PwC’s 2025 Customer Experience Survey carries a sharp title, “The Loyalty Illusion”, and two numbers that matter for anyone automating customer contact. Customers leave over bad experiences, and most of them are not yet fully comfortable with AI when they deal with a brand. Read together, those findings make customer experience a measure to manage next to cost, not after it.

We read the survey from the position of a company that runs automated telephone contact for regulated enterprises. The aim is to state exactly what PwC found, how far it applies outside its sample, and what it changes in the way an automated process should be designed and judged.

What PwC surveyed#

PwC published the survey on 29 September 2025. It covers 5,511 consumers and 406 executives in the United States. The scope is customer experience across channels: stores, websites, apps, chat and phone. It is not a study of contact centres or of voice automation specifically.

PwC, “2025 Customer Experience Survey: The Loyalty Illusion”, 29 September 2025
FindingFigurePopulation
Stopped using a brand because of poor customer experience, online or in person29%US consumers
Only somewhat or not at all comfortable using AI tools to engage with brands58%US consumers
Sample5,511 consumers and 406 executivesUnited States

Source: PwC, 29 September 2025. Customer experience across channels; the consumer and executive samples are separate populations.

Two caveats follow. The sample is American, and European customers, especially in regulated sectors with strong consumer-protection and data rules, may answer differently. And because the survey spans all channels, the 58% describes comfort with AI in brand contact generally, not with a voice on the phone in particular. PwC’s own wording is that these consumers are only somewhat or not at all comfortable using AI tools to engage with brands, so the figure includes many who are partly comfortable. Neither figure should be carried over to a single process as if it had been measured there.

Why “illusion” is the right word#

The title points at something many organisations know but rarely measure. Loyalty that looks stable in retention data can be thinner than it appears. A customer who has not yet left is not necessarily satisfied; they may simply not have hit the experience that tips them over. When 29% of consumers say they have already stopped using a brand because of poor customer experience, the risk is not abstract.

Telephone contact is often where that tipping point sits. People call, or are called, when something is wrong: a missed payment, a claim, a cancelled appointment, a disputed bill, a renewal that went unnoticed. These are high-stakes moments with little goodwill in reserve. An automated conversation that handles them badly does more damage than a weak marketing e-mail, because it arrives exactly when the relationship is under strain.

Regulated sectors add a second layer. In banking and insurance, a customer who leaves is not only lost revenue. A debtor who stops engaging drifts into a later stage of delinquency, where the bank may have to build reserves; a policyholder who hears from the insurer only at renewal, often by e-mail alone, has little reason to stay. In healthcare, a patient irritated by a confirmation call may simply not come, and an empty slot was someone else’s paid visit. Experience in these settings has a balance-sheet consequence, and the contact process is one of the few places where the organisation speaks to the customer directly.

That is also why the executive sample in PwC’s survey is worth noting, even though we do not quote its figures here. The survey asked both sides of the relationship, and the title suggests the two do not see loyalty the same way. Organisations that judge their customer contact by internal dashboards alone risk the same blind spot.

The 58% is a design brief, not a stop sign#

It would be easy to read the 58% as a reason to keep AI away from customers. We read it differently. Most consumers are not fully comfortable with AI in brand contact today; that is the starting condition. The question is what makes an automated conversation acceptable to someone who began sceptical.

In our experience four things matter most, and none of them is the sophistication of the model.

  1. Honesty from the first sentence. Vee always introduces itself as AI. Since 2 August 2026 that transparency is also required in the EU under Article 50 of the AI Act, with Commission guidelines published on 20 July 2026.
  2. A short, purposeful call. Customers tolerate automation that respects their time. In banking, cases ending with a promise to pay averaged 116.5 seconds of handling time (historical); short is not rude when it is clear.
  3. Memory of what was agreed. A customer who promised to pay last month should not be greeted as a stranger. Behaviour history adjusts the conversation, and a broken promise is named calmly.
  4. A route to a person. Requests for a human, disputes, hardship and other exceptions lead to an agreed transfer to the client’s own staff, not to a loop.

Each of these is something a customer notices within the first minute. Comfort with AI is built, or lost, there.

What it means for how you measure#

The practical consequence of PwC’s survey is that cost per contact cannot be the only scorecard for automation. If bad experiences drive defection, then a process that is cheap per call but irritates customers is not cheap. The measurement has to capture both the business effect and the experience that produced it.

We use a four-rung ladder for every process. The first rung is reach: without reaching the customer, there is nothing else. The second is continuation, the inverse of hang-ups after the company name is heard, which is the customer’s first verdict on the experience. The third is obtaining the key information: a promise to pay, an expression of interest, a confirmation, a completed survey. The fourth is the real business effect, such as a payment, a contract or attendance at a visit.

43.5%First-attempt completion rate (FACR)Surveys completed at the first permitted attempt, divided by eligible attempted cases.Scope: CSAT survey, energy sector, historical.

Experience is visible on every rung. A high hang-up rate after the greeting tells you the opening is wrong. A low completion rate on surveys tells you the questions are too long or badly timed. In one CSAT process in the energy sector, 43.5% of eligible cases completed the survey at the first permitted attempt (historical); that figure reflects both the reachability strategy and a conversation customers were willing to finish.

Where we would add to PwC’s reading#

Surveys measure attitudes; operations measure behaviour. A consumer may say they are uncomfortable with AI and still complete a short, honest, well-timed call that solves their problem. Equally, a consumer who says they are comfortable may hang up on a poorly designed one. That is why we listen to real calls every day in every project, rather than relying on what customers say they would do.

We would also stress the comparator. Claims that automation improves or harms experience mean little unless they are measured against the alternative on the same customers: the same portfolios, the same buckets, the same time windows. In early-stage collections, our historical results show a 7–13% increase in on-time payment frequency against a human-agent comparator. The point of quoting it is the method, not the number: an effect is only real when it is measured against something.

We play on the client’s side of the table. – Dawid Wójcicki, Founder & CEO of Vee

What to ask before you automate a customer-facing process#

  • How does the conversation disclose that it is AI, and in which words?
  • What is the hang-up rate in the first seconds, and how is it tracked from day one?
  • Which exceptions lead to a person, and how quickly?
  • Does the process remember previous agreements with the same customer?
  • Against which comparator will the effect on payment, retention or attendance be measured?
  • Who listens to real calls, and how fast does a change reach every live conversation?

PwC’s survey reminds us that loyalty is more fragile than it looks and that comfort with AI must be earned. For automated telephone contact that is not an argument against automation. It is an argument for running it as a customer-experience discipline with its own measures, which is how we run it for regulated enterprises.

Follow The First Second

Occasional issues on the mechanics and economics of AI-powered customer contact. No noise, no vendor news.

Follow via RSS