduling benefit that never shows up in an hours-saved calculation.
The cost side, stated honestly
Three line items, and the third is the one vendors leave out.
Subscription cost. Usually priced per document, per provider, or per site. Get the pricing model in writing before you build the model, because per-document pricing changes the shape of the ROI curve as volume grows.
Implementation and EHR integration. Connecting the fax feed is quick. The EHR side — document filing, patient index lookup, task creation — is the variable. Practices on modern document APIs land at the short end; those needing HL7 interface work or agent-driven workflows take longer. Ask who does that work and who maintains it when your EHR ships an update.
Your own team's time in month one. Somebody experienced has to map document types to chart categories, name routing destinations, define the exception process, and correct mismatches while confidence thresholds settle. Budget 20 to 40 hours. Including this line is what makes your model credible to a skeptical CFO, because its absence is the tell that a calculator was built by a vendor.
Add a fourth if it applies: duplicate-record cleanup. If your patient index has duplicate charts, the matcher has no correct answer available, and that cleanup is real work whether or not you buy the software.
The pressure making this worth doing isn't going away. MGMA's 2026 Regulatory Burden Report found roughly 95% of practices reporting increased administrative burden over the prior three years, with 40% carrying multiple full-time administrative staff per physician. Honey Health sits in this category as one reference point for capability and pricing — an input to your model rather than its conclusion.
Two ROI traps worth naming before you present
Crediting the software with hours nobody reallocated. You rarely eliminate FTEs outright. What you get is capacity, and its value depends entirely on what you do with it — a hire you didn't make, overtime you stopped paying, a device-clinic backlog you finally cleared, referral follow-up that finally happens. Name the destination in the business case. If you genuinely intend to reduce headcount, put that in writing too, because savings nobody acts on show up as zero.
Modeling a straight-through rate the vendor only hits on the easiest document class. A vendor quoting 95% automation on clean lab results and one quoting 95% across a cardiology mix that includes device transmissions and multi-study packets are describing very different products. Ask for the rate by document type, on your documents, and model the hard categories conservatively.
There's a third worth mentioning: assuming accuracy is flat from day one. It isn't. Easy document types perform well early; multi-study packets, degraded manufacturer-portal scans, and patients with no prior record improve as the system sees more of your specific senders. Model the first 90 days at a lower automation rate and step it up.
Frequently Asked Questions
What payback period should a cardiology practice expect?
Most practices modeling honestly land somewhere between six and eighteen months, driven mainly by daily volume and document mix rather than practice size. Build your own number from the five inputs rather than accepting a vendor range, and present it with the year-one haircut applied so the forecast survives contact with reality.
How do we pressure-test a vendor's savings claim?
Ask three questions: what residual review time did you assume on exceptions, what automation rate did you assume by document type, and what implementation time did you charge to month one. A calculator assuming zero residual review and instant accuracy produces a number roughly double what you'll see. If a vendor won't show the inputs, treat the output as marketing.
Does fax triage ROI depend on practice size?
Less than you'd think. The math turns on total touch minutes rather than provider count. A practice with 60 faxes a day that are mostly device transmissions and referral packets can clear higher annual savings than one with 150 simple lab results. Run your own volume and time inputs before assuming you're too small.
Should we model headcount reduction or redeployment?
Decide before you present, because they show up differently on the P&L. Redeployment is the more common outcome and the easier one to defend — recovered hours going to referral follow-up, device clinic coverage, and prior auth work that used to sit. If you intend headcount reduction, state it explicitly rather than leaving it implied.
What automation rate is realistic in year one?
It depends heavily on document mix. Structured, recurring documents from consistent senders often clear at high rates within weeks. Multi-study hospital packets, degraded portal scans, and documents for patients with no chart take longer. A reasonable first-year plan assumes strong performance on your top two or three document types and modest performance on the rest.
How do we account for the exception queue in the model?
As a real, permanent line item. At a given automation rate, the documents that don't clear still need review time, and that time doesn't trend to zero. Estimate it per document type — a one-click confirmation costs far less than a genuine research case — and include a small QA sampling allowance for auditing auto-filed documents.
