By Sagar Shankaran, Founder of CallSphere
Discharge planners call three suppliers until one answers. What DME loses to abandoned calls, and what an instant-answer line changes about that number.
Key takeaways
How many discharge planners called your main line last Friday after 4 p.m. and got a voicemail greeting?
You do not know. That is not an insult; almost nobody in home medical equipment knows, because the missed ones do not appear anywhere. A hospital case manager working a Friday discharge has a printed list of contracted suppliers. She calls the first one about a semi-electric hospital bed and an over-bed table for a patient going home at 6 p.m. If it rings four times, she hangs up and calls the second name on the list. Nothing on your side records that this happened. The referral simply belongs to a competitor now, along with the mattress upgrade, the follow-up walker, and whatever that household needs for the next three years.
Every DME owner knows the shape of this. The delivery technicians are out, the customer service reps are on other calls or already gone, the intake coordinator is trying to close out the day's authorizations, and the phone rings with the highest-value call of the week: a same-day bed order tied to a discharge that has to happen tonight.
The same window is crowded with calls that do not make you money but cannot be ignored. A patient whose oxygen concentrator is alarming. A daughter asking where the delivery is. A physician's office calling back with the chart note you have been chasing for six days, which if it goes to voicemail costs you another week of a rental sitting unbilled.
The workaround is the on-call rotation and the after-hours answering service. The answering service takes a name and a number and reads it back to you on Saturday morning; it cannot look up a delivery ticket, it cannot tell a discharge planner whether you deliver to that ZIP code tonight, and it certainly cannot book a Saturday delivery window. It is a message pad with a phone bill.
Here is the definition worth pinning down: a real-time voice agent is a phone answerer that hears speech and speaks back directly, in about two-tenths of a second, and can look something up in your system or book something while the conversation is still going. No hold music, no "let me transfer you," no reading a message into a pad.
Two things changed to make this ordinary rather than experimental. Speech-to-speech models now answer in roughly 200 milliseconds — quicker than an average human reaction time — because the listening, the thinking and the speaking happen in one pass instead of three separate steps bolted together. And they can call out to your systems mid-sentence, so the agent can check whether a bed is in stock, whether the patient's ZIP is inside tonight's delivery radius, and what the next open Saturday window is, all while the case manager is still talking. Google's Gemini 3.1 Flash Live arrived in March 2026 with strong performance on exactly that kind of look-up-while-talking work, and OpenAI's realtime line pushed the same capability into ordinary business phone lines.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent for healthcare in your browser — 60 seconds, no signup.
The 2024 version of this was a voice menu with better manners. It could not read your inventory, it introduced an awkward pause after every sentence, and any caller with an accent or a bad connection ended up mashing zero. That pause is what made people hate it. The pause is what went away.
flowchart TD
A["Call lands on the main line, 4:41 p.m. Friday"] --> B["Agent answers on the first ring"]
B --> C{"Who is calling?"}
C -->|Discharge planner| D["Takes bed order, checks the delivery ZIP, books a Saturday window"]
C -->|Concentrator alarming| E["Runs the alarm questions, pages the on-call respiratory therapist"]
C -->|Where is my delivery| F["Reads back the delivery ticket status"]
D --> G["Referral written into the intake queue with the callback number"]
F --> G
E --> H["Therapist calls back with the case already summarized"]
4:41 p.m. The case manager at the regional hospital dials. The line is answered before the first ring finishes. She says she has a discharge tonight, needs a semi-electric bed with rails, an over-bed table and a bedside commode, patient going home to a ZIP code twenty-two miles out.
The agent confirms it delivers to that ZIP, tells her the last technician run tonight is committed but there is a 9 a.m. Saturday window available, and asks whether the family can manage overnight. She says yes. The agent takes the patient's name and date of birth, the ordering physician, the insurance, and asks her to fax the order to the number she already has. Then it books the Saturday window against the delivery schedule, creates the referral in the intake queue, and texts the on-call branch manager a two-line summary: bed, table, commode, Saturday 9 a.m., 22 miles, fax pending.
At 4:52 a patient's wife calls: the concentrator is beeping. The agent walks her through the two questions that separate a clogged filter from a real fault, establishes the patient still has a full backup cylinder, and pages the on-call respiratory therapist with the model, the serial number, the alarm description and the answers to both questions. The therapist calls back four minutes later already knowing what she is dealing with, instead of starting from "tell me what it is doing."
At 5:10 the physician's office calls back with the chart note. The agent confirms which patient, confirms the fax number, and flags the pending authorization file so your specialist sees on Monday morning that the document is coming rather than chasing it a seventh time.
Assumptions, illustrative. Your own call reporting from your phone system will give you the first three lines in about ten minutes.
| Inbound calls per month | 3,600 |
| Abandoned or sent to voicemail | 16%, or about 590 |
| Share of those that are new referral calls | 3%, or about 18 |
| Captured by an agent that answers instantly | 60%, or about 11 |
| Of those, orders that actually convert | 55%, or about 6 new patients per month |
| First-year contribution per new patient (illustrative) | $850 |
| Monthly value of captured referrals | about $5,100 |
| Status and delivery calls handled without a rep, at 4 minutes each | about 220 calls, 15 hours per month |
Six patients a month is not a dramatic number, and that is the point — small enough to be credible, large enough to matter on a business where a bed rental and the equipment that follows it compound over years. The fifteen hours is the quieter benefit: time your customer service reps spend on calls that need a person during respiratory season, when volume is up and staffing is not.
Draw this line before you turn anything on, and write it down.
Still reading? Stop comparing — try CallSphere live.
See the healthcare AI agent handle a real call — complete, industry-specific, and live in your browser. No signup.
Anything clinical goes to a human. A patient whose ventilator is alarming, a caregiver describing shortness of breath, a wound that looks infected around a negative-pressure dressing — the agent's only job on those is to gather the facts fast and get a respiratory therapist or clinician on the phone. It should never advise, never adjust a setting, never suggest a therapy change.
Anything about coverage and cost gets a careful answer, not a confident one. "What will this cost me" is a question with a real answer only after benefits are verified, and a wrong number given at 5 p.m. Friday turns into an angry call and sometimes a complaint. The right behaviour is for the agent to explain that a benefits check has to run and to book a callback from your billing specialist.
And keep a person on complaints and on anything that sounds like a patient safety event or an equipment failure that caused harm — those have reporting obligations, and they need a human being's judgement and a documented trail from the first minute. Two more practical limits: the agent needs to hand off cleanly and quickly when a caller asks for a person, and you should listen to a sample of recorded calls every week for the first month. The failure mode is not usually a wrong answer; it is an agent that is too eager to keep handling a call it should have passed on.
Probably, within a sentence or two, and it matters less than owners expect. What loses you the referral is a fourth ring and a voicemail beep, not a machine that answers immediately, knows your delivery area and books the window. Have the agent identify itself plainly and offer a person whenever the caller wants one.
It can write in, and if it cannot, the deal is not worth much. The whole benefit is that a referral taken at 4:41 p.m. exists as a referral, not as a message. Ask any vendor to demonstrate it creating a real record in your system — during the trial, on your data, not on a demo tenant.
That is the case for it, not against it. When power goes out across a county, your oxygen patients all call at once and no reasonable staffing plan covers it. An agent that answers every one of those calls simultaneously, confirms who has backup cylinders and how many hours are left, and hands your respiratory therapist a prioritized list is worth more in that one afternoon than in a normal quarter.
Pull your abandoned-call count for the month before and the month after, and count referrals that came in through the agent and turned into delivered orders. Both numbers are in your phone system and your billing system already. Do not accept a vendor dashboard as the measurement.
Ask your phone provider for a report of calls answered, abandoned, and sent to voicemail, broken out by hour of day, for the last ninety days. Look at the 4 p.m. to 6 p.m. band and at Monday mornings. If the abandoned number in those windows is small, you do not have this problem and you should spend your money elsewhere. If it is not small, you now know exactly what you are buying.
That is the work CallSphere does: AI voice and chat agents that answer the phone and web chat every hour of the day, take referral and patient details, book delivery and fitting windows, and pass a complete record to your team — with the clinical calls routed straight to your on-call therapist rather than handled by the agent.
Written by
Sagar Shankaran· Founder, CallSphere
LinkedInSagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
North Platte, Nebraska family practices serve Union Pacific railroaders on 24-hour schedules. An AI phone agent answers, books, and triages around the clock.
Sixty calls a day, one person at the counter. See what unanswered banner and yard-sign calls cost a print shop, and what an instant-answer voice agent changes.
Luxury buyers in Scottsdale call nights and weekends, and the first team to answer usually wins. How AI voice and chat agents keep Arizona brokerages first.
Late-night lockouts, dead AC, and towed cars define after-hours calls at Key West, FL rentals. An AI night desk fixes most 2 a.m. calls without waking you.
What a device maker loses to unanswered support calls, and what an instant-answer voice line changes about complaint records, RMAs and service parts revenue.
Community bank calls that hit voicemail at noon cost real deposits. A 2026 voice agent answers in 200ms, reads holds and payoffs, and hands the rest to a banker.
© 2026 CallSphere Inc. All rights reserved.
Made within San Francisco
Watch how CallSphere handles real customer calls, schedules appointments, and processes payments — live.
Try Live DemoBook a DemoCalculate Your ROI