About 20% of inbound calls answered by a voice AI agent — roughly one in five — end with the caller asking to speak to a person, on average across Sales Workshop clients. That is an average across all industries, not a norm for each one: where people call to book or check a price, such requests are usually rarer; where they discuss a large purchase, a contract, their health or a complaint, they are more common. The share on its own doesn't tell you whether the agent is good or bad. What matters is why people ask for a human and what happens to them after they ask.
- About 20% of inbound calls include a request for a person; the remaining roughly 80% of conversations happen without one.
- Most reasons have nothing to do with the agent's quality: habit, high stakes, haggling, unusual questions and existing customers calling “their” account manager.
- The industry matters more than the settings: the share depends on why people call, how much is at stake and who is calling — new or existing customers.
- The goal isn't 0%. The goal is for everyone who needs a person to get one quickly — without repeating themselves.
Where the number comes from
We measure the share of inbound calls in which the caller explicitly asked to be connected to a person: “put me through to a manager”, “can I talk to an operator”, “I want a real person”. About 20% is the average across Sales Workshop clients whose voice AI agent answers an inbound line. Individual companies sit well above or below that average — below we break down why.
A note on the data: these are internal platform statistics, not independent research, and Sales Workshop builds the product discussed here. So we treat the figure as a benchmark and back up conclusions about customer behaviour with public research from Gartner and Five9 — links at the end of the article.
Why callers ask for a human
Asking for a manager isn't always a sign that the agent failed. We see eight typical reasons: some requests can be prevented with a good answer, others should be fulfilled as fast as possible.
| Reason | What it sounds like | What to do |
|---|---|---|
| Habit and distrust of automation | “Get me a person” in the first seconds, before any question | Say right away what the agent can help with, and promise a person on the first request |
| High stakes | A large purchase, a contract, money, health | Answer the factual questions and hand the decision to a specialist |
| The answer isn't in the knowledge base | An unusual situation, a rare product, an exception to the rules | Don't make up an answer: transfer, then add it to the knowledge base |
| Haggling and custom terms | “Can you give me a discount?”, instalments, an unusual volume | Give the agent agreed rules; anything beyond them goes to a manager |
| “My account manager” | An existing customer calling the person they already work with | Transfer without questions: this caller needs a specific person |
| Complaints and emotions | Frustration, conflict, “sort out my problem” | Don't argue or hold on: capture the issue and transfer |
| Bad experiences with robots | Expecting a “press 1” menu and an endless loop | Speak naturally and don't force a rigid script |
| The agent's own hiccups | Pauses, asking again, answers that miss the question | Review recordings of these calls and fix the setup |
The first reason is the most common and the least under the company's control. In a Gartner survey of 5,728 customers (December 2023), 64% said they would prefer companies didn't use AI in customer service, and the top concern was that it would become harder to reach a person. Five9's 2026 research shows the same picture: 80% of consumers are willing to use AI-powered customer service, yet two-thirds still prefer speaking with a human.
The last reason is the only one that depends entirely on configuration. Every pause before an answer, every “sorry, could you repeat that” and every off-target reply nudges the caller to say “get me an operator”. That's why we work on response speed separately: in 2026 the Sales Workshop voice agent became twice as fast to respond, so conversations flow without awkward pauses.
Early and late requests are different stories
It helps to look not only at the share but at the moment the request comes. An early one — in the first seconds, before the caller has asked anything — is almost always about habit and distrust: people react to the very fact of talking to AI. A late one — after a few answers — more often means the conversation has reached the limits of the agent's authority: haggling, an exception to the rules, readiness to buy.
A late transfer is often a good sign. If the caller got answers, became interested and wants to discuss terms, handing them to a manager is the whole point: the agent has qualified the lead and passes it on at peak interest. That's why the Sales Workshop agent transfers the call not only when asked, but also when the caller is ready to discuss a deal.
Why the share depends so much on the industry
The 20% average is made up of very different lines. Five factors affect the share of requests the most:
- Why people call. Booking, checking a price, opening hours or an order status are tasks with a clear outcome that the agent handles end to end. Negotiating terms or sorting out an unusual situation needs a person with authority.
- Money and risk. The more expensive a mistake — a home, an annual contract, medical treatment — the more often callers want a responsible person, even when the agent answered correctly.
- New or existing customers. New customers call for information; existing ones often call “their” manager or about a problem with something they already bought.
- How standard the product is. Standard services with a price list are easy to describe in a knowledge base; custom projects and quotes are not.
- The audience's age and habits. According to Five9, the phone is the most preferred customer service channel overall, and that preference rises with age — from 23% of Gen Z to 47% of Baby Boomers and 66% of the Silent Generation. It's reasonable to expect an audience used to sorting things out by voice with a person to ask for a manager more often.
Below are qualitative benchmarks by call type. They aren't measurements for each industry, but how the factors above usually combine; the exact share for your line is best measured in the agent's first weeks.
| Call type | Share of requests for a person | Why |
|---|---|---|
| Booking a service: gyms, salons, car service, clinics | Usually below average | A clear goal and a standard outcome — a calendar booking |
| Information: prices, hours, locations, availability | Usually below average | The answer is in the knowledge base and the call is short |
| Online stores: order status, delivery, payment | Below or around average | The agent handles routine questions; returns and disputes go to a person more often |
| Online schools and courses | Around average | The agent helps choose a programme, but instalments and discounts invite haggling |
| Real estate | Usually above average | A large ticket, a specific property, the habit of working with one estate agent |
| B2B and complex services | Usually above average | Custom quotes, proposals, several people involved in the decision |
| Legal, financial and insurance services | Usually above average | High responsibility; callers expect advice from a specialist |
| Existing-customer lines and complaints | Usually above average | People call “their” manager or about a problem only a person can solve |
A word on healthcare: booking a doctor's appointment is one of the simplest scenarios for an agent, while questions about symptoms and treatment are a reason to hand the call to a person right away. An AI agent shouldn't give medical advice, and such a transfer is a success, not a failure. We cover how the agent works in specific industries on our solution pages for fitness clubs, real estate, online schools and ecommerce.
What counts as normal for your business
Comparing your share with someone else's average is close to useless: a clinic booking line and a B2B support line aren't comparable. Compare yourself with yourself — week by week and reason by reason. A practical routine:
- Measure the baseline share over the agent's first two to three weeks.
- Tag the reasons: take the calls with a request for a person and sort them into the eight reasons from the table above — call recordings are saved in the lead card.
- Separate early requests from late ones: early ones go down with better wording and faster answers, late ones with a fuller knowledge base and more authority for the agent.
- Look at what happened after the transfer: did a manager pick up, how long did the caller wait, how did the call end.
How to cut unnecessary requests without losing people who need a human
- Introduce the agent honestly. The agent says up front that it's the company's AI assistant and what it can help with: according to Five9, 71% of consumers consider it very or extremely important to know when they're interacting with an AI agent.
- Promise a person in advance. Gartner explicitly recommends telling customers they'll be connected to an agent if the AI can't help — it removes the fear of never reaching a human.
- Be useful from the first line. The sooner the caller gets a helpful answer, the less likely they are to ask for a manager.
- Grow the knowledge base from real calls. Every transfer caused by a missing answer is a ready-made knowledge base entry: add the answer, and the agent will handle the next such call itself.
- Give the agent authority within the rules. If instalment terms or promo discounts are in the knowledge base, the agent can handle part of the haggling; decisions beyond the rules stay with a manager.
- Ask no more than one clarifying question. “May I ask what it's about? I might be able to help right away” is fine once. If the caller repeats the request, transfer without persuasion.
- Decide what happens when nobody is available. At night and at weekends the agent shouldn't promise a live connection: it's better to agree on a convenient time for a call back or book a meeting, and pass the manager a lead card with the call summary.
What happens after the request
A request for a manager is the moment a company either earns trust or loses a customer. According to Five9, 83% of consumers still have to repeat themselves at least sometimes after being transferred. With Sales Workshop the agent transfers the call to a manager within the same conversation — no hanging up, no call back — and the manager sees who is calling and what has already been discussed. How to set up the transfer so calls aren't lost and the caller's number is kept is covered in a separate article on transferring calls from AI to a human.
Bottom line
About 20% of inbound calls with a request for a manager is a benchmark, not a KPI. It's lower on booking and information lines and higher for big-ticket deals, B2B, legal and financial services and existing-customer lines — and that's normal. Manage the reasons, not the number: early requests go down with an honest greeting and a fast answer, late ones with a better knowledge base and more authority for the agent, and everyone else needs a fast transfer without repetition. For the full picture of how inbound calls are handled, see our guide to AI agents for inbound calls.
Frequently asked questions
- What share of callers ask to speak to a person when AI answers the phone?
- According to Sales Workshop data, about 20% of inbound calls — roughly one in five — on average across clients. The share is usually lower for bookings and information calls, and higher in real estate, B2B, legal and financial services and on lines for existing customers.
- Is it bad when callers ask for a real person?
- No. Some requests are a normal part of the process: haggling, an unusual question, readiness to buy. What's bad is when the person never comes, the wait is long, or the caller has to explain everything again.
- Can I stop the AI agent from transferring calls to a manager?
- Technically you can leave the transfer switched off, but we don't recommend it. In Gartner's survey, customers' top concern about AI in customer service is that it will become harder to reach a person. A safer rule is one clarifying question, then an immediate transfer if the caller asks again.
- How do I reduce transfers without losing customers?
- Introduce the agent honestly and say what it can help with; give a useful answer from the first line; add answers from calls where people asked for a person to the knowledge base; give the agent agreed rules on discounts and instalments; ask no more than one clarifying question.
- Why do callers ask for a manager more often in some industries than in others?
- The share depends on why people call, how much money and risk is involved, how many callers are existing customers, how standard the product is and the habits of the audience. The agent handles bookings and information end to end, while people want to discuss large and custom decisions with a person.