Why 78% of customers try to bypass your AI agent

Why 78% of customers try to bypass your AI agent

Right now, somewhere in your support queue, a frustrated customer is mashing zero on the keypad. They’re typing "agent" over and over. Or saying "talk to a human."

At first, it’s polite, maybe even patient. Over time (and countless experiences), your customers get more and more irritated. And it’s not their fault.

New Twilio research finds that 78% of consumers have tried to bypass an AI agent to reach a human, and 63% want an easy way to escalate at any point. However, only 47% of brands give them one.

Your customers definitely aren’t the problem here, and your AI system might not be either. Regardless, there’s a gap here. Customers are routing around AI customer service because the systems in front of them haven't earned the benefit of the doubt (yet).

3 reasons customers reach for a human

From the dashboard, the bypass instinct looks like impatience. Up close, it's three separate problems, and they compound:

1. They can't tell who they're talking to

Ask consumers whether they can spot a text-based AI agent and 75% say yes. Test them and 90% get it wrong.

Consumers want to be told. 

• 76% think an AI agent should identify itself at the start of a conversation. 

• Brands think they're already doing it, with 81% claiming their agents disclose immediately. 

• Consumers don’t agree—they put that disclosure rate at 22%.

Somebody is wrong by 59 points. And only 32% of businesses say they prioritize AI transparency when designing agents. 

This suggests who’s really right here.

Call a bot a bot. It costs nothing, it removes the moment a customer feels tricked, and it lets people escalate before they waste five minutes.

2. The stakes feel too high

Consumers are more open to automation than most brands assume but very specific about where it stops.

They'll hand over:

• Scheduling and reservations (80%)

• Returns and exchanges (79%)

• Picking concert seats (75%)

They pull back on:

• Medical inquiries (60% won't)

• Purchases or payments made on their behalf (56% won't)

• Loan applications (54% won't)

That’s money, health, and anything with a signature at the end. Push AI into those moments and you teach customers to look for the human before they state their problem.

Remember, this is point in time research. Who knows what your customers will be comfortable with 3, 5, 10 years down the road. For now, meet them where they are—just know these things can (and will) change overtime. 

3. It doesn't work well enough

Globally, 80% of consumers report a negative experience with AI, and 19% say an AI agent made an outright mistake. Inaccurate answers are the top driver of dissatisfaction, which is why 44% of brands now list accuracy as a design priority.

These experiences are relatively new, and it isn’t necessarily a reflection of your brand or AI systems. Still, a consumer who has a bad experience with another AI sometimes puts that blame on all AI.

There's a time cost too. When an AI agent can't resolve an issue, it wastes 81 seconds per interaction on failed attempts before escalating. That's a minute and a half a customer spends proving to a machine that it can't help them, and it lands on a human anyway.

Woof.

Bad automation costs more than no automation. It burns the interaction, burns the goodwill, then hands your agent a customer who's already annoyed. Nobody wins here.

What customers want versus what brands deliver

Here are the big discrepancies between what your customers want and what they’re getting:

However, there’s a silver lining: 92% of consumers agree that brands can take specific steps to make them more comfortable with AI, and they've told you what those steps are. The escalation gap is the widest one and the cheapest to close.

Will AI replace customer service?

No, it won’t. Business leaders estimate AI agents handle around half of customer service interactions today and expect that to reach 63% by 2027, but consumers simply don’t trust AI with high-stakes tasks. Not yet, anyway. 

The more realistic model has AI absorbing volume and routine resolution while humans take anything involving money, health, judgment, or a customer who's already frustrated. Replacement isn't on the table, but a better division of labor—that is.

How to build AI agents customers don't try to escape

Now, there are steps you can take before your customers start trying to dodge your AI agent:

1. Disclose up front: Have the agent identify itself as AI in the opening line. Simple, honest, effective. This is the highest-return change on the list and it's just a copy edit.

2. Make escalation obvious and warm: Twilio Conversation Orchestrator applies rules that trigger AI-to-human handoffs based on what's happening in the conversation. The payload reaching a human agent in Twilio Flex carries the conversation history and profile so the customer never restarts from zero.

3. Give the AI something to work with: Twilio Conversation Memory maintains persistent customer profiles across sessions and channels, so an agent recognizes a returning customer instead of interrogating them.

4. Watch conversations live: Twilio Conversation Intelligence runs language operators in real time to surface sentiment shifts and escalation risk mid-conversation (which is the only useful time to catch them).

For voice, Twilio Conversation Relay handles the speech pipeline and lets you bring your own LLM, with interruption handling so customers can cut in the way they would with a person. It's a HIPAA Eligible Service when configured properly with a signed Business Associate Agreement (BAA).

Trust sets the ceiling on AI adoption

Automation is scaling faster than confidence in it. Those 76% of consumers trying to bypass your AI agent are telling you that. The brands that pull ahead won't be the ones with the most convincing bots. They'll be the ones that told customers what they were talking to, made the exit obvious, and built systems that remember the conversation. 

That's a design problem before it's a model problem, and every fix here is already available today.

Get the full picture by reading the Customer Insights Series report: How to Scale Automation Without Losing Customer Confidence. Or start building with Twilio for free.

Frequently asked questions

Why do customers try to bypass AI agents? 

Twilio research found 78% of consumers have tried to reach a human instead of an AI agent. The drivers are unclear disclosure, discomfort with high-stakes tasks, and previous inefficiencies. 80% report a negative AI experience, so it’s no wonder many still prefer to interact with a human agent

Should an AI agent tell customers it's AI? 

Yes. Twilio research shows 76% of consumers think an AI agent should identify itself at the start of a conversation. Only 22% of consumers say brands do this, against 81% of brands who claim they do. AI transparency builds customer trust and prevents customers from feeling deceived if the system makes a mistake or reaches its limits.

When should an AI agent hand off to a human? 

An AI agent should hand off to a human during high-stakes tasks, repeated failed attempts, or after any explicit customer request for a human agent. Giving customers this option at any point in their conversations satisfies the 63% of consumers who expect seamless escalation. Just make sure their full conversation context carries over to the human agent, so the customer doesn’t have to repeat themselves.

What tasks are customers comfortable letting AI handle? 

Twilio research shows consumers accept AI for scheduling and reservations (80%), returns and exchanges (79%), and seat selection (75%). Comfort drops for medical inquiries, payments, and loan applications.

How do I stop my chatbot from asking the same questions every time a customer returns? 

Twilio Conversation Memory keeps persistent customer profiles across sessions and channels, so an AI agent reads prior context at the start of a conversation. Repeat questioning happens when memory resets between sessions.

What tools support real-time handoff to a human agent with full conversation history? 

Twilio handles this through Conversation Orchestrator, which triggers AI-to-human handoffs on conversation rules. The payload delivered to a human agent in Flex includes conversation history and profile data, plus an AI-generated summary.

How do I pass conversation context from a web chat to a live phone call? 

Twilio Conversation Orchestrator connects channels into one continuous conversation, so context moves with the customer from chat to voice. The human agent receives the prior history rather than an empty screen.