Analytics

Losing customers quietly: what your phone calls show before they leave

Few customers say they are leaving. How missed calls, repeat calls and tone on the phone warn you early, and what AI call analytics can and can't tell you.

Chris Popescu, Director of Operations Updated November 17, 2026 7 min read

In short

  • Customers rarely announce that they are leaving. The warning usually shows up on the phone first: a call nobody returned, a second call about the same problem, a change in tone.
  • AI call analytics reads your recorded calls and sends a short daily brief, so those warnings reach the owner while there is still time to act.
  • It is an estimate, not a verdict. It only works on recorded calls, and callers have to be told why they are being recorded.
On this page

Few customers call to say they are leaving. They call about a late delivery, a bill they don't understand, or an appointment that was moved twice. Then they stop calling. By the time the owner hears about it, the account is gone and nobody can say when it went wrong.

The warning was usually there, on the phone, days or weeks earlier. This guide covers what those warnings look like, why most businesses miss them, what AI call analytics does about it, and where its limits are. It is written for owners and managers of Canadian businesses that take customer calls. Nothing here is legal advice.

Why a kept customer is worth the effort

Most owners know roughly what a new customer costs them in advertising and sales time. Fewer know what a lost one costs. Research by Frederick Reichheld of Bain & Company, summarized in Harvard Business Review, found that raising customer retention by 5% raised profits by 25% to 95%. The range is wide because industries differ. The point holds for most businesses with repeat customers: keeping the ones you have pays better than it looks.

Four warnings that show up on the phone

  • A missed call nobody returned. The customer tried to reach you and couldn't. If no one calls back, they draw their own conclusion.
  • A second call about the same problem. One call is a question. A second call about the same thing means the first one didn't fix it.
  • A change in tone. A caller who was friendly in March and short with your staff in June is telling you something, even if every call ends politely.
  • A promise that wasn't kept. "I'll call you back this afternoon with a price." If that call never happens, the customer remembers.

None of these is dramatic on its own. That is why they get missed.

Why most businesses miss them

A call log tells you how many calls came in, how long they lasted and which ones were missed. It doesn't tell you what was said. Recordings do, but someone has to listen to them.

An office that takes 60 calls a day at four minutes each produces four hours of audio every day. Nobody has four spare hours. So the recordings sit untouched until there is a dispute, and the owner only hears about the customers who were angry enough to ask for the owner by name.

The quiet ones, who were polite on the phone and then moved their business, never come up.

What AI call analytics does with a recorded call

AI call analytics is software that reads recorded calls so a person doesn't have to listen to all of them. The steps are the same in most products.

From a call to a daily brief

  1. The call is recorded, and the caller is told at the start.
  2. The recording is turned into a written transcript.
  3. The transcript is summarized: what the call was about, what was promised, and a sentiment score for how the call went.
  4. Scores are tracked for each caller, so a customer whose calls keep getting worse stands out.
  5. The day is rolled up into a short brief.

The brief is for the owner, who wants to know in two minutes whether yesterday went well. The dashboard behind it is for the manager, who needs to open the specific calls and deal with them.

What changes, signal by signal

Warning What a call log shows What analytics adds What to do about it
Missed call nobody returned A missed call entry, if anyone looks Who is still waiting for a callback, most urgent first Call back the same day
Second call, same problem Two calls from one number What each call was about and what was promised Give that customer one person who owns the problem
Tone getting worse Nothing A sentiment score for each call and the trend for that caller Have someone senior call before the next invoice goes out
Promise not kept Nothing Action items pulled from the call Check yesterday's action items every morning

A good brief is short enough to read every day. Vantact's daily brief covers how many calls were analyzed, which calls were missed, trends, friction points, one direct quote from a caller and one recommended action.

What it can't do

Call analytics is useful, and it is easy to oversell. These are the limits worth knowing before you rely on it.

  • It estimates. A sentiment score is a judgment made from the transcript. It can miss sarcasm, and it can misread a customer who is blunt but perfectly happy.
  • Transcripts contain errors. Names, numbers and bad connections are the usual trouble spots. The recording is the record. The transcript is a fast way to find the part worth hearing.
  • It only sees recorded calls. Calls taken on a personal cell phone, or on an extension where analytics isn't turned on, are not in the picture.
  • It doesn't fix anything. Someone still has to make the callback.
  • It is not a way to catch staff out. People should know it is there and what it is used for. It works best as coaching, with the person who took the call in the room.

A sentiment score is a reason to listen to a call. It is not a verdict on the customer or on the person who answered.

What Canadian privacy rules expect

Analytics runs on recorded calls, so the recording rules come first. The Office of the Privacy Commissioner of Canada updated its guidance on recording customer telephone calls on October 8, 2026. For a business covered by PIPEDA, the main points are these.

  • Tell the caller the call is being recorded, and say why.
  • Describe the purpose clearly and cover every use. A greeting that says "quality assurance" does not cover marketing or profiling.
  • Where recording isn't essential to the service, give a caller who objects another way to deal with you.
  • Keep recordings only as long as the purpose needs.
  • Treat voiceprints as sensitive. A voiceprint identifies a person by the sound of their voice, and the guidance says express consent may be required. Analytics that works from a written transcript has no need for one, so ask any provider whether its system creates them.

If your greeting was written years ago for "quality and training purposes", read it again before turning analytics on. It should describe what you now do with the recording.

Two more things to settle with a provider, in writing: where recordings, transcripts and summaries are stored and processed, and who can open them.

Health care providers in Ontario have a second set of rules under PHIPA. Those are covered in the guide to recording calls in Ontario health care.

Where to start

You don't need every call in the building analyzed. Start with the lines your customers call.

A first month

  1. Turn on recording for the customer facing lines, with a greeting that says why calls are recorded.
  2. Write down what a bad call looks like in your business. A restaurant, a law office and a repair shop will not describe it the same way.
  3. Give one person the list of unreturned missed calls every morning, and make it their job to clear it.
  4. Read the daily brief for two weeks before changing anything, so you know what a normal day looks like.
  5. Once a week, look at the customers whose scores are dropping and have someone senior call them.

The fifth step is where customers get kept. Everything before it is only how you find out who to call.

How Vantact handles this

Vantact's AI call analytics is turned on per extension, on recorded calls, so it runs on the seats that need it. It includes the missed calls report, AI call transcripts, AI call summaries with action items and a sentiment score, customer loss prevention alerts based on a caller's sentiment across calls, coaching for each call, and the daily brief by email. Sentiment definitions can be set for each business.

More detail is on the AI call analytics page, or you can contact Vantact to see it on real calls.

Talk with the Vantact team

Questions about business phone systems, call analytics or Canadian compliance? Contact us, or call the Toronto team.