These days, if you are active online you will never be too far away from an advert or article promoting the use of AI as a business tool. Such uses include training AI to function as an AI voice agent in customer service scenarios. This article provides a balanced view of the benefits and drawbacks of AI voice agents, drawing on input from AI itself for pros and cons as well as use case scenarios.
A link to the AI research report is provided at the end of this article. I have also included some thoughts and opinions on AI as a research tool in the narrative.
It should also be noted that the AI report clearly references larger national and multinational organisations, so some of the comments may not be as applicable to the SMEs serviced by Paperclip’s human agents.
In the opening paragraph I mentioned ‘balanced view.’ For clarity’s sake I should declare that I am firmly in the ‘human’ camp when it comes to any marginal decision regarding ‘human v machine.’
.
Benefits
The key benefits proposed for AI voice agents are as follows:
- 24/7 availability
- Faster response
- Handling high call volumes
- Scalability
- Cost savings
- Efficiency
- Consistency (uniformity)
- Accuracy
- Hybrid – empowering human agents for complex tasks
- Continual learning
- Knowledge retention
- Analytics
Review
There is certainly an argument for using a hybrid human-AI agent model if you have occasional need for customer engagement outside normal office hours. In some clear-cut cases – e.g. RAC, emergency medical care, multinational IT support – there is clearly a need for 24/7 coverage. The question in such scenarios is whether AI voice agents are suitable for the task in hand.
Within the AI report it references research that states that ‘62% would rather get immediate help from a bot than be on hold’. In such a scenario, I think I would too! However, the customer service objective should be to ensure all calls are answered rather than placing callers on hold in the first place.
This raises one of the key issues for me in using AI as a research tool. It is unable, at the first level at least, to evaluate the relative merits of the different pieces of market research it references i.e. it will report the ‘headline’ findings without evaluating research methodology, sample size, margin of error etc. This can lead to a skewed perspective.
The second benefit in the AI list is ‘faster response.’ Given that professional call answering/call centre service providers like Paperclip will answer most calls within five seconds this benefit is marginal at best. Furthermore, if you select the appropriate call answering/call centre service provider they will have the capacity to handle ‘high call volumes.’
A clear benefit of AI voice agents, assuming the IT capacity and bandwidth are in place, is their ability to scale instantaneously. In the case of human call answering services, workforce scheduling changes to accommodate scaling would take at least a day to take effect. However, it should be stated that this is acceptable in most SME use case scenarios.
In the AI report, cost savings are attributed primarily to shorter call durations and reduced staffing costs with humans only dealing with calls that have been escalated by the AI voice agents. When AI voice agents are being used to deliver system-driven interactions that can be broken down into relatively straightforward process steps – e.g. fast-food orders – there are undoubtedly opportunities to reduce call durations. However, as a recent news story on the BBC website indicates – ‘Taco Bell rethinks AI drive-through after man orders 18,000 waters’ – this is still far from foolproof.
Reported benefits for AI voice agents are consistency and uniformity. This strikes me as being the same argument that is put forward for using scripts in human call answering scenarios. The problem with this argument for me is that humans are notoriously inconsistent and far from uniform; isn’t that what makes the world go round? Trying to shoehorn callers into a ‘one size fits all’ model sits uncomfortable with me.
What makes AI so clever from a systems perspective is that, if designed correctly, it will be continually retaining new knowledge which feeds into continual learning. In other words, over time the AI voice agent will become more effective. However, isn’t that exactly what humans do over time anyway?
At Paperclip we are looking into AI applications that can assist our human call agents to deliver a better service. For example, using AI to analyse bulk call recordings to highlight any service issues as quickly as possible e.g. the mispronunciation of names, products and so on. AI is certainly the clear winner, in terms of speed, when it comes to collecting, reviewing, and presenting data analytics. Personally, I still feel it needs human input to evaluate such data, particularly if there is a qualitative element to it.
On the subject of analytics, the report notes that ‘AI speech analytics can flag when customers sound frustrated.’ I would argue that human voice agents can identify such frustration instantaneously and take steps to address it within a heartbeat.
Drawbacks
The AI report identifies the following factors as drawbacks of using AI voice agents (effectively, the benefits of using human voice agents):
- Lack of empathy
- No personal touch
- Customer frustration
- Errors – hallucinations and misinformation
- Barrier to reaching a human agent ‘obstacle effects’
- Negative reactions
- PR risks – damage to brand equity
- Sector-specific limitations
- Technical limitations
- Sycophancy (AI wants to please)
- Technicalities of introducing an AI phone system
Review
Whether you perceive ‘lack of empathy’ and ‘no personal touch’ as drawbacks or problems will depend on whether you view your customers as ‘data points’ or humans. Many of Paperclip’s clients are active in sectors where emotions and stresses can run high, including law firms, estate agents, IT support and healthcare. In such sectors, the empathy of the human call handler is critical.
Not all AI voice agents are created equal. Some are far more effective than others in escalating a call to a human agent when they are unable to respond effectively. Others, however, will try repeatedly to resolve the problem leading to considerable customer frustration and negative feedback. At the extreme this can significantly damage brand equity. For example, fintech firm Klarna heavily automated its customer support (and achieved considerable efficiencies) but has now admitted that ‘nothing is as valuable as humans’ for empathetic services.
Perhaps the most serious drawback of AI voice agents is the risk of unidentified errors – e.g. hallucinations and misinformation – and the financial and service implications of such errors. For example, an Air Canada chatbot mistakenly promised a refund to a passenger, who relied on the promise. A court held that the airline was, not unreasonably, liable for the bot’s false promise.
It is rare, but there are examples of AI voice agents ‘going rogue.’ The following is an extract from a BBC News story ‘How to stop AI agents going rogue’ (Sean McManus, 26/08/25):
“An AI agent consists of a few things,” says Donnchadh Casey, CEO of Calypso AI, a US-based AI security company.
“Firstly, the agent has an intent or purpose. Why am I here? The second thing: it’s got a brain. That’s the AI model. The third thing is tools, which could be other systems or databases, and a way of communicating them.”
“If not given the right guidance, agentic AI will achieve a goal in whatever way it can. That creates a lot of risk.”
As the AI-generated report notes, ensuring AI only provides vetted, correct information is a major challenge.
Of course, human agents will occasionally make errors regardless of how good the training is, but the human agent is far more likely to recognise they have made an error and take suitable steps to correct it. An apology from a bot does sound – and feel – hollow.
My experience of using AI for background research is that it has a bias towards ‘sycophancy.’ In other words, AI wants to help and wants to solve the problem it has been set. There is a risk, therefore, that it will edit and use a tone of voice to ‘please’ the listener. Passing on negative news – e.g. your appointment has been cancelled because the physiotherapist is ill – is not easy for human agents, but they can deliver the message with genuine empathy and work with the patient to reach a solution.
Link to the ChatGPT article on the benefits and drawbacks of AI voice agents
Click on the following link to access the Chat GPT report: LINK
Conclusion
AI is undoubtedly clever and has a huge contribution to make in certain areas. For example, when researching this article I came across the following AI-related headlines on the BBC website:
- AI supercomputer to help cancer vaccine research
- AI stethoscope could detect major heart conditions in seconds
However, when it comes to customer interactions over the phone, I believe that this article has laid out the reasons why humans prefer to communicate with other humans in most cases.
In ten years, however, who knows…!
You may also like:
Hi, I’m Louise Bellwood, Director at Paperclip. I am responsible for delivering Paperclip’s customer service and growth strategies for 2024 and beyond, I am a crusader for the benefits call answering can deliver for organisations of all kinds; customer service, sales generation, customer retention and more. If you have a question about call answering – pricing, set-up, benefits – I would love to hear from you at louise@paperclip.co.uk or 01246 418 181.