Medsender
Medsender is a New York based healthcare workflow automation company whose product reads the documents and phone calls arriving at a medical practice and files what it finds into the electronic health record. It began in 2014 as a HIPAA compliant digital fax service and the automation layer came later, which is still visible in the product shape: fax transport, document editing, electronic signature and storage sit underneath, and the artificial intelligence is sold on top of them in named tiers.
The platform has four parts. Fax automation reads each inbound fax, titles it, categorises it, tags it and uploads it to the correct patient chart, with new chart creation where none exists. Referral management classifies and routes inbound referrals into a dedicated workflow before a coordinator touches them, and reports status back to the referring provider. The AI Medical Assistant, branded MAIRA, answers inbound patient calls, schedules appointments, handles routine questions and produces recordings, transcripts and call summaries. A developer fax interface exposes the same extraction to other health technology companies through typed software development kits, webhooks and sandbox access.
Integration is the strongest published surface for a company of this size. Named ambulatory electronic health record connections include athenahealth, eClinicalWorks, ModMed, NextGen, Practice Fusion and Ezderm, and an integration guide is published.
The company raised a 5 million dollar Series A led by Ballast Point Ventures, announced January 2025, and is led by co chief executives Zain Qayyum, who founded it, and Salman Haque. Sources disagree on the founding year: company registries record 2014 while the founder dates it to 2015 in interviews.
Extraction accuracy is claimed at 99 percent, but the figure appears as a customer testimonial in marketing copy rather than as a measured result, with no test set, denominator, document mix or independent validation. No model is named or described anywhere in public material.
Capability Axes
An AI Health Index grade measures what a buyer can verify from public sources on the date shown. It is not a rating of how good the product is. A vendor can build an excellent system and grade low on an axis because it publishes nothing an outsider can check. How grades read
The automation is what the company sells and what its pricing tiers are named after, and the marketing describes proprietary automation and custom trained models. The mechanism is a document classification and entity extraction pipeline across fax and voice channels, plus a conversational agent on the telephone line.
Held below A for two reasons. The business began in 2014 as a HIPAA compliant fax service, and the transport, document editing, electronic signature and storage layers are ordinary software that would function with no model at all. And no model is named or characterised anywhere in public material, which separates this record from its closest competitors that do name one, such as the DextractLM model behind 314e Dexit and the RaeLM vision language model behind Tennr.
Autonomy is high and the oversight architecture is unpublished. Inbound documents are read, classified, matched to a patient and written to the chart with the stated aim that no member of staff reads them first, and the voice agent conducts complete patient conversations including scheduling.
A customer testimonial refers to a handful of exceptions needing review, which implies an exception queue exists, but no confidence threshold, routing rule or escalation criterion is published. Nothing published describes what the voice agent does when a caller reports a clinical emergency or acute distress, or whether it identifies itself as artificial at the start of a call.
No model, architecture or training approach is described. The developer interface advertises custom trained models without saying what they are trained on, by whom, or against what.
The single quantitative claim, 99 percent extraction accuracy, is carried as a customer quotation in marketing copy with no test set, no denominator, no document mix and no independent check. A buyer evaluating this product has nothing to assess other than a demonstration on its own documents.
The published control detail is specific rather than generic, which is rarer in this segment than it should be. Data is stated to be encrypted in transit and at rest with the strengths named rather than merely asserted, the security page states that patient data is never transmitted or stored in an unencrypted state, and role based permissions, inactivity timeouts, centralised user management and audit trails are each described.
Servers are stated to be in the United States, which settles residency without requiring a separate assurance. Naming key strengths and stating an absolute about unencrypted handling are both checkable in a way that a general claim of encryption is not. The unanswered question is training and the product's own developer interface raises it.
Custom trained models are advertised while nothing published states what they are trained on, whether customer protected health information contributes, or what de identification applies. For a product whose entire input is patient documents and recorded patient calls, that is not a peripheral question: the corpus and the customer data are plausibly the same thing, and a custom model trained for one customer on that customer's documents raises the further question of whether it stays with them.
No model or provider is named and no sub processor list was located. Ask whether customer documents train or tune models, whether a custom model is exclusive to the customer that funded it, and for retention on documents and call audio.
Operational claims are specific and several attach to named customers, which is better than the category norm. Anne Arundel Dermatology is named as having doubled referral conversion, and the company publishes figures for two to three times more referred patients scheduled, a 15 percent increase in patients scheduled through automated scheduling, and referral processing falling from seven days to same day.
All are vendor published. None carries a baseline, a measurement period or a denominator, and no independent or peer reviewed evaluation was located. This is an administrative product, so clinical outcome evidence is not the relevant test, but operational claims of this size are directly measurable and the measurements are not shown.
The published control detail is specific rather than generic. Data is encrypted in transit at 256 bit and at rest under a 2048 bit private key, the security page states that patient data is never transmitted or stored in an unencrypted state, and role based permissions, inactivity timeouts, centralised user management and audit trails are all described. Servers are stated to be in the United States.
The unanswered question is training. The developer interface advertises custom trained models while nothing published states whether customer protected health information is used to train or improve them. For a product whose entire input is patient documents and recorded patient calls, that is the question to settle in writing before signing.
A signed business associate agreement is included with every account rather than gated to an enterprise tier, and the company states this plainly on its published security page. That is a materially stronger position than the common pattern of offering an agreement on request or above a price threshold, because the smallest practice gets the same contractual footing as the largest customer.
HITRUST certification adds independent validation of a control set mapped to HIPAA, which is the closest thing to third party assurance available in a framework where no body certifies compliance itself.
Among the strongest published postures in this category. SOC 2 Type II was completed in July 2025 and the report type is named rather than left as an unqualified SOC 2 claim, HITRUST certification is stated with independent assessor validation, and continuous compliance monitoring is attributed to a named platform rather than described vaguely.
A trust centre exists and both the Type I and Type II reports are made available on request. Single sign on with Microsoft is supported. The gaps are that no penetration testing statement was located and that the reports sit behind a request rather than a self service portal.
No device pathway applies and none is claimed. The product moves administrative documents and handles patient communication, and does not make or support a diagnostic or treatment decision, so the absence of a clearance is correct rather than a gap.
The regulators that do matter are health privacy enforcement under HIPAA and, for the telephone agent, telephone consumer protection rules, which have been enforced against healthcare callers. Nothing published addresses consent for automated calling, or what a patient is told about speaking to an artificial agent.
Nothing published on model evaluation, error rates, subgroup performance, or how a mistake is detected and corrected once a document has already been filed to a chart.
The exposure is concrete rather than theoretical. Speech recognition accuracy varies with accent, dialect, first language, age and emotional state, and a practice telephone line reaches exactly that spread of callers. Document extraction carries its own version of the same problem, since handwriting, poor fax quality and non standard forms are unevenly distributed across referring practices, and a referral that extracts badly is a patient who waits longer. No subgroup performance data was located.
Two passes located no model, architecture or training approach, no evaluation methodology, no published limitations and no warranty, indemnity or remediation commitment. The developer interface advertises custom trained models without saying what they are trained on, by whom, or against what, so a buyer cannot establish what kind of system they are integrating. The single quantitative claim deserves naming for its form as well as its content.
A ninety nine per cent extraction accuracy figure is carried as a customer quotation in marketing copy, with no test set, no denominator, no document mix and no independent check. A performance figure attributed to a customer rather than measured by the vendor is a weaker artefact than a vendor figure, not a stronger one: the customer had no obligation to define a method, the vendor has not adopted the number as its own claim, and neither party can be held to it.
A buyer therefore has nothing to assess other than a demonstration on their own documents, which is worth doing and is not a substitute for a stated error characteristic. The document mix matters more than the headline for extraction products, since a faxed handwritten referral and a clean electronic record are different problems. Ask for accuracy by document type and quality band, the test set, and what happens to an extraction the system is unsure of.
Integration breadth is unusually good for a company of this size and it is published rather than asserted. Named connections include athenahealth, eClinicalWorks, ModMed, NextGen, Practice Fusion and Ezderm, an integration guide is public, and write back into the chart including creation of a new patient chart is described as automatic.
The developer interface adds typed software development kits, webhooks and sandbox access so other health technology companies can consume the same extraction. Held at B because the depth sits on the ambulatory specialty side, no Epic or Oracle Health integration was located, and the write back mechanism is not described at the standards level.
Cloud only, with servers stated to be in the United States. That answers the residency question cleanly for a domestic buyer and forecloses it for anyone else. Auto scaling infrastructure and a 99.99 percent uptime figure are published.
There is no on premise or private cloud option, no region selection, and no published retention schedule for documents, call recordings or transcripts. Retention matters more here than in most records, because inbound faxes and recorded patient calls accumulate continuously and raw audio is identifying independently of what is said.
Genuinely mixed. The consumption mechanism is published: 1,500 fax pages included at the entry tier with additional pages at 10 cents, and the developer interface at 3,000 pages included with additional pages at 3.5 cents. A pricing calculator and a return on investment calculator are both public, and the company advertises no auto renewal traps and no surprise increases.
What is absent from the vendor's own pricing page is the subscription price of the automation tiers, which resolve to a demonstration request. Those prices do reach buyers, but through a third party software directory. A price a vendor allows a directory to publish while omitting it from its own page cannot be relied on as current.
Primarily ambulatory specialty practices, with dermatology the most frequently named and the case studies concentrated there, extending to multi location groups. The company states it also serves hospitals and health systems, though no named health system deployment was located.
A second and quite different buyer exists through the developer interface, where health technology companies embed the extraction inside their own products. Functional coverage spans the front office, health information management and referral coordination, with the telephone agent reaching the patient directly.
Pricing
Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.
| Entry Price | Pricing Basis | BAA Tier | Implementation | Source |
|---|---|---|---|---|
|
299 dollars per month, AI Fax Automation tier
$299 baseline
|
Flat monthly subscription per named tier, plus per page overage on fax volume. Three product tiers are listed by a third party directory: AI Fax Automation at 299 dollars a month, AI Referral Automation at 549, AI Call Automation at 2,249, with Enterprise on quotation. The developer fax interface is priced separately on included volume with per page overage. | Included with every account at all tiers, stated on the published security page | None published. Onboarding and training are described as included from the Pro tier upward, with expedited support at Enterprise. Most teams are stated to be operational in two to four weeks. | Third Party Estimated |
The transparency here splits in an unusual way and buyers should know which half they are reading. The consumption mechanism is published by the vendor: 1,500 fax pages included at entry with additional pages at 10 cents, and 3,000 pages included on the developer interface with additional pages at 3.5 cents. The subscription price of the automation tiers is not on the vendor's own pricing page, which resolves to a free trial, a demonstration request and a contact form.
The tier prices recorded here come from a third party software directory rather than from the vendor, so they may be stale and should be confirmed in a quotation. A free trial exists but no free version.