Most oncology practices should buy oncology fax triage and document routing software rather than build it, because keeping document classification accurate across constantly shifting payer, lab, and referral fax formats takes ML engineering that few practices can staff full-time. Building buys you control; buying buys you speed, a maintained classifier, and a lower total cost at the volume most independent and mid-size groups handle. Very large multi-site oncology organizations are the exception — not the rule.
Why fax is still the front door in oncology practices
Ask any front-office lead at an oncology practice what their day looks like, and fax comes up in the first two minutes. Referrals arrive by fax. Pathology reports arrive by fax. Prior authorization decisions, lab values, and imaging summaries arrive by fax. Roughly 89% of healthcare organizations still operate fax machines, and 56% of referrals still move through fax rather than an electronic channel.
Why hasn't this changed? Interoperability gaps are the honest answer. Different EHRs, different payer portals, different reference labs — fax is the one channel that works with everyone, so it never goes away. Oncology practices feel this more than most specialties do. Oncologists carry some of the highest inbox and message volumes in medicine, driven by frequent lab reviews, imaging updates, and treatment plan changes, according to a national study of oncology EHR inbox trends published in a peer-reviewed informatics journal.
So what happens when that volume hits a front desk staffed for a normal day? Documents get misfiled. Urgent lab values sit in a queue behind routine correspondence. Industry data suggests 30% of medical tests get reordered because the original results were lost or buried in a fax backlog — a number that should worry any CFO running an oncology P&L, since redundant testing burns both money and patient goodwill.
Prior authorization compounds the problem, and oncology carries a heavier PA load than almost any other specialty because of the drug-approval requirements tied to chemotherapy and immunotherapy regimens. The 2024 CAQH Index found that a manually processed prior authorization costs practices roughly $3.41 per transaction, against $0.05 for a fully automated one — and a large share of that manual cost is staff time spent chasing down faxed documentation that never arrived or arrived illegible.
This is the specific problem oncology fax triage and document routing software is built to solve: sorting incoming faxes by type and urgency, matching them to the right patient chart, and routing them to the right queue without a human reading every page first. The question isn't whether a practice needs this capability. It's whether to build the classification engine in-house or buy one that's already built and maintained. That decision has real financial weight, and it deserves more than a gut call.
Build vs. buy: what each path actually costs
Building your own fax triage system isn't a weekend project — it's an ongoing ML program disguised as a document management tool. What does that actually mean in dollars and headcount? Fax formats change constantly: payers redesign their cover sheets, labs update their report layouts, referring practices switch EHRs and their outputs shift with them. A classifier trained on last year's formats degrades quietly, and nobody notices until a misrouted document causes a missed treatment window. Who catches that before it becomes a clinical problem instead of an IT one?
Here's what the build path typically requires:
- Model development — training and validating an OCR/ML classification pipeline against oncology-specific document types (pathology, labs, imaging, prior auth, referrals)
- Ongoing accuracy tuning — a dedicated engineer (or team) retraining models as payer and lab formats change, indefinitely
- EHR integration and maintenance — building and maintaining the interface layer so routed documents land in the right chart field, then re-testing it every time the EHR vendor pushes an update
- HIPAA and security overhead — a HIPAA risk analysis alone runs $5,000 to $30,000-plus per engagement with an outside consultant, and that's before ongoing risk management, which HHS's Office for Civil Rights made clear in its 2025 enforcement priorities is now scrutinized separately from the initial analysis
- Staff to babysit exceptions — someone has to review low-confidence classifications every single day, because no model hits 100%
Buying shifts that cost structure. You pay a subscription instead of carrying engineering headcount. The vendor's classifiers get updated across their whole customer base when a payer changes a form, not just for you. Time-to-value is measured in weeks, not the 12-to-18-month build-and-tune cycle typical of custom ML projects. The tradeoff is real too — you get less bespoke control over exactly how documents route, and you're dependent on a vendor's roadmap. But for most practices, that tradeoff is worth it: the total cost of ownership on custom healthcare software routinely runs two to three times the initial estimate once integration and maintenance are counted, a pattern documented across healthcare IT build-versus-buy analyses.
A decision framework: four questions before you commit
How do you actually decide, instead of guessing? Four factors do most of the work.
What's your fax volume? A single-site oncology practice processing a few hundred documents a week doesn't generate enough scale to justify a dedicated ML team. A large MSO or academic-affiliated network processing tens of thousands of documents monthly across multiple sites starts to change that math — the fixed cost of a build spreads across more volume.
Do you have in-house engineering capacity today, not hypothetically? Building a document classifier isn't a side project for your EHR analyst. It requires people who've shipped ML systems before. If your IT team is two people who also handle helpdesk tickets, you already have your answer.
How complex is your EHR integration? Oncology-specific EHRs (and the handful of instances layered on top of them) each have their own quirks for where routed documents need to land — which flowsheet, which document type tag, which encounter. A vendor who's already built connectors for your EHR removes months of integration work; if you're on a rare or heavily customized instance, that calculus shifts.
How fast do you need production-grade accuracy? A 90% classification rate sounds good until you calculate what the other 10% costs in missed pathology results or delayed prior auth documentation. Buying gets you a model that's already been tuned against real-world payer and lab formats. Building means you accept a rocky accuracy curve while your own model learns — on your patients' documents.
Weigh those four factors honestly, and the answer usually falls out on its own. Most oncology fax triage and document routing software evaluations come down to volume and engineering capacity more than anything else on the vendor's feature list.
Where the buy path leads — and how Honey Health fits
Once a practice decides to buy, the market splits into two camps: generic OCR/RPA tools repurposed for healthcare, and vendors built specifically for medical fax and document workflows. Does that distinction actually matter, or is fax classification fax classification no matter who built it? It matters quite a bit. Generic OCR wasn't trained on oncology-specific documents — it doesn't know the difference between a pathology addendum and a routine lab panel, and it won't route a stat critical value any differently than a refill request. So how do you tell the two categories apart during a vendor demo?
Honey Health's fax triage agent is one example of the purpose-built category: an AI agent that classifies incoming faxes by document type and urgency, matches them to the correct patient record, and routes them into the right queue in the practice's EHR — without a human pre-sorting every page. It's built as a subscription service, so the classifier improvements a vendor makes for one oncology customer roll out to every customer, rather than sitting in one practice's private codebase waiting for a developer to get around to it.
That's the general shape of what buying gets you across any vendor in this category: less bespoke control in exchange for a model that's already been trained on real oncology fax volume, maintained without your headcount, and live faster than an internal build. Practices evaluating vendors here should ask the same four questions from the decision framework above — volume, capacity, EHR complexity, and speed-to-accuracy — of each vendor's actual track record, not just their pitch deck.
The honest take: when building actually makes sense
Building isn't always the wrong call — it's just the wrong call for most oncology practices. So who is it right for? A handful of very large, PE-backed MSOs or academic cancer networks have both the fax volume and the engineering bench to make a custom build pencil out. If you're processing enough documents that a 2-point accuracy improvement saves real staff hours across dozens of sites, and you already employ ML engineers for other purposes, the marginal cost of a document classifier is lower for you than it is for everyone else.
But that's a narrow band. Most independent oncology groups and mid-size specialty networks don't have ML engineers on staff, don't want to hire them for this specific problem, and can't justify a 12-to-18-month build cycle when a vendor solution is already trained and running. What's the actual downside of admitting that and buying instead? Mostly pride — the financial math rarely argues for building. For those practices, buying isn't a compromise — it's the financially rational choice. The document classification problem doesn't go away once you buy either; it just becomes someone else's job to keep current, which is exactly what a subscription for oncology fax triage and document routing software is paying for.
Frequently Asked Questions
What is oncology fax triage and document routing software?
It's software that automatically reads incoming faxes — referrals, labs, pathology reports, prior authorization decisions — classifies them by document type and urgency, and routes them to the correct patient chart or staff queue without a person manually sorting each page first. It's built specifically for the document types and urgency levels common in oncology, unlike generic office fax tools.
How much does it cost to build fax triage software in-house?
Costs vary by scope, but a custom build typically includes ML model development, an EHR integration layer, a HIPAA risk analysis ($5,000–$30,000-plus per engagement), and ongoing engineering time to retrain models as fax formats change. Total cost of ownership on custom healthcare software commonly runs two to three times the initial estimate once integration and maintenance are factored in.
Is fax still a major channel for oncology referrals and lab results?
Yes. Roughly 89% of healthcare organizations still use fax, and 56% of referrals move through it rather than an electronic channel. Oncology practices see this acutely because of high lab, pathology, and imaging message volume relative to other specialties.
How long does it take to get a fax triage vendor live?
Vendor implementations for purpose-built fax triage tools typically run weeks, not months, since the classification models are already trained and the EHR connectors are usually pre-built for common oncology EHR instances. Custom in-house builds typically take 12 to 18 months to reach production-grade accuracy.
Should a small oncology practice ever build its own fax triage system?
Generally no. Building makes financial sense mainly for very large, multi-site MSOs or academic cancer networks with existing ML engineering staff and enough fax volume to spread the fixed cost of a build across many sites. Independent and mid-size practices almost always come out ahead buying a vendor-maintained solution.
What happens if a fax triage vendor misclassifies a document?
Reputable vendors build in confidence scoring — low-confidence classifications route to a human review queue rather than being auto-filed, and misclassifications get fed back into the model to improve accuracy over time. This is one of the practical advantages of buying: the vendor's model improves from every customer's exception cases, not just your own.
