69% of people won't typically answer a call from a number they don't recognise. So what happens after your AI receptionist takes the message?
By Paul Gandar, founder of Stitch AI · October 2026
Prefer to watch? I walk through the evidence in a video on this page. Watch it on YouTube, read the full transcript at the end, or download the evidence review (PDF) with every source.
AI telephone answering services are becoming increasingly sophisticated. They can sound remarkably human, understand enquiries, ask questions and collect customer information. For businesses struggling with missed calls, the appeal is obvious.
But there's a question that often gets overlooked. Just because technology can answer a call, does that mean the customer actually wants to speak to it? And perhaps more importantly, what happens to that customer relationship when the call ends?
For businesses where every new enquiry could be worth thousands of pounds, these questions matter considerably more than how many calls an AI assistant can answer. So we reviewed the published UK research. Here's what it says.
We found no published study that tests AI voice answering against a WhatsApp handover for the same callers, so the comparison below draws on related evidence. The full evidence review lists every source, sample and date.
Imagine you're thinking about selling your house. You're considering three local estate agents and decide to telephone the first one. Instead of speaking to somebody at the branch, you're greeted by an AI voice assistant. It's polite, sounds natural and offers to help.
But you wanted to speak to an estate agent: somebody who understands your local market, can discuss your property and give you confidence that they're the right business to represent you. So you hang up and call the next agent. The first agency might never know it lost a potential instruction, and its AI answering service could have worked perfectly. That's the danger of measuring successful call handling instead of successful customer engagement.
Research supports the importance of this distinction. In research for Zoopla, YouGov surveyed 1,000 people who had sold a home in the past five years. 49% invited three different estate agents to value their property before deciding who to appoint, and 63% preferred to communicate by phone. Two-thirds (67%) expected a reply within four hours.
These customers aren't simply requesting information. They're evaluating which business deserves their trust. For an estate agent trying to win a valuable sales instruction, replacing the opportunity for an initial human conversation with an AI interrogation could create unnecessary friction.
This is particularly interesting when comparing property sales and lettings. Consider a prospective tenant trying to secure a rental property in a competitive market. Their priority may be securing a viewing as quickly as possible. They may be willing to answer qualifying questions, use an automated service or complete a structured process to progress their application.
Now consider a homeowner choosing who should market their £750,000 property. That person has a different objective. They're choosing a professional adviser and assessing the quality of service they can expect. If their first interaction feels impersonal, they may simply move on to another agency.
Of course, not all tenants behave alike, and many landlords, buyers and sellers welcome automation. The important distinction is between customers who need access to a particular service and those actively choosing between competing providers. The more choice a customer has, and the more valuable their business, the more carefully you should consider the impression your automation creates.
This applies well beyond estate agency: mortgage advisers, private healthcare providers, travel specialists, recruitment agencies and other businesses where trust and personal service influence purchasing decisions.
AI voice technology has improved enormously, but technical capability and customer acceptance are two different things.
Deloitte's 2026 Digital Consumer Trends survey of 4,156 UK consumers found that 61% would be less inclined to engage if they knew they were conversing with a generative AI assistant. Invoca's 2026 UK Buyer Experience Report surveyed 663 UK consumers who had recently made a high-stakes purchase, across industries such as financial services, insurance and travel. It found that:
This doesn't mean consumers universally dislike AI. Invoca also found that 46% of UK respondents had spoken, or possibly spoken, to an AI voice agent in the past year, and most of them rated it about the same as a human. Many people welcome AI for simple questions, routine tasks and faster service.
But it raises an interesting question. Could customers be more comfortable exchanging messages with an AI assistant than having a spoken conversation with one? There's a plausible behavioural distinction. We're already used to communicating with businesses through messaging, automated replies and digital assistants. An automated WhatsApp conversation lets customers read a question, think about their answer and respond at their own pace.
A voice AI interaction is different. It demands immediate attention: the caller must listen, speak, wait and possibly repeat themselves if something is misunderstood. And if they called specifically because they wanted human contact, discovering they're speaking to another automated system may be frustrating.
Direct evidence that customers generally prefer WhatsApp AI over voice AI is still limited. The different interaction styles are worth testing, particularly for high-value enquiries.
Suppose your AI telephone assistant has done its job. It's collected the customer's details, established what they want and generated an email summary for your team. What happens now?
Someone must read the email and decide what action to take. Often, that means calling the customer back. But the customer may now be at work, driving, in a meeting, or simply not recognise the number calling them. YouGov found that 69% of GB adults wouldn't typically answer a call from an unknown number, and Ofcom found that two-thirds of people who screen calls simply don't want to speak to callers they don't recognise.
So your employee leaves a voicemail. Later, they try again, or they send an email, which may sit unread among dozens of others. The AI has successfully captured the enquiry, but the business is still playing telephone tag.
We've seen this first-hand. In our mystery shop of 154 estate agency enquiries, our number never answered, so every agency call was a missed call. Where agencies rang again, the median gap was almost six hours on seller enquiries, and 23 agencies called once and gave up.
More advanced AI voice solutions can book appointments, transfer calls and interact with other systems directly. Where they genuinely complete the customer's task, this problem may not arise. But when the end result is simply an email requesting human follow-up, much of the original communication problem remains: the customer and employee still need to be available at the same time.
Rather than automatically forcing callers into an AI voice conversation, Voice2WhatsApp can offer them another way to communicate. For example:
"Sorry to keep you waiting. You can continue holding, or press 1 and we'll help you through WhatsApp."
If the customer chooses WhatsApp, the business can send a WhatsApp message through an appropriately configured, consent-based process. Meta lists phone flows using interactive voice response as a supported way to collect WhatsApp opt-in, as long as the opt-in names the business and complies with local law.
That changes the customer journey. The caller no longer needs to stay on hold. They can communicate with the business through WhatsApp at their convenience, and the business now has an ongoing messaging channel through which it can engage and re-engage the customer, subject to WhatsApp's messaging rules.
Instead of collecting information and hoping the customer answers a callback, you establish a channel where the conversation can continue. This is the real value of asynchronous communication. The customer doesn't have to be available at the same moment as your employee, and the conversation remains available for both parties. It's also a channel people already use: Ofcom found that 90% of UK online adults used WhatsApp in 2025, and 74% used it every day.
An AI telephone answering assistant and an AI WhatsApp assistant can draw on the same business knowledge, qualification criteria and customer service processes. Both can be configured to:
The underlying AI instructions need to be adapted for voice or messaging, but the business objectives can be virtually identical. What differs is the experience for the customer and what happens when the AI hands over.
With a conventional AI answering service, a human may receive an email summary and need to make a return call. With AI operating through WhatsApp, the human receives the conversation itself. They can read the history, understand the enquiry and pick up exactly where the AI left off. The customer isn't forced to start again, and the employee doesn't need to make another telephone call simply to continue the discussion.
This is arguably one of the most commercially important differences. A telephone call is a real-time interaction. Once it ends, further engagement normally requires another call, voicemail or other method. A WhatsApp conversation is persistent.
Suppose an estate agent receives a valuation enquiry on Monday. The AI collects some initial information, but the seller isn't quite ready to book an appointment. On Wednesday, the agent wants to follow up. With a conventional answering service, the agent may need to try calling the prospect again. With msgboxx, the agent can re-engage through WhatsApp using the appropriate approved message. The customer can read and respond when convenient, and both parties have the previous conversation available.
There are WhatsApp Business Platform rules governing business-initiated messages. Once a customer messages you, you can reply freely for 24 hours; after that, business-initiated messages must use an approved template, and marketing messages must also meet PECR and UK GDPR. Those rules don't remove the value of establishing a business-owned messaging relationship in the first place. The commercial opportunity lies in being able to continue the relationship after the first enquiry.
If you're operating a large contact centre handling thousands of routine enquiries, automating as much of the customer interaction as possible may make commercial sense. But what if your business receives relatively few new enquiries, each with significant potential value?
Would you rather have an AI voice assistant that answers 100% of calls, or a communication journey that maximises the number of customers who stay engaged and ultimately become clients? Those aren't necessarily the same outcome. A business needs to look beyond the number of calls answered and consider:
For a premium service business, improving the customer journey may be worth considerably more than simply reducing call-handling costs. The simplest way to find out is to test both routes on your own calls: split unanswered calls between an AI voice assistant and a WhatsApp handover over the same weeks, and track those four measures for each.
Stitch's msgboxx platform helps businesses convert unanswered telephone enquiries into WhatsApp conversations. Through integration with compatible telephone systems, callers can be offered the option to move from voice to messaging when staff are unavailable.
Once the conversation moves into WhatsApp, businesses can use automation or AI to engage with the customer, collect relevant information and route the enquiry to the appropriate team. When a human is required, they can take over directly within the same conversation. There's no need for the customer to repeat everything they've already explained, no separate email for somebody to notice, and no unnecessary callback simply to restart the conversation.
Staff can access conversations through a shared WhatsApp inbox on the web, mobile devices or the msgboxx Chrome extension, and supported CRM integrations keep customer communications connected to existing business records.
AI voice assistants will continue to improve. For certain tasks and customer environments, they'll be highly effective. But businesses should consider more than whether the technology can answer the telephone. They need to consider how their customers feel about that experience, how easily enquiries can be handed over and whether the business can keep in touch afterwards.
For high-value businesses where personal relationships matter, giving customers a choice may be more valuable than forcing every caller through automation. Voice2WhatsApp offers an alternative: use the telephone as the starting point, WhatsApp as the continuing channel, AI for qualification and your people for the conversations that matter most. The goal should be to give every valuable enquiry the best possible chance of becoming a customer.
See Voice2WhatsApp in action, or read our estate agency mystery shop.
This article is based on a desk review of published UK research, checked against each publisher's own page on 23 October 2026. It is not new consumer research. Survey answers describe stated attitudes, which don't always match behaviour, and the Invoca survey covers high-stakes purchases across seven industries rather than property. Invoca and InStep sell call and AI phone technology. Stitch AI sells WhatsApp messaging software, including Voice2WhatsApp, so we have a commercial interest in this question. Download the full evidence review (PDF, 7 pages), free with no email needed.
Many are wary. Deloitte found 61% of UK consumers would be less inclined to engage if they knew they were talking to a generative AI assistant, and Invoca found 58% of UK high-stakes buyers prefer a human when both are equally available. People accept AI for routine tasks, and 46% of Invoca's UK respondents had spoken to an AI voice agent in the past year.
Often a person has to call the customer back. YouGov found 69% of GB adults wouldn't typically answer a call from an unknown number, so callbacks can turn into voicemail and telephone tag. Voice systems that book appointments or transfer calls during the call avoid this.
Voice2WhatsApp lets a caller who can't get through choose to continue the conversation on WhatsApp, for example by pressing 1 on the phone menu. The business can then use AI to qualify the enquiry, and a person can take over in the same WhatsApp conversation.
Yes, with consent. Meta lists phone flows using interactive voice response (IVR) as a valid way to collect WhatsApp opt-in, as long as the opt-in names the business and meets local law. After the customer messages, a business can reply freely for 24 hours, then needs approved templates.
It depends on the enquiry. AI voice suits high-volume, routine calls where the AI can finish the task during the call. A WhatsApp handover suits low-volume, high-value enquiries where customers compare providers and the conversation continues over days. No published study yet compares the two directly, so test both on your own calls.
Ofcom's Online Nation 2025 report found 90% of UK online adults used WhatsApp in 2025, and 74% used it every day in May 2025.
Yes. Please credit "Source: Stitch AI evidence review, Oct 2026" and link to this page. The full PDF is free to download.
This is the full transcript of the video Your AI receptionist answered the call. Then what?, narrated with a voice clone of Paul Gandar, founder of Stitch AI.
Hello, I'm Paul Gandar, founder of Stitch AI. AI telephone answering has come a long way. These systems can sound remarkably human, understand what you're asking, and collect your details. If your business misses calls, the appeal is obvious. But there's a question that often gets overlooked. Just because technology can answer a call, does the customer actually want to speak to it? And what happens to that customer relationship when the call ends? For businesses where a single enquiry might be worth thousands of pounds, those questions matter far more than how many calls an AI can answer. So we went through the published UK research on this, checked every figure against its original source, and pulled it together into an evidence review. Over the next few minutes I'll take you through what it says, where the evidence is strong, where it's thin, and what I'd do about it. I'll be upfront: we sell a product in this space, called Voice2WhatsApp, so I'll be clear about what the research shows and what it doesn't.
Let's start with a scenario. Imagine you're thinking about selling your house. You're considering three local estate agents, and you ring the first one. Instead of a person at the branch, you get an AI voice assistant. It's polite, it sounds natural, and it offers to help. But you wanted to speak to an estate agent. Someone who knows the local market, can talk about your property, and gives you confidence they're the right people to sell it. So you hang up and call the next agent on your list. The first agency may never know it lost a potential instruction. And its AI answering service worked perfectly. That's the danger of measuring calls handled instead of customers engaged. The research backs up this picture. In research for Zoopla, YouGov surveyed a thousand people who'd sold a home in the last five years. Forty-nine per cent invited three agents to value before choosing one. Sixty-three per cent said they preferred to communicate by phone. And two thirds expected a reply within four hours. These people aren't just asking for information. They're deciding which business deserves their trust.
Not every customer has the same reason to persevere. Compare lettings and sales. A prospective tenant chasing a rental in a competitive market mainly wants a viewing, fast. They may be quite happy to answer qualifying questions, use an automated service, or work through a structured process to get there. Now picture a homeowner deciding who should market their seven hundred and fifty thousand pound property. Their objective is different. They're choosing a professional adviser, and judging the service they can expect. If the first contact feels impersonal, they can simply move on to another agency. Of course, not every tenant behaves the same way, and plenty of landlords, buyers and sellers are happy with automation. The useful distinction is between customers who need access to a particular service, and customers who are actively choosing between competitors. The more choice a customer has, and the more their business is worth, the more carefully you should think about the impression your automation makes. And this goes well beyond estate agency. Think about mortgage advisers, private healthcare, travel specialists, recruitment agencies, anywhere that trust and personal service decide who wins the work.
So do people actually want to talk to AI? Voice technology has improved enormously, but what it can do and what customers will accept are two different things. Deloitte's 2026 Digital Consumer Trends survey asked over four thousand UK consumers about generative AI. Sixty-one per cent said they'd be less inclined to engage if they knew they were talking to a generative AI assistant. Invoca's 2026 UK Buyer Experience Report surveyed six hundred and sixty-three people who'd recently made a high-stakes purchase, in sectors like financial services, insurance and travel. Eighty-three per cent said human connection was important during that kind of purchase. Fifty-eight per cent preferred a human when a human and AI were equally available. And eighty-two per cent said it mattered that a brand's AI clearly identifies itself. I want to be fair here. This doesn't mean people dislike AI. The same Invoca survey found that nearly half of UK respondents had spoken, or possibly spoken, to an AI voice agent in the past year, and most rated it about the same as a human. People are fine with AI for simple questions and routine tasks. The risk is narrower than that. It's the high-value customer who wanted a person, and who's free to ring someone else.
That raises an interesting question. Could customers be more comfortable exchanging messages with an AI assistant than having a spoken conversation with one? There's a plausible reason they might be. We're all used to dealing with businesses through messaging, automated replies and digital assistants. In a WhatsApp conversation, you read the question, think about your answer, and reply when it suits you. A voice AI call is different. It needs your full attention, right now. You listen, you speak, you wait, and sometimes you repeat yourself when it misunderstands. And if you rang precisely because you wanted a human, finding yet another automated system can be frustrating. I should be honest about the evidence. We didn't find any published study that directly compares a WhatsApp AI with a voice AI for the same customers. So this is a question worth testing, especially for high-value enquiries, rather than a settled fact.
Now to what I think is the hidden problem with AI telephone answering. Suppose your AI assistant has done its job perfectly. It's taken the customer's details, worked out what they want, and sent your team an email summary. What happens next? Someone has to read that email and decide what to do. Very often, that means ringing the customer back. And that's where it runs into a habit almost all of us share. YouGov asked over six thousand adults whether they'd typically answer a call from an unknown number. Sixty-nine per cent said they wouldn't. Ofcom found that two thirds of people who screen their calls simply don't want to speak to callers they don't recognise. So the customer is at work, or driving, or in a meeting, or just doesn't recognise your number. Your colleague leaves a voicemail. Tries again later. Maybe sends an email that sits unread among dozens of others. The AI captured the enquiry, but the business is still playing telephone tag.
We've seen this first-hand. Earlier this month we mystery-shopped a hundred and fifty-four estate agency enquiries in London, Birmingham and Newcastle. Our tracking number never answered, so every call an agency made to us was a missed call. Where agencies rang a second time, the median gap was almost six hours on seller enquiries, and over eighteen hours on buyer enquiries. Twenty-three agencies made one call and gave up. And not one of the hundred and fifty-four enquiries got a WhatsApp message. To be fair to AI voice systems, the more advanced ones can book appointments, transfer calls, and work directly with other systems during the call. Where they genuinely finish the job for the customer, there's no callback to chase, and this problem doesn't arise. But where the end result is an email asking someone to ring back, the original problem is still there. The customer and your colleague both have to be free at the same moment.
Voice2WhatsApp takes a different approach. Rather than putting every caller straight into an AI voice conversation, it offers them another way to talk to you. For example: sorry to keep you waiting. You can continue holding, or press one and we'll help you through WhatsApp. If the customer chooses WhatsApp, the business sends them a WhatsApp message through a properly configured, consent-based process. And that's allowed. Meta's own developer documentation lists phone menus, what's called interactive voice response, as a supported way to collect WhatsApp opt-in, as long as the opt-in names the business and meets local law. That changes the customer's journey. They don't have to stay on hold. They can carry on the conversation on WhatsApp when it suits them. And the business now has an ongoing channel it can use to engage and re-engage that customer, within WhatsApp's rules. Instead of taking details and hoping the customer picks up a callback, you've opened a conversation that can carry on. That's the real value of asynchronous communication. Nobody has to be free at the same moment, and the whole conversation stays there for both sides.
It's also a channel people already live in. Ofcom's Online Nation report found that ninety per cent of UK online adults used WhatsApp in 2025, and seventy-four per cent used it every single day, up from sixty-four per cent the year before. So when you offer a caller WhatsApp, you're not asking them to download anything or learn something new. And here's something businesses often overlook. An AI phone assistant and an AI WhatsApp assistant can draw on exactly the same business knowledge, qualification questions and processes. Both can understand why the customer got in touch, answer common questions, collect the right information, qualify the enquiry, work out the right person or team, and summarise everything for a human. The instructions need adapting for voice or for messaging, but the business goals can be almost identical. What differs is the customer's experience, and what happens at the handover. With a conventional AI answering service, a person gets an email summary and has to ring back. With AI working through WhatsApp, the person gets the conversation itself. They read the history, understand the enquiry, and pick up exactly where the AI left off. The customer doesn't start again, and nobody needs another phone call just to carry on.
This is arguably the most commercially important difference. A phone call is real time. Once it ends, getting back in touch means another call, a voicemail, or some other method. A WhatsApp conversation persists. Say an estate agent gets a valuation enquiry on Monday. The AI gathers some initial information, but the seller isn't quite ready to book. On Wednesday, the agent wants to follow up. With a conventional answering service, that probably means trying to ring them again. With msgboxx, the agent can re-engage on WhatsApp with the appropriate approved message. The seller reads it and replies when it suits them, and both sides can see the earlier conversation. There are rules, and they're worth knowing. Once a customer messages you, you can reply freely for twenty-four hours. After that, messages the business starts have to use an approved template. And marketing messages also have to meet PECR and UK GDPR, which are stricter than Meta's own rules. But none of that takes away the value of owning a messaging relationship with the customer in the first place. The opportunity isn't just capturing the first enquiry. It's being able to continue the relationship afterwards.
So what should you actually measure? If you run a big contact centre handling thousands of routine enquiries, automating as much as possible may well make commercial sense. But if your business gets relatively few new enquiries, each worth a lot, ask yourself this. Would you rather have an AI that answers a hundred per cent of calls, or a customer journey that keeps the most customers engaged until they become clients? Those aren't necessarily the same thing. Look beyond the number of calls answered. How many callers abandon before giving you their details? How many enquiries turn into a real two-way conversation? How many customers can you reach again after that first contact? And how many qualified enquiries end up as appointments or sales? The simplest way to find out is to test both routes on your own calls. Over the same weeks, split unanswered calls between an AI voice assistant and a WhatsApp handover. Keep everything else the same, and track those four numbers for each. Report them by enquiry type, because a tenant chasing a viewing and a homeowner choosing an agent may behave very differently.
Let me sum up. AI voice assistants will keep getting better, and for some tasks and some customers they'll be very effective. But it's worth thinking about more than whether the technology can answer the phone. How do your customers feel about that experience? How easily can the enquiry be handed to a person? And can you keep in touch afterwards? For high-value businesses where personal relationships matter, giving customers a choice may be worth more than pushing every caller through automation. Voice2WhatsApp is one way to do that: the phone as the starting point, WhatsApp as the channel that carries on, AI for qualifying, and your people for the conversations that matter most. A few honest caveats. This is a review of published research, not a new experiment. Surveys measure what people say, which isn't always what they do. Some of the studies cover all consumers or several industries, not just property. And as I said at the start, we sell Voice2WhatsApp, so we have an interest. That's why every source is listed in the full evidence review, which you can download free from the Stitch AI blog, with no email address needed. If you'd like to see Voice2WhatsApp working, you can book a demo on our website. Thanks for watching.
