What to Measure on an Automated Front Office Besides Call Volume

Every vendor selling an automated front office will show you a volume chart. Calls answered, messages handled, a line going up and to the right. It is the easiest number to produce and it tells you almost nothing about whether your staff have less work than they did last quarter.

The number that replaced it is not much better. Containment, the share of interactions handled without a person, has become the headline metric in patient access, and on its own it is misleading in a specific way.

A contained call that ends with a message somebody has to read, interpret, and act on has not been contained. It has been moved. The volume chart goes down and the work sits in a different queue wearing a different name.

What a practice needs is a small set of measures that survive that trick. Each one should answer a question a manager actually has, and each one should be something you can ask a vendor to show you on live data rather than describe.

Start with what the phone is actually carrying

Before choosing measures, know what the volume is made of. In a March 10, 2026, MGMA Stat poll of practice leaders, the most time-intensive phone tasks were eligibility and prior authorization at 45%, scheduling at 31%, intake at 9%, prescription refills at 6%, and an other category at 9%.

That distribution is the argument against counting calls. Eligibility and authorization work is not one interaction, it is a chain of them, and a practice that handles the same chain in fewer touches will look worse on a volume chart than one that handles it badly in many.

For a multi-specialty group the mix also differs by department, which makes a single practice-wide number close to meaningless. The orthopedic line and the behavioral health line carry different work and should be graded on different things.

So the first measure is not a rate at all. It is a breakdown: what share of interactions belonged to each category, by department, this month compared with last.

Ask whether the work finished, not whether the call ended

The measure worth building the whole program around is completion. For each category of request, what share ended with the thing the patient wanted actually done, in the system where it counts.

A scheduling request completes when an appointment exists on the calendar with the right type, provider, and department. A refill request completes when it is sitting in the correct clinical staff queue for a provider to act on. A records request completes when the release paperwork is filed and the request is in the right work queue.

None of those definitions mention a phone call, which is the point. They are all readable from athenaOne rather than from a telephony report, and that is what makes them hard to game.

Pair completion with its opposite and you have the whole picture. Of everything that did not complete, how much was handed to a person with the work already assembled, and how much was handed over as a note saying the patient called.

The measure most practices skip is time to touch

Speed of answer is a call center metric. The one that matters to a practice is how long an open item sits before somebody works it.

A referral received Friday afternoon that nobody opens until Tuesday is a five-day gap that no phone report will show, because nobody called about it. The same is true of an order sitting in the queue, a form waiting on a signature, and an authorization approved with nothing booked against it.

Measure the age of the oldest open item in each queue, and the share of items older than whatever your practice decides is too old. Both are simple, both are readable from the systems you already run, and both move when the process improves.

This is also the measure that exposes relocated work fastest. When containment rises and the age of open items rises with it, the automation is not finishing anything. It is filling a queue quietly.

There is already a vocabulary for the patient side

Practices do not need to invent access measures from scratch. The CAHPS Clinician and Group Survey measures access as the patient experienced it, including getting timely appointments, care, and information, along with how well office staff handled the interaction.

That framing is useful because it maps onto things a front office controls. Getting a timely appointment is a slot availability and outreach problem. Getting information is a callback and status problem. Both are operational and both are invisible in a call log.

The practical version for a multi-specialty group is a short list. How long from request to booked appointment, by department. What share of pre-visit confirmations completed. How often a patient had to call back to get a status they should have been given the first time.

That last one is the single best proxy for whether the automation is any good. Repeat contacts about the same request are the cost of an incomplete first interaction, and they are countable.

Why the workload result is a coin flip today

It helps to know that disappointment with automation is common rather than personal. An August 5, 2025, MGMA Stat poll of 244 applicable responses found 71% of practice leaders reported some use of AI in patient visits, but among those using it, 44% said it had not reduced staff workload, 39% said it had, and 17% were unsure.

That spread is not mysterious. A tool that automates one step inside a process leaves the process, and the person running it, exactly where they were. A tool that owns a workflow from intake through closure removes hours.

Both get sold with the same vocabulary. The measures above are how you tell them apart before you sign, and how you tell whether the one you bought is working after.

So build the scorecard before the pilot, not after. A baseline taken from your own queues in the four weeks before anything changes is worth more than any number a vendor reports to you later.

What to ask a vendor to show you

Every question here is a request to see something on live data rather than to hear a claim.

Show me completion by request category, defined as the record existing in athenaOne rather than the call ending. Show me the age of the oldest open item in each queue you touch. Show me last week’s handoffs and let me read three of them. Show me repeat contacts about the same request.

Then ask the harder one. When your tool cannot finish something, what does the person receive? If the answer is a notification, you are buying an alerting system and staffing the rest yourself.

All of this depends on the automation being able to read and write the systems where the work actually lives. PGA works across 440+ of athenahealth’s roughly 800 endpoints, which is why completion can be defined as a record in athenaOne instead of a status in a vendor dashboard.

Grade on depth of access, not on the length of the feature list. Every measure worth having on this page requires it.

Key Takeaways

Stop reporting calls handled, and break interactions down by category and department, because a multi-specialty average hides everything useful.
Define completion as a record existing in athenaOne rather than as a call that ended, which is the one definition automation cannot game.
Measure the age of the oldest open item in every queue, since relocated work shows up there long before it shows up anywhere else.
Count repeat contacts about the same request, the cleanest proxy for whether a first interaction actually finished.
Take a baseline from your own queues in the four weeks before a pilot starts, because a vendor-reported number after the fact has nothing to compare against.
Ask to read three real handoffs, and treat a notification with no assembled context as a sign you are buying an alerting system.

Call volume is the number that is easiest to produce and hardest to act on. Measure completion in the record that matters, watch the age of what is still open, count how often patients have to call back, and make every vendor show you the answer on live data rather than a slide.

Sources