Blog

What happens after the AI transfers the call?

By Anand Gohel · October 7, 2026

An AI receptionist is judged on how well it answers. The caller judges the practice on something else: whether they ended up talking to the person they needed.

Every demo of an AI receptionist follows the same arc. The phone rings, the AI answers at once, it understands the request, and it says "connecting you now". The demo ends there, and everyone is impressed.

On a real phone line that sentence is the midpoint of the call. What comes next is a staff extension ringing on someone's desk, laptop or mobile. That person may be with a patient, on another call, at lunch or off today. If they do not pick up, the caller lands in a voicemail box, or the line rings out, or the call drops.

From the caller's side, the practice failed to help them. From the practice's side, nothing happened at all. Most phone systems record that a call was transferred. Very few make it easy to see that the transfer went unanswered, and almost none tie that back to who the caller was and what they wanted.

Why a faster front door makes this more visible

A human front desk hides the problem. When a receptionist can take one call at a time, the bottleneck is the front desk and the queue forms there. Put an AI in front that answers every call immediately and routes it quickly, and the queue moves downstream to the people being transferred to. The practice has not got worse at answering. It can now see a problem it always had.

This is the most useful thing we have learned from running an AI receptionist in production: the answer rate is the easy number. The number that matters is the share of transferred calls that reached a person.

What to measure

If you run any kind of automated front door, human or AI, these are the figures worth having for every week:

• Transfers attempted, by destination. Which desks and which individuals are being sent calls?
• Transfers answered by a person. Voicemail does not count as answered.
• Time to answer. A caller who waits ninety seconds for a pickup has had a different experience from one who waits ten.
• What happened to the misses. Was the caller called back? By whom, and how long did it take?
• Patterns by time of day. Missed transfers tend to cluster: before a certain hour, over lunch, on one day of the week. That is a staffing question, and it is easy to fix once it is visible.

What to build into the transfer

Measurement only tells you about the failures. A few design choices prevent many of them.

Tell the caller where they are going. "Connecting you to billing now" sets an expectation. "One moment" followed by ringing does not.

Make shared queues accept the call. When a call goes to a group line, the person who picks up should hear a short introduction and press a key to accept. That way a voicemail greeting or an auto-attendant cannot swallow the caller, because a machine never presses the key.

Give the staff member context. A few seconds of "this is a daughter calling about her mother's medication refill" means the caller does not have to explain twice.

Have a plan for no answer. If nobody accepts within a set time, tell the caller honestly that the team will call back, and put the call on a list a named person works through. A missed transfer with a call-back is an inconvenience. A missed transfer that nobody knows about is a patient who thinks you ignored them.

The question to ask a vendor

When you evaluate an AI receptionist, ask to see the report of last week's transfers that were not answered. If the vendor cannot produce one, their product stops at "connecting you now", and the part of the call your patients care about is still unmanaged.