Key takeaways
- Call counts tell you how busy you are. Call reasons tell you why, and that is what you can change.
- Phones remain the front door for many patients and still consume a disproportionate share of staff time[1].
- Plan for seasonality: US flu activity most often peaks in February[2], and January brings insurance changes.
- Recording and analyzing calls needs the right consent. Some states, including California, require consent from all parties to record a confidential call[3].
Beyond counting calls
Most practices know roughly how many calls they get. Fewer know why patients call, which calls end without the patient getting what they needed, or how those patterns change by hour, day and season. That gap matters because MGMA describes the phone as the real front door for a large share of patients, even where portals and reminders exist, and notes it still consumes disproportionate staff time[1].
Conversation data, meaning transcripts or summaries of calls tagged by reason and outcome, fills the gap. It can come from an AI phone tool, a call recording system with analytics, or simply a structured tally kept by staff for two weeks. The source matters less than the questions you ask of it. This article covers the questions worth asking, what the answers usually reveal, and the consent and privacy limits that apply.
Five questions call data can answer
| Question | Data you need | What it usually drives |
|---|---|---|
| Why do patients call? | Calls tagged by reason | Which calls to automate, which web pages to fix, which reminders to change |
| When do they call? | Calls by hour and day, with answer rates | Staffing at peak times, lunch coverage, overflow routing |
| Which calls fail? | Outcome per call: resolved, transferred, abandoned, callback | Broken routes, missing answers, training needs |
| Which new patients do we lose? | New-patient calls and whether they booked | Script changes, faster callbacks, availability fixes |
| What frustrates callers? | Requests for a person, repeat calls, complaint words | Greeting changes, shorter hold, clearer information |
Build a simple call reason list
Call analytics is only as good as its reason codes. Keep the list short enough that staff or software can apply it consistently, usually ten to fifteen reasons, and review it quarterly. A typical outpatient list:
- New appointment: new patient
- New appointment: established patient
- Reschedule or cancel
- Confirm or check appointment
- Refill request
- Results question
- Referral or prior authorization status
- Insurance or eligibility question
- Billing or payment
- Records or forms
- Office information: hours, location, directions
- Clinical question or symptom
- Other or unclear
Tag outcomes alongside reasons: resolved on the call, transferred, callback promised, abandoned. The combination is where the insight lives. A high share of refill calls that end in callback promises, for example, suggests patients do not know the refill process, which a website page or a voicemail message might fix more cheaply than more staff.
Expect the mix to match what practice leaders report nationally. In MGMA polling, eligibility and prior authorization and scheduling were named the most time-consuming phone tasks, far ahead of intake and refills[1]. Your own data will show whether that holds for you, and in which hours.
Getting data without an analytics platform
You do not need expensive software to start. Many practices learn more from two weeks of disciplined tallying than from months of unused dashboards. If your phone system only reports volumes and hold times, add the reasons by hand.
- Print the reason list on a single sheet with columns for each reason and rows for each hour of the day.
- Ask every person who answers the phone to make one mark per call, with a circle if the call ended in a callback promise or transfer.
- Run it for two normal weeks, avoiding holidays.
- Enter the totals into a spreadsheet at the end of each day; it takes ten minutes.
- Combine the tally with your phone system's abandoned-call report by hour.
If you use an AI phone tool, its summaries usually include a reason and an outcome for every call it handles. Check a sample against the transcripts before trusting the tags, since misclassification is common at first, and remember that calls answered by staff are not in that data unless you tag them too. The most useful view combines both: what the AI handled, what reached people, and what was lost in between.
Whatever the source, share the results with the front desk team. Staff often know why a pattern exists the moment they see it, and they are the ones who will make most of the fixes. A five-minute look at the monthly chart in a team huddle turns data into action faster than a report sent by email.
Seasonal patterns you can plan for
US flu season usually runs through fall and winter, with activity most often peaking between December and February[2]. Across 40 seasons, the peak month was most often February, in 17 seasons, followed by December, January and March[2].
Source: Centers for Disease Control and Prevention, current[2]
Show the data
| Item | Value |
|---|---|
| December | 7 seasons |
| January | 6 seasons |
| February | 17 seasons |
| March | 6 seasons |
January adds insurance churn. ACA Marketplace open enrollment runs to January 15, and plans chosen by December 15 start on January 1[4], so early January brings a wave of eligibility questions and new-patient calls. Use last year's call data by week to set staffing, overflow coverage and reminder timing for these months, and compare actuals each year.
What sentiment can and cannot tell you
Many analytics tools score calls for sentiment. Treat the score as a pointer to calls worth reviewing, not as a measure of patient satisfaction. Sentiment models struggle with accents, sarcasm, background noise and patients who are calm but unhappy. They also flag distress that is about the patient's health, not about your service.
More reliable frustration signals are behavioral: the caller asked for a person twice, called back within a day about the same request, or used words like again, still or nobody called me back. Count those, read a sample each week, and you will find the real problems faster than a dashboard of sentiment scores.
Missed opportunities hiding in the data
- New-patient calls that did not book. New patients made up 16.8% of physician office visits in 2019[5]; losing a share of their calls is lost growth. Check why: no availability, plan not accepted, call abandoned or no callback.
- Abandoned calls at predictable times, usually opening, lunch and late afternoon.
- Repeat callers chasing the same request, which point to slow callbacks or unclear answers.
- Questions answered on the website that patients still call about, which suggest the page is hard to find or unclear.
- Calls about reminders, such as which location, which suggest the reminder text needs a detail added.
Consent and privacy limits
Recording is the foundation of most call analytics, and recording law varies by state. California, for example, prohibits intentionally recording a confidential communication without the consent of all parties, with fines of up to $2,500 per violation for a first offense[3]. Practices that receive calls from several states commonly play a recording notice at the start of every call. Ask counsel what your notice should say.
Treat recordings and transcripts as patient information. Limit who can listen, set a retention period, make sure any analytics vendor has signed a business associate agreement, and avoid exporting raw transcripts into general-purpose tools. For analysis, reason codes and outcomes are usually enough; you rarely need the full conversation outside the review sample.
A monthly call review
Pull the numbers
Calls by reason, outcome, hour and day for the month, against the same month last year where you have it.
Find the top three problems
The biggest reason-outcome combinations that end badly: abandoned, repeated or unresolved.
Listen to a sample
Ten calls for each problem. Note the moment it went wrong.
Choose one fix per problem
A new approved answer, a route change, a website update, a staffing shift or a reminder edit.
Check next month
Did the problem shrink? If not, try a different fix.
Example: a fictional family practice
Example: a fictional family practice tags a month of calls and finds results questions are its fourth largest reason, and most end in a callback promise. Listening to a sample, the manager hears the same thing: patients see a portal notification and call to ask what it means. The practice changes the portal notification text to say a clinician will message them within two business days, and adds a matching approved answer on the phone line.
Example, continued: the next month, results calls fall and repeat calls about results fall further. The data also shows new-patient calls on Monday mornings are often abandoned, so the practice routes overflow to its AI line from 8 to 10 a.m. on Mondays and captures those callers for a same-day callback. For using the same data to catch AI mistakes, see when AI gets it wrong.
Frequently asked questions
What is call analytics in a medical practice?
It is the practice of tagging calls by reason and outcome, and looking at patterns by time, so you can see why patients call, which calls fail and which new patients you lose. It can use AI transcripts, recording software or a simple staff tally.
Do we need patient consent to record calls?
It depends on state law. Some states, including California, require consent from all parties to record confidential calls. Many practices play a recording notice at the start of every call; ask counsel what yours should say.
Is sentiment analysis useful for practice calls?
As a pointer to calls worth reviewing, yes. As a satisfaction score, less so. Behavioral signals such as repeat calls and requests for a person are more reliable measures of frustration.
References
Medical Group Management Association (MGMA) (2026). Phones are still a bottleneck costing medical practices time they can't afford (opens in a new tab) โ
Centers for Disease Control and Prevention (current). Flu Season (opens in a new tab) โ
California Legislature (code text via California Public Law) (current). California Penal Code ยง 632 - Eavesdropping on or recording confidential communications (opens in a new tab) โ
HealthCare.gov (Centers for Medicare & Medicaid Services) (2026-2027). Dates and deadlines for 2027 health insurance (opens in a new tab) โ
CDC National Center for Health Statistics (2019). National Ambulatory Medical Care Survey: 2019 National Summary Tables (opens in a new tab) โ