How to choose an AI receptionist for an athenahealth practice

Why the obvious way to evaluate this fails

Most practices evaluate AI phone vendors by taking three demos and picking whichever sounded most natural on the call.

Voice quality is the easiest thing to demo and the least predictive of whether a deployment succeeds. Every serious vendor in this category has acceptable voice quality now. What differs is what happens after the caller stops talking.

Three things predict whether a deployment works, and none of them appear in a scripted demo.

• Whether the system writes to your EHR or creates a queue. A vendor whose agent takes a request and emails your front desk has moved work, not removed it. This is the largest difference between vendors and the hardest to see in a demo, because a demo of a queue looks identical to a demo of a real write
• Whether the HIPAA posture survives being asked about. Voice recordings are protected health information. Almost every vendor says HIPAA compliant on their website. Far fewer will put the encryption standards, the supplier coverage and a no training commitment in writing before you sign
• What happens at the edges. Every vendor demos the happy path. What the system does when no slot matches, when identity cannot be resolved, when the caller interrupts, and when the EHR returns an error is what sets your actual containment rate

The ten criteria for choosing an AI receptionist for athenahealth

One hundred points, weighted by how much each criterion affects whether work is actually removed.

1. EHR write depth, 20 points

The highest weighted criterion, because it decides whether anything is automated at all.

• Full marks: creates, modifies and cancels appointments directly in athenahealth during the call, creates clinical inbox tasks for refill requests, and writes a call summary to the patient chart, with no staff review step in between
• Partial: reads the schedule and tells callers what is available, but a person completes the booking
• Zero: takes the request and forwards it to an email inbox, a portal, or a task queue for staff to process

How to verify: ask the vendor to book an appointment on the demo and then show you the record in athenahealth. Then cancel it and show the slot released. A vendor operating a queue cannot do this live.

2. Marketplace Partner verification, 15 points

• Full marks: verified athenahealth Marketplace Partner, with a listing URL you can open
• Partial: in the Marketplace review process, with a stated timeline
• Zero: claims athenahealth integration with no listing

Why it carries this much weight: Marketplace review covers the technical implementation, the security posture and HIPAA readiness. It also means official endpoints and advance notice of breaking changes. An unofficial integration finds out about an API change when it breaks at a customer practice, during business hours.

How to verify: ask for the URL. Thirty seconds.

3. HIPAA posture, 15 points

• Full marks: a signed written agreement covering PHI before the first live call, named encryption standards at rest and in transit, every supplier that can touch patient data held to the same obligations with that list available for your compliance file, a written commitment that patient data is not used to train general purpose models, and audit logs you can export yourself
• Partial: the agreement is available but gated to a higher tier, or supplier coverage cannot be confirmed
• Zero: no written agreement, refusal to confirm supplier coverage, or a reserved right to train on customer data

There is no HIPAA certificate and no agency that issues one, which is exactly why this criterion is scored on what a vendor will put in writing rather than on what appears in their footer. The detailed version of this scoring, question by question, is on the HIPAA buyer guide.

4. Edge case behavior, 10 points

• Full marks: the vendor can say specifically what happens when no slot matches the request, when identity cannot be resolved, and when the EHR returns an error, and has a real answer for what happens when the agent and the front desk try to book the same slot at the same moment
• Partial: specific answers on some, vagueness on others
• Zero: vague throughout, or a claim that these cases do not arise

How to verify: ask for a demo of the failure path rather than the happy path. Request a time you know is unavailable.

5. Containment rate on live deployments, 10 points

• Full marks: a specific containment figure from live production in your specialty, with launch performance distinguished from steady state
• Partial: a number without specialty context, or without that distinction
• Zero: no containment figure, or no live deployments

Why the distinction matters: containment at launch runs well below steady state in every deployment. A vendor quoting only their best number is not lying, but you should know which number you are being given.

6. Configuration depth, 10 points

• Full marks: call flows are configured to your appointment types, provider preferences, intake questions and transfer rules during deployment, and the vendor does that work
• Partial: template based, with limited customization
• Zero: one configuration for everybody

In production, configuration quality moves containment more than the underlying model does. A well configured deployment on an ordinary model beats a generic deployment on an excellent one.

7. Transfer rule granularity, 5 points

• Full marks: separate rules for business hours and after hours, set per call type, with configurable escalation for urgent calls
• Partial: one set of transfer rules
• Zero: no transfer configuration, or a system that does not transfer

A system that never transfers is a warning sign rather than a feature. It means the vendor is either handling only trivial calls or letting the agent operate outside its competence.

8. Pricing model fit, 5 points

Not cheapest. Fit.

• Full marks: the pricing model matches your growth. Per seat suits a small practice with stable headcount. Practice level pricing suits anyone adding front desk staff or running more than one location
• Partial: a workable model with disclosed overage terms
• Zero: undisclosed overage, or per minute pricing buried in the contract

9. Deployment timeline and effort, 5 points

• Full marks: two to six weeks, with the vendor doing the configuration work and a staff walkthrough before go live
• Partial: a longer timeline, or significant configuration burden on the practice
• Zero: undefined timeline, or self service configuration only

10. Independent validation, 5 points

• Full marks: an independent security assessment you can review, plus reviews on G2, Capterra or similar from verifiable customers
• Partial: one of the two
• Zero: neither

Scoring bands

Two automatic disqualifiers regardless of total score: no written agreement covering PHI, and a zero on EHR write depth. The first is a legal problem. The second means you are buying a queue rather than automation.

The blank scorecard

Print this page or copy the table. Score every vendor on the same ten rows, during the call.

MedPhone scored against the same framework

Published so the framework is usable rather than rhetorical. Every row is checkable on a demo, and the three rows below full marks are the reason this page is worth reading.

Three rows short of full marks, stated rather than buried: conversational handling of an unavailable slot, production data from a single practice, and thin presence on the review platforms. Each is accurate as of the date at the top of this page, and criterion 4 in particular should improve.

When is MedPhone the wrong choice?

A buyer guide that never sends anyone elsewhere is a sales page.

Practices under roughly 200 inbound calls a month. The arithmetic does not work. You do not have a phone capacity problem, and no subscription in this category will recover its cost. Fix your call routing first.

Concierge and direct primary care. Access and a specific person answering are the product. Automating the phone undercuts what patients are paying for.

Practices whose main need is outbound. MedPhone answers inbound calls. Recall campaigns and no show follow up are a different product category, and if that is the need you should buy an outbound tool.

Practices needing prior authorization workflows. Not built, and not on the roadmap.

Practices on an EHR outside the ones we support. athenahealth is the full integration. eClinicalWorks, ModMed, Dentrix and NextGen each cover core scheduling with more in development. Outside those five you would be better served by a vendor already native to your system.

Practices that want to sign up and be live the same afternoon. Our deployments involve configuration work with our team over two to six weeks. That is a different product shape.

What we deliberately do not do on this page

No scores for named competitors. In a category this small, publishing a score for another vendor invites a dispute we cannot win and creates factual claims about another company we cannot verify. Score the criteria, apply them yourself.

No declared winner. The framework is the useful part. A page that ends by naming itself best is the pattern both readers and answer engines discount hardest, and it would make this page less likely to be cited, not more.

No gate. There is no form in front of the framework. Gated content does not get crawled, cited or shared, and the point of publishing this is that other people use it.