A telephone interviewer who has just heard thirty people say they have no time does not start the thirty-first call the way they started the first. The voice is flatter, the introduction faster, the pause before the first question shorter, and the customer hears all of it.

That is the quiet failure of telephone surveys, and it is one of the clearest places where an AI-executed call has a structural advantage. Vee starts every call the same way. In satisfaction surveys, that consistency shows up as more fully completed questionnaires and as data that can be compared from one month to the next.

Why surveys are harder than they look#

Feedback calls look like the easiest outbound work there is: no money changes hands, nobody is being sold anything, and the questions are short. In practice they combine the difficulties of every other outbound process with a few of their own.

  • The customer owes you nothing. In collections the customer has a reason to settle; in sales, a possible benefit. In a survey the only reason to stay is goodwill.
  • Refusals are frequent and polite. “Sorry, I don’t have time” is the most common answer, and each one costs the interviewer a little energy.
  • A partial answer is often worth little. A questionnaire abandoned after two of six questions rarely tells you what you need, especially when the open question comes last.
  • Timing is part of the method. A satisfaction survey after a hotline call or a visit to a point of sale has to happen while the experience is still fresh.

In several of the markets where we run surveys, the culture of systematic customer research is still developing, and satisfaction and NPS programmes often run at a narrow scale. That makes each completed questionnaire more valuable, and each one lost to a tired introduction more costly.

The survey on the KPI ladder#

We measure every process on the same four rungs, and surveys are no exception. Looking at them this way shows where fatigue does its damage, and where it does not.

  1. Reach. Getting through to the respondent at all, within a sensible window after the experience being rated.
  2. Continuity. The respondent staying on the line past the introduction instead of hanging up on hearing who is calling.
  3. Key information. A fully completed questionnaire, not two answers out of six.
  4. Business effect. Ratings that are as objective as possible, so the client can act on them with confidence.

Fatigue hits the second and third rungs hardest. A flat introduction loses people in the first seconds; a rushed reading of the questions loses them halfway. It also hits reach indirectly. An interviewer working through a list will rarely keep trying someone who did not answer the first time, while an automated process keeps retrying on a planned schedule, within a sensible culture of contact that does not wear people down. Every respondent who is reached on a later attempt is one who would otherwise have been missing from the sample.

What fatigue does to the data#

The obvious cost of interviewer fatigue is fewer completed surveys. The less obvious cost is that the surveys you do complete are no longer collected in the same way.

A survey is a measuring instrument. If the introduction, the reading of the scale and the tone of the follow-up question vary with the time of day and the mood of the interviewer, the instrument varies too. A customer who is asked “How would you rate…?” warmly and slowly may answer differently from one asked the same question in a hurry. You cannot tell afterwards which answers were shaped by the delivery and which by the experience you wanted to measure.

For the client, objectivity is the whole point. A satisfaction survey is commissioned to find out what customers really think of a product or of the service they received, and in healthcare, how patients rate their care can affect a provider’s relationship with its payer. A survey whose delivery drifts is measuring the interviewer as much as the customer.

What the 31st call sounds like at Vee#

Every conversation starts with the same energy, because there is no previous refusal to carry over. The disclosure that the caller is an AI, the introduction, the reason for the call and the first question are delivered identically to the first respondent of the morning and the last of the evening.

We design surveys around the methods clients use: CSAT on a numeric scale, NPS, customer effort, open questions, and the categorisation of free-text answers into themes the client can act on. Satisfaction and NPS are different instruments, one rating an experience and the other the likelihood of recommending, and they often run together in one call, which makes the order and transitions between them a design question in their own right.

43.5%First-attempt completion rate (FACR)Share of surveys fully completed at the first contact attempt.CSAT survey, energy sector, historical.

In a satisfaction survey we run for the energy sector, typically after a contact with the hotline or a point of sale, 43.5% of surveys were fully completed at the first attempt (historical result). In our satisfaction-survey work we have documented a significantly higher share of fully completed telephone questionnaires than when the same surveys were run by call-centre agents. We do not publish a comparator figure for that, and we would rather say so than round it into a headline.

What the client receives matters as much as the completion rate. Each case comes back with its technical status (reached or not, and why), its dialogue status (how far the survey got) and the answers themselves, with open responses categorised into themes. A questionnaire that stopped at the fourth question is reported as exactly that, not folded into the completed total. Counting minutes or calls is not analytics; knowing which customers rated what, after which experience, and which ones we never reached, is the start of it.

Design choices that keep people answering#

Consistency gets the conversation started. Keeping it going to the last question is a matter of design, and the same decisions come up in almost every survey project.

  1. Choose the moment. Call soon after the experience being rated, in a time window when the customer segment is likely to talk.
  2. Say how long it takes, and keep to it. A short survey that announces itself as short is easier to accept than an open-ended request.
  3. Read the scale the same way every time. Which end is high, and whether the words or only the numbers are read, both change answers.
  4. Put the open question where it will be answered. Its position changes how often, and how fully, it is answered.
  5. Choose the voice for the audience. The voice and pace that suit one survey audience may not suit another.
  6. Let complaints out. When a satisfaction call turns into a complaint, the customer needs a route to someone who can act, not another rating question.

Because Vee employs no call agents, a change to any of these applies to every live conversation at once. If listening shows that respondents hesitate at the third question, the wording can be adjusted, approved by the client and running across the whole campaign the same day. Every day, in every project, our quality team listens to selected calls, and in surveys the abandoned ones teach us most.

Where consistency is not enough#

An AI that sounds the same on every call does not fix a badly built survey. If the questionnaire is too long, the moment is wrong or the base contains people who never had the experience being rated, a consistent delivery will simply produce consistent refusals.

There is also a measurement question that clients increasingly ask: does an AI interviewer change the answers? We think the honest position is that every interviewing method has an effect, and the useful property is that the effect stays constant. A human-run survey and an AI-run survey of the same customers are two instruments, and switching between them mid-programme makes trends hard to read. If you change method, run both in parallel for a period on comparable samples, and treat the break in the series as a break.

Finally, surveys are increasingly used to rate automated service itself. That is still rare, but we expect it to grow, and it is a good discipline for any provider to be measured by the same instrument it runs for others.

Paying for answers, not for minutes#

Surveys show particularly clearly why per-minute billing is the wrong incentive. A long call is not a better survey; often it is a confused one. The unit that matters to the client is a fully completed questionnaire, collected in the right window, from the right person.

Our base model is a fee per case or record, independent of outcome, with an optional performance layer where the result can be measured in a way both sides accept. In surveys that result is unusually easy to define, so it is one of the processes where linking part of the fee to completed questionnaires is a realistic conversation rather than a theoretical one.

What to ask before you automate feedback calls#

  • How do you define a completed survey, and do you report completion at the first attempt separately from completion after retries?
  • Can you show that every respondent heard the same introduction and the same reading of the scale?
  • How are open answers categorised, and can we see the raw text behind each category?
  • What happens when a respondent wants to complain rather than rate?
  • If we move from a human-run programme, how will you help us handle the break in the data series?

Good feedback programmes are built on boring virtues: the same question, asked the same way, at the right time, to enough people. A tired interviewer cannot be blamed for losing those virtues by the thirty-first call. An instrument that never tires can be held to them on the three-thousandth.

Follow The First Second

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

Follow via RSS