Key takeaways
- The most damaging failure is a script that lets the caller decide what is urgent; in one survey 93% of services worked that way[1].
- Measure the service with recordings, not impressions: a weekly sample of messages scored against the call, plus timestamps for every handoff.
- Stale on-call calendars, late batch delivery and garbled callback numbers are usually fixable inside the current contract.
- Advice creep, repeat errors after a written warning and no access to recordings are reasons to start a switch.
- Some complaints are the practice's own: forwarding left off, greetings out of date, and morning messages nobody owns.
Complaints arrive one at a time
Most problems with an answering service reach the practice manager secondhand. A patient mentions at check-in that she called twice on Saturday and nobody rang back. A physician says the service woke him at 2 a.m. for a refill question, then failed to reach him about a post-op fever the following weekend. The morning report shows a callback number with nine digits. Each one looks like a one-off, nobody adds them up, and the contract renews.
The weak points are well documented. In a survey of 91 primary care offices' after-hours lines, about two in three calls that got past the recorded greeting were picked up by an answering service[1]. In 93% of those services, the patient decided whether the on-call physician should be contacted: only calls the caller described as an emergency were passed on[1]. And 95% of services faxed reports on every call to the office the next business day[1]. The survey dates from 2001 and delivery methods have moved on, but the three questions it raises still decide whether a service is any good: who judges urgency, how fast the message travels, and whether anyone checks the work.
The cost of the first question showed up in one office's own calls. Over a year, 10% of its after-hours clinical calls, 288 in all, were held for the morning because the caller did not call them emergencies. When four family physicians reviewed those calls without knowing the outcome, they would have wanted to speak to the patient that night in a mean of 50% of them, with individual reviewers ranging from 22% to 77%[1]. That is the problem worth fixing first, ahead of every courtesy and typo issue.
Eight failure points, and how to spot each one
Sort complaints by where the call broke rather than by who complained. Almost everything lands in one of these eight rows, and the right-hand column tells you whether to fix it with the vendor or plan an exit.
| Failure point | What it looks like | How to check | Usually |
|---|---|---|---|
| Caller decides urgency | The only screening question is βIs this an emergency?β | Listen to every call held for morning in one week | Fixable if the script can change |
| Wrong or stale on-call | The page goes to the physician who swapped weekends | Compare the service's calendar with yours every Friday | Fixable |
| Late or batched delivery | Messages arrive in one 7 a.m. batch | Call time versus time received, per message | Fixable |
| Garbled details | Missing digits, misspelled names, wrong date of birth | Score 20 messages a week against recordings | Fixable if errors fall |
| Long rings and holds | Hang-ups on Monday mornings and storm nights | Ring-to-answer and abandonment by hour | Partly fixable |
| Stale script | Old holiday hours, departed providers, dropped plans | A test call after every change | Fixable |
| Advice creep | Operators telling callers βit can probably waitβ | Recording review | Reason to leave if repeated |
| Language gaps | Spanish-speaking callers get an English message form | A test call in another language | Depends on the contract |
Urgency comes first because it carries clinical risk. A service is not a triage nurse and should not act like one, but it can follow a written escalation list. Replace the single emergency question with a list your physicians write. For example: any call from a patient who had surgery in the past 14 days goes to the on-call physician; any call about a child under three months with a fever goes through; anything the caller describes as getting worse goes through. The guide to writing answering service instructions covers the wording.
Stale on-call data is the most common fixable problem and the most embarrassing one. Swaps happen by text between physicians, and the service hears about them last. Give one person ownership of the calendar, send it every Friday whether or not it changed, and ask the service to confirm receipt in writing. When a page goes to the wrong person, ask for the calendar they were working from before you discuss anything else.
Garbled details look trivial until a callback fails. Most of the damage is in the callback number and the date of birth, because those are the fields nobody can guess. An operator who reads the number back digit by digit almost never gets it wrong; one who types while the caller talks often does.
Where a message breaks on its way to you
- Caller
- Your team
- System
01Caller
Patient calls at 9:40 pm
Hears the office greeting, then reaches the answering service.
02Your team
Operator takes the message
Name, date of birth, callback number and reason: the first chance for a typo.
03Your team
Urgency is decided
By the caller, the script or the operator: the weakest step in most audits.
04System
On-call is reached, or not
Paged from the service's calendar, which may not match yours.
05System
Morning report lands
Email, fax, app or portal; note the time each message actually arrives.
Example. Each handoff is a place to put a timestamp and an audit question; the failure points in the table map onto these steps.
Time every handoff, not just the first one. A service that answers in 20 seconds and then takes 40 minutes to page the physician has a speed problem that its own answer-time report will never show. Ask for four timestamps per urgent call: call received, page sent, page acknowledged, physician callback. If the service cannot produce them, that tells you something about how it manages its own work.
A one-month audit any manager can run
Pull four weeks of data
Message reports, the service's call logs, its on-call calendar and any patient complaints that mention after-hours calls.
Sample the messages
Twenty a week, plus every message marked urgent and every call held for morning. Request the recordings for all of them.
Score each one against the recording
Five fields: name, date of birth, callback number, reason in the caller's words, and urgency handled per your list. One wrong field is one error.
Time the chain
For urgent calls, the four timestamps above. For routine calls, call time against the time the message reached your staff.
Make six test calls
Weeknight, weekend, holiday, one in Spanish, one with an urgent post-op story, one office question. The mystery caller audit has scoring sheets.
Put it in writing
Send the vendor a one-page summary with the error rate, the timing gaps and the calls that went wrong, and ask for a fix date for each item.
Example: a four-physician practice samples 100 messages in a month and finds 7 with a wrong field, 5 of them callback numbers. Two urgent calls took more than 30 minutes from call to page, both on a Saturday with a calendar swap. The fixes are a read-back rule, a Friday calendar confirmation and a weekend staffing question for the vendor. Nothing yet justifies leaving.
Fix it, or start planning the switch
Most problems in the table can be fixed within the contract if the vendor engages. The test is what happens after the written summary. A good service sends recordings without argument, names a supervisor for your account and shows the error rate dropping within a month. A poor one disputes the sample, blames the caller and asks for a meeting next quarter.
- Fixable: calendar errors, script updates, read-back discipline, delivery format, a weekend staffing gap the vendor admits and addresses.
- Fixable with a contract change: escalation rules, a named account supervisor, recording access on request, interpreter access.
- Reason to leave: operators giving clinical advice after a warning, repeat errors in the same field after retraining, refusal to share recordings, or urgent calls held for morning against your written list.
- Reason to leave regardless of performance: the vendor will not sign or honor a business associate agreement on the terms you need.
If you do leave, overlap the services for at least two weeks and move one call type at a time. Switching answering services without losing calls walks through the forwarding changes, and how to choose a medical answering service covers the scorecard for the replacement.
Problems that start inside the practice
Some complaints are the practice's own. Forwarding that was never switched on after a phone system change. A greeting that still lists a provider who left in the spring. A morning message batch that lands in a shared inbox nobody owns, so a 6 p.m. message is read at 2 p.m. the next day. Before blaming the service, check that the forwarding schedule matches your hours and that one named person clears the overnight messages by a set time; callback time standards suggests targets.
Callbacks are the other gap. Eight in ten adults say they generally do not answer calls from unknown numbers, and only 19% say they generally do[2]. A nurse returning a message from a blocked or unfamiliar number may never reach the patient, and the patient then reports that nobody called back. Set your outbound caller ID to the practice name and number, and agree what a voicemail may say.
Surge hours and language access
Services are staffed for average nights, and complaints cluster on the nights that are not average. Flu activity most often peaks between December and February[3], and more than half of emergency department visits arrive outside weekday business hours[4], so worried callers are not rare after 5 p.m. Hold time matters because callers give up: in an audit of Texas primary care offices, researchers posing as new patients hung up when hold reached five minutes, and 42 of 1,081 calls ended that way[5]. Ask the service for ring-to-answer time and abandonment by hour for your account, not a monthly average, and compare them with call abandonment benchmarks.
Language access is a contract question, not a courtesy. Covered health entities must offer a qualified interpreter when interpretation is required for a patient with limited English proficiency[6], and 9.0% of US residents aged five and over speak English less than very well[7]. Ask how the service connects an interpreter at 11 p.m., then test it.
Splitting the line by call type
One common fix is to stop sending every call to the same place. Many after-hours calls are office questions that never needed a nurse: hours, holiday closures, directions, the fax number, whether a plan is accepted. In the orthopedic after-hours example, about three in ten after-hours calls were office questions, yet each one became a message for a nurse to return the next morning.
An SI Employee can take that slice. It answers office and new-patient questions from the information your practice approves, gives your 911 instruction in your words when a caller describes an emergency, and transfers anything about a patient to the number you choose, such as the on-call physician or your answering service. If nobody picks up, the caller is told so plainly and an urgent message is saved. Every call leaves a recording, transcript and emailed summary. It takes no health details and does not triage, and AssistCall does not sign business associate agreements today, so clinical calls stay with clinicians and the services that cover them (what it handles). For the full comparison, see after-hours calls: answering service vs SI Employee.
Frequently asked questions
What are the most common answering service problems in healthcare?
Callers left to judge their own urgency, pages sent to the wrong on-call physician, late or batched message delivery, wrong callback numbers, long holds on busy nights, out-of-date scripts and operators drifting into advice. The first one carries the most clinical risk[1].
How many message errors are acceptable?
Set your own threshold, but count them. Score a weekly sample against recordings and treat any error in a callback number or an urgent call as serious. A falling error rate after feedback matters more than any single month.
Can we ask an answering service for call recordings?
Ask before you sign, and write access into the contract. Recordings are the only way to tell whether an error was the caller's or the operator's. A vendor that will not share them makes every dispute your word against theirs.
When is it time to switch answering services?
When a problem survives a written warning, when operators give clinical advice, when urgent calls are held for morning against your written list, or when the vendor will not honor a business associate agreement. Overlap the old and new service during the move.
Is an SI Employee a replacement for an answering service?
For office questions, often yes. For clinical calls, no: it does not triage and takes no health details, so patient calls are transferred to the on-call physician or the service you keep.
References
The Journal of Family Practice (Hildebrandt DE et al.) (2003). After-hours telephone triage affects patient safety (J Fam Pract. 2003;52(3):222-228) (opens in a new tab) β
Pew Research Center (2020). Most Americans don't answer cellphone calls from unknown numbers (opens in a new tab) β
Centers for Disease Control and Prevention (2026). Flu Season (opens in a new tab) β
CDC National Center for Health Statistics (2022). National Hospital Ambulatory Medical Care Survey: 2022 Emergency Department Summary Tables (Table 3) (opens in a new tab) β
JAMA Network Open (via PubMed Central) (2020). Association of Simulated Patient Race/Ethnicity With Scheduling of Primary Care Appointments (JAMA Network Open) (opens in a new tab) β
Legal Information Institute, Cornell Law School (Electronic Code of Federal Regulations text) (2026). 45 CFR Β§ 92.201 - Meaningful access for individuals with limited English proficiency (opens in a new tab) β
U.S. Census Bureau (2024). GCT1603: Percent of People 5 Years and Over Who Speak English Less Than 'Very Well' (2024 ACS 1-year geographic comparison table) (opens in a new tab) β