AI voice agents for customer calls: what they can do
By Keith Hamilton · · 6 min read
Missed calls cost UK businesses customers, and hiring more people to answer the phone is not always practical. AI voice agents for customer calls can now answer every call, deal with routine questions and pass the rest to a person. This article explains what they can and cannot do, how to keep callers informed and safe, and how to measure whether one is working.
What is an AI voice agent?
An AI voice agent is software that answers a phone call, listens to what the caller says in their own words, and replies in natural speech. Unlike a "press 1 for sales" menu, it understands questions phrased in different ways.
It works best when connected to your order system, booking diary or customer records, so it can look things up rather than just talk.
What can AI voice agents do on customer calls?
Most of the value comes from routine, repeatable calls. Typical examples include:
- Answering every call, including at busy times, lunchtimes and out of hours
- Giving order, delivery or appointment updates by looking up the caller's record
- Answering common questions about opening hours, policies and services
- Taking simple bookings, changes and cancellations
- Sending a text message with a link or confirmation after the call
In our own work, an SME telecoms business used AI voice agents to take calls for several companies, and call answering went from under 80% to virtually 100%. You can read more in the telecoms AI voice agents case study. In a proof of concept for a European fashion retailer, an AI voice agent found orders in the courier network and told callers when to expect delivery.
What can they not do well?
An AI voice agent is not a replacement for good judgement. It is weaker on:
- Complaints where the caller is upset and wants to be heard by a person
- Complicated problems that cross several departments or need a decision
- Vulnerable callers, or calls where someone's health, safety or finances are at risk
- Anything you have not given it accurate information or system access to answer
It can also mishear a name or give an out-of-date answer, which is why the hand-over to a person matters as much as the AI itself.
How should triage and transfer to a person work?
Good triage means the AI works out early whether it can help and, if not, passes the call on quickly rather than keeping the caller going round in circles.
Make it easy to reach a person
Callers should be able to ask for a person at any point. Where nobody is available, the agent can take a message, book a callback or send a text with useful information.
Hand over the context
When a call is transferred, your colleague should see who is calling, what they asked and what the AI has already done. Nobody likes repeating themselves. A UK district council proof of concept we worked on sent callers an instant text link with answers from the council website on 3 of its 27 customer journeys, with transfer to a person available. See the council AI voice service case study.
Should callers be told they are speaking to AI?
Yes. Be open with people that they are dealing with AI, ideally in the first sentence of the call. It builds trust.
Call recordings and transcripts contain personal data, so UK GDPR applies. Decide what you record, why, how long you keep it and who can see it, and update your privacy information to match.
Can an AI voice agent take payments?
It can, but this needs care. Card payments taken by phone must be handled securely under the PCI DSS standard. As a rule, an AI agent should not hear or store full card details unless the system has been built for that purpose. Safer approaches include sending the caller a secure payment link by text, or passing them to a secure payment step that keeps card details away from the AI and from call recordings.
A UK gardening company proof of concept showed an AI agent taking phone orders and payment and reconciling the order in the distribution system. That only works when the payment step is designed properly from the start. See the garden retailer phone orders case study.
How do you measure whether it is working?
Decide your measures before you switch anything on, and record today's figures so you have a baseline. Useful measures include:
- Answer rate: the share of calls answered rather than missed or abandoned
- Resolution rate: calls fully dealt with by the AI without a person
- Transfer rate and reasons: which calls go to a person, and why
- Repeat calls: whether callers ring back about the same thing
- Customer feedback: a short survey or a sample of reviewed calls
Listen to a sample of calls every week, especially early on, to find wrong answers and gaps in what the agent knows.
A simple worked example
As a hypothetical example with round numbers, suppose one person spends 20 hours a week answering routine calls on a salary of £25,000. Add about 20% for employer on-costs and that is £30,000 a year. Spread over about 1,650 paid working hours, it comes to a little over £18 an hour. Twenty hours a week for 52 weeks is 1,040 hours, or roughly £18,900 a year. If an AI agent removed between a third and two thirds of that work, the saving would be somewhere between about £6,300 and £12,600 a year, before the cost of the system. Calls no longer missed may be worth more still.
Frequently asked questions
Will an AI voice agent replace my reception or customer service team?
Typically, no. It takes on the routine, repetitive calls so your team can spend more time on the conversations that need a person. Most businesses use it to cover busy periods and out-of-hours calls rather than to remove people.
Do callers mind speaking to an AI voice agent?
Many callers are happy to, as long as it answers quickly, solves their problem and lets them reach a person when they need one. Telling them up front that they are speaking to AI, and keeping the experience simple, helps a great deal.
What happens if the AI voice agent gets something wrong?
It will make mistakes from time to time, so plan for them. Limit it to call types where the information is reliable, make transfer to a person easy, and review a sample of calls regularly so errors are spotted and fixed quickly.
Where Forwardcycle fits
Forwardcycle helps UK businesses design, test and run AI voice agents that answer calls well and hand over to people at the right moment. Find out more about our customer experience services, take the free AI Readiness Assessment or arrange a free call.