AI Customer Service: Why 78% of Customers Try to Bypass Your AI Agent
78% of customers have tried to get past an AI agent to reach a person. Here's why it happens, and how to build AI customer service they'll actually use.
A customer tells your chatbot they've been charged twice and gets a link to the billing FAQs. They rephrase the question in the chat and get the same link back. So they ring up, wait in the queue, and have to explain the whole problem again to your agent.
By the time it's sorted they've told the story twice and are a good deal more fed up than when they started. Experiences like that are why 78% of customers in Twilio's latest customer service research have tried to get round an AI agent to reach a person.
Customers are happy to let AI handle the routine things. What they don't appreciate is being stuck with it.
So how do you get AI handling the routine queries in a way customers are happy to use, with faster answers and a real person there the moment they need one?
Why customers go looking for a way round
Most AI agents are judged on containment, the share of contacts that never reach a person. On that measure, a customer who gives up and rings back tomorrow counts as a success. Twilio's research found that 86% of businesses think customers are happy with their AI customer service, while only 51% of customers agree. That's a 35-point gap between the experience businesses think they're delivering and what their customers actually get.
Customers also want to know what they're dealing with. 76% say an AI agent should introduce itself as AI at the start, and 63% want a quick way to reach a person at any point, yet only 32% of businesses make that kind of transparency a priority when they design their agents.
Reaching a person doesn't fix things if the handover goes badly, and that's where most customers lose patience. In TCN's survey of 1,000 UK adults, 41% named repeating themselves to several people or departments as one of their biggest frustrations with customer service, and a chatbot that passes a customer on without a record of what they've already said causes exactly that.
What changes when the AI hands over well
When customers know what they're dealing with and can reach a person whenever they need one, they have less reason to work around it, and it gets the chance to deal with the things it handles well, like checking a balance or updating someone's details.
The customers who need a person reach one with the conversation they just had with the AI already on your agent's screen, including the account number and what the AI has tried, so your agent can start on the fix straight away and the customer gets their answer faster.
It also changes what you measure. If you track which queries the AI resolves and which it hands over, and why, you can see what it handles well and give it more of those, while the queries that need a person stay with your team.
What to look for in an AI customer service agent
Do customers know what they're dealing with, and can they ask for a person whenever they want? If the only way out is to fail three times, customers will fail on purpose, and they reach your agent already annoyed.
When it hands over, does your agent see the whole conversation? If not, the customer starts again with someone who knows nothing about them.
Does it run inside your contact centre? An AI tool that sits beside it has its own routing and reporting, and context tends to get lost between the two systems.
Is it measured on resolution? Containment figures on their own can't tell a helped customer from one who gave up.
How MAIA can help
MAIA is Ciptex's AI agent for customer service. It can be added to the contact centre you already run, and it handles routine contacts on voice, chat, email, SMS and RCS. It makes sure customers know who they're dealing with at all times and can reach a person whenever they ask, so they always feel comfortable. When they do, MAIA passes them to the right queue with a short summary and everything they've shared, and your agent picks up where MAIA left off.
Because MAIA works inside your contact centre rather than beside it, it uses the same queues, routing and agent desktop as your team. Someone who starts on webchat and rings later is still one conversation on the agent's screen.
Why your AI agent needs to cover every channel
A lot of AI customer service starts and ends with a webchat widget. Customers don't stick to one channel, though. They message, then call, then email to chase, and if the AI only covers chat, everything else lands on your team and the customer explains it all again on the phone.
An AI agent that works across voice, chat, email and messaging can take the same routine query wherever it turns up and keep the history in one place. Customers get the same answer whichever way they get in touch, and when a person does need to step in, they can see everything that's been said so far.
It also gives you a truer picture of demand. When the AI sees every channel, you can tell which contact reasons drive the most volume overall, not just on your website, and choose which ones are worth handing to AI next.
Start with your busiest contact reason
Pick the contact reason that fills your queue most often and has a clear answer, such as "where's my refund?" or a change of address. That's usually the best first job for an AI agent. It won't handle everything straight away, but it gives you a sensible place to start and build from.
Book a demo and we'll show MAIA handling that journey, including the handover to one of your agents.
Not sure where you stand yet? Start with the AI Readiness Checklist, or read more on our AI readiness page.
