Review Request Routing: Who the Checkout Survey Should Skip

Most multi-specialty groups have decided they want more online reviews and have never decided who gets asked. So review request routing defaults to whatever the front desk remembers to do, which in practice means a burst of requests the week after a bad review and nothing for the four months after that.

The two failure modes look opposite and produce the same result. One group asks almost nobody, because the ask lives in a person’s memory at the end of a shift. The other asks everybody every time, so a patient with a standing infusion appointment gets a request every three weeks, stops reading anything the practice sends, and the channel is spent for the reminders that actually matter.

Routing is the part in the middle nobody builds. It decides which finished encounters are eligible for an ask today, and it is a rules problem rather than a marketing one.

The trigger is the checked-out appointment, not the monthly export

Reputation work stays undone because it is nobody’s stated priority, and the survey data says so plainly.

In a December 9, 2025 MGMA Stat poll, practice leaders put no-shows first at 27% when asked where to focus patient access effort, followed by online scheduling at 24%, phone access at 22%, and wait times at 21%. Reputation and survey work does not appear on that list at all. It is real work with no owner, which is exactly the kind of work that gets done in bursts or not at all.

The fix is to stop treating it as a campaign. An appointment that reaches checked-out status is an event, and the event is the trigger. Reading appointment status off athenaOne through GET /appointments/booked means the ask is evaluated per encounter, within the hour, against rules that do not depend on anyone remembering.

A monthly export cannot do this. By the time the list is pulled, the visit is three weeks stale, the patient has forgotten which provider they saw, and the request reads as a form letter because it is one.

Filtering by predicted sentiment is the one rule you cannot bend

There is a version of this that practices ask for by name, and it has to be refused.

The request is usually phrased as sending the survey first and only routing the happy responders to a public review page. That is review gating. Every major review platform prohibits it, and it is the fastest way to get a practice’s reviews suppressed or its listing penalized after someone has spent a year building it.

So the rule is simple and it is not a matter of taste. Everyone eligible gets the same ask. You do not branch on how you expect the patient to answer, and you do not use a satisfaction score as a gate on the public request.

What you can do is decide who is eligible, which is a completely different axis. Eligibility runs on facts about the encounter and the contact history rather than on a prediction about sentiment. That distinction is the whole article, and it is what makes the automation both effective and defensible.

Eligibility is what the routing actually decides

Once sentiment is off the table, the useful rules are the boring ones, and there are more of them than a practice expects.

The frequency rule comes first. A patient who received a request in the last several months is out, regardless of how many visits they have had since. This one rule does most of the work in a multi-specialty group, where an established patient may touch four departments in a quarter.

Then encounter facts. A cancelled or no-showed appointment never reaches checked-out status, so it drops out on its own. A patient seen twice on the same day for linked visits should generate one ask, not two. Patients who have opted out of non-clinical outreach are excluded, and so are proxy relationships where the person at checkout is a parent or caregiver rather than the patient of record.

All of that resolves against surfaces athenaOne already holds. Departments and provider groups scope the campaign, communication preferences carry the opt-out, and custom demographic fields are where practice-specific flags tend to live. Writing the outcome back with POST /patients/{patientid}/documents keeps the record of what was sent attached to the chart, so the next evaluation can see it. A suppression list that lives in a marketing tool cannot see any of this, which is why it re-asks people it should not.

The complication: the chart says one provider, the visit had another

Here is where a rules engine that looked clean on a whiteboard starts producing wrong output.

Review requests are usually meant to name the provider the patient saw, because a request that names the wrong clinician reads as a mistake and gets ignored. The obvious field to use is the patient’s primary provider on the chart. That field is stale in most practices, sometimes badly, and it is not a reliable routing key. A patient whose chart still lists a physician who left eighteen months ago will get an ask naming that physician.

The fix is to route off the rendering provider on the checked-out appointment rather than the chart header, because that is a fact about the visit that happened. In a multi-specialty group this also handles the ordinary case of a patient who saw cardiology in March and dermatology in June, where the chart header can only ever be right about one of them.

The handoff is specific. When an encounter has more than one rendering provider, or was closed by a covering provider who is not on the public roster, the automation does not guess a name. It holds the request and routes it to a person with the encounter and both provider records attached, and a scheduler or office manager decides in a few seconds. That is a small queue in practice, and it is the difference between a system that is trusted and one that gets switched off after the first embarrassing send.

A survey and a review request are two different asks

Collapsing these two is common and it costs you the more useful of the two.

An internal survey exists to tell you where the operation is failing. It can be long, it can ask about the phone experience and the wait and the billing call afterward, and it belongs to the practice. Validated instruments exist for exactly this: the CAHPS Clinician and Group Survey is built around a defined measure set covering access, provider communication, and office staff, and it gives a practice something comparable over time rather than a mood reading.

A public review request is promotional. It is short, it goes to one place, and its entire job is to lower the friction between a patient who is satisfied and a listing that does not reflect it.

Run them on different clocks and different eligibility. The survey can go to a wider set more often because it is private and its burden is lower. The review request is the scarce resource, governed by the frequency rule, and it should be spent on encounters where the patient completed a full visit rather than a nurse call or a records pickup. Sending both to the same person in the same week is how you train patients to ignore both.

Measure the routing, not the star rating

The star rating is a lagging number with a long tail, and steering by it produces bad decisions for months.

The numbers that tell you whether the routing works are upstream. Start with eligible encounters as a share of checked-out appointments, which tells you whether your rules are so tight that almost nobody qualifies. Then requests sent as a share of eligible, which catches technical failures quietly eating the volume. Then completion rate, which is the honest read on whether the timing and the wording are right.

Suppression deserves its own line. Track how many encounters were skipped and why, broken out by rule. A frequency rule suppressing most encounters in a chronic-care department is working correctly, because those patients are seen constantly. A stale opt-out flag suppressing most of a primary care panel is a bug. In a single total the two look identical, and they only come apart when you count by reason.

This matters more than it sounds, because the general track record is not good. An MGMA Stat poll found that among practices using AI in patient visits, about 44% said it had not reduced staff workload, 39% said it had, and 17% were unsure. Automation that nobody instruments tends to land in the first group. Counting suppression by reason is the cheapest way to stay out of it.

Key Takeaways

Trigger the ask off appointment status reaching checked-out, evaluated per encounter, rather than off a monthly list export.
Never branch the public review request on predicted or measured sentiment. Review gating is prohibited by the major platforms and puts the listing at risk.
Put the frequency rule first. In a multi-specialty group one patient can touch four departments a quarter, and re-asking is what burns the channel.
Route on the rendering provider from the checked-out appointment, not the primary provider field on the chart, which is stale in most practices.
Hold and hand off any encounter with multiple rendering providers or a covering provider instead of guessing a name.
Run the internal survey and the public review request on separate clocks with separate eligibility. They have different jobs.
Report suppression counts broken out by rule. Without that, a correct frequency rule and a broken opt-out flag look identical.

Review request routing is unglamorous and it is almost entirely rules. That is why it does not get built: there is no single decision to make, just twenty small ones that have to be applied consistently to every finished encounter, forever, in a group where the encounters arrive from a dozen departments. An AI team working checked-out appointments inside athenaOne can hold those rules and apply them the same way at 9am and at 6pm, write what it sent back to the chart, and pass the genuinely ambiguous encounters to a person with the context already attached.

Sources