The feedback you never hear is the feedback that matters

You only hear from the customers who speak up

Every business builds its picture of its customers from the same few sources. Complaints. Survey responses. What account managers report back after a call.

All of them draw from a self-selected group: the customers engaged enough to say something.

Research by John Goodman at TARP found the average business hears from about one in twenty six of its dissatisfied customers. It is old work, and mostly consumer research, so take it as a direction rather than a measurement. The shape of it holds up in almost any business you look at.

The customers who complain are the ones who still care. They are giving you a chance to fix it. The risk sits with the ones who say nothing, because there is nothing to respond to and no moment at which anyone is prompted to act.

The survey does not close the gap

The usual answer is to send something out and ask. CustomerGauge puts the average B2B response rate for an NPS survey at 12.4%.

The problem is which twelve percent reply. People fill in forms when they hold a strong view, so you hear from your advocates and from the seriously annoyed. The customer drifting quietly toward a competitor does not respond at all.

That is not a small distortion. It means the group most likely to leave is the group least represented in the data you use to decide whether anyone is likely to leave.

Ask while they still remember

There is a second problem with the periodic survey, and this one is about timing rather than sampling.

Ask a customer in November how a delivery went in March and you get an impression. Fine, I think, no complaints. Ask them the day after and you get the detail. The pallet arrived at four when it was booked for eleven. Two cases were short. The driver would not wait while anyone checked.

The first answer tells you nothing you can act on. The second is a work order.

Recall fades fast, and it fades unevenly. What survives longest is how the customer felt, not what actually happened, which is why long-interval surveys produce sentiment and almost never produce a cause. If you want to fix something, you need to hear about it while the specifics are still in the room.

The window closes quietly

The stronger argument for asking straight away is that it buys you time.

One bad delivery does not lose the customer that week. It sits there. It gets added to the last one. At some point the customer decides, without telling anyone, that you are more trouble than the alternative. After that, anything you do is a win-back rather than a fix, and win-backs are much harder.

Goodman's research found the reverse effect as well. Customers whose problems were resolved quickly stayed loyal at rates of around 95%. A 2025 study by Netigate found that 85% of customers who had already left a supplier said they would have stayed if their issue had been dealt with.

That is the whole opportunity in one number. The customer was recoverable. Nobody knew there was anything to recover.

It gets harder as you grow

A business with forty customers can talk to all of them. Someone knows every account by name and hears about the late delivery on the day it happens.

At four thousand, that stops being possible. A coverage model gets built, and it is a sensible one: put the people where the revenue is. The largest accounts get a named contact who spots a problem early. Everyone else gets an invoice.

So the customers you hear least from are the ones you understand least. They are also the ones you are most likely to lose, and the ones whose reasons you will never learn.

And it is not only problems you are missing

The gap is wider than dissatisfaction.

Uncontacted customers cannot tell you what they wanted and could not get from you. They cannot tell you what a competitor has started offering, or that the person who used to sign off orders has left, or that they would happily buy three other things from you if anyone had mentioned them.

That is not customer service information. It is demand data, and most businesses only collect it from the fraction of the base that already has someone's attention.

Asking everyone quickly used to be impossible

There were two options and both failed at scale.

Surveys are cheap to send and almost nobody answers, because the effort sits with the customer. Calls work, and people do talk, but the arithmetic stops you. Reaching a base of 15,000 customers every fortnight is around 30,000 conversations a month, or roughly 37 full-time staff for one market. That headcount does not get approved to serve the accounts at the bottom of the ledger, and it would be hard to keep staffed if it did.

So the same customers get asked once a year, and everyone else stays a blank.

Make it a by-product, not a project

What changes with a voice agent is not really the cost of a call. It is that headcount stops being the ceiling. One agent can reach the whole base on a regular cycle, in the language each customer speaks.

That solves both problems at once. You are no longer sampling, because everyone gets asked. And you are asking days after the last delivery rather than months, so the answers still contain facts.

Once it is running, feedback stops being something you launch. If you are already on the phone taking an order, asking whether the last delivery arrived on time adds ten seconds. The customer does not opt in, click a link or remember a password. They answer a question in a conversation they were already having.

A problem in one region then shows up within a fortnight, while there is still something you can do about it.

Start with the ones who already went quiet

Pull every account whose orders have dropped or stopped in the past six months. In most businesses the list is longer than anyone expects.

None of them complained. None of them cancelled. Nobody has ever asked them what changed.

Ask them.