The #1 AI behavior hurting your customers’ experiences

The #1 AI behavior hurting your customers’ experiences

91% of brands say customer satisfaction has dropped because of something their AI did.

Well, that’s pretty vague, so we dug a bit deeper: which behaviors specifically? We got the answers back, and here’s what we found.

It's not AI hallucinations. Nor is it tone. It's the conversation falling apart when a customer changes channels.

4 AI behaviors hurting your brand the most

These are ranked by the share of businesses citing each as a top driver of dissatisfaction:

1. Losing context between channels or interactions: 40%

2. Lack of empathy or emotional understanding: 39%

3. Inability to handle exceptions or non-standard requests: 38%

4. Failing to understand customers or provide relevant responses: 36%

These are all fairly close, so you can treat them as a cluster of complaints. Really, though, what’s useful for teams build AI experiences is looking at what’s not on the list

You don’t see speed or clunky models. You don’t even see accuracy or transparency.

Three of the four are context problems. An agent that can't handle an exception usually lacks the history that would explain the exception. An AI agent that fails to provide relevant responses usually can't see what would make a response relevant.

Why changing channels breaks your AI

Customers are rarely tied to a specific channel. They simply use the one that’s most convenient at the time. If they’re on their computer experiencing a SaaS problem, they’ll probably open a webchat, submit a ticket, or send an email. If they’re on their phone, they’re more likely to make a call or send a text message.

Sometimes, though, they go through all of these different channels over the course of a day, a week, or even a month. However, your tech stack isn’t connecting all of these interactions as one ongoing conversation. No, it likely sees three sessions across three tools with possibly three different identifiers.

Here’s what’s going wrong:

• Session boundaries: Most conversational systems scope memory to a session. Close the window, and the context closes with it.

• Identity mismatch: The phone number on Thursday's call and the email from Tuesday's chat resolve to different records unless something stitches them together.

• Channel-specific tooling: Voice runs on one platform, chat on another, SMS on a third. Each holds a complete record of its own slice.

• Batch timing: Even when systems do sync, an overnight job means today’s agent is reading yesterday’s picture.

This isn’t a problem with your agent model. This is an infrastructure issue.

What broken conversations costs your business (and customers)

Before, the businesses cited what AI issues are hurting their brand. Now, let’s look at what the customer-side numbers say:

• 74% of consumers report repeating themselves to AI agents.

• 76% say the human agent they reach has little-to-no context about their issue.

• 71% will abandon a conversation entirely if an AI fails to recognize who they are.

In Twilio's published report Inside the Conversational AI Revolution, 49% of consumers say AI never resolved their issue, and 40% say it repeats itself or gets stuck in loops.

People abandoning a conversation isn’t a drop in satisfaction. That’s people completely leaving mid-conversation, and some of those disgruntled customers are never coming back. They don’t even leave a complaint you can count in your metrics. They’re just gone.

How to make context survive a channel change

The fix is structural. Context has to be attached to the customer instead of the session.

• Model the conversation as one thread: Twilio Conversations treats chat, messaging, and voice as a single lifecycle, so a chat that becomes a call carries the earlier exchange with it.

• Keep the profile current: Twilio Conversation Memory syncs new interaction details into existing profiles in real time, which closes the batch-timing gap.

• Route on conversation state: Twilio Conversation Orchestrator applies rules across the whole conversation, so escalation and handoff decisions account for what happened on other channels.

• Resolve identity before you need it: The stitching between a phone number, an email, and an anonymous web session has to happen ahead of the conversation. Doing it at runtime is too late.

Start with the transition that generates the most repetition. For most companies that's chat to voice, because it crosses both a channel boundary and a tooling boundary at the same time.

What to measure

Containment and CSAT don’t surface these problems because both average over conversations that never crossed a channel. Here are a few things that’ll help you better monitor the situation:

1. Repeat-explanation rate after a channel change, approximated by customers restating their issue in the first two turns of a new session.

2. Recognition rate, meaning how often an agent opens with prior history attached versus a blank profile.

3. Abandonment at the transition point, tracked separately from overall abandonment.

If your cross-channel numbers look materially worse than your single-channel numbers, you've found the problem.

Fix the structure before you change the model

The first instinct when AI underperforms is to improve the AI. That’s not what this data supports, though. 

The ranked data points somewhere less interesting (but more fixable): the places where one system hands a customer to another but drops everything it knows.

Pick one seam. Make context survive it. Then measure whether people stop repeating themselves, because that's the behavior customers are describing when they say AI support doesn't work.

However, if you do need to change models as well, Twilio Agent Connect makes it simple and easy. You’re never tied to a single model.

Don’t just take our word for it, though. See for yourself. Read the Customer Insights Series guide on navigating the data deluge, or see how context and trust interact in How to Scale Automation Without Losing Customer Confidence. Ready to build? Start with Twilio for free.

Frequently asked questions

What is the biggest problem with AI customer service? 

Twilio found losing context between channels or interactions is the top AI behavior damaging customer satisfaction, named by 40% of businesses. 91% of brands have seen satisfaction drop from an AI misstep.

Why does my AI forget the conversation when a customer switches channels? 

Most conversational systems scope memory to a session and run voice, chat, and SMS on separate tools. Twilio Conversations models those touchpoints as one lifecycle so context follows the customer instead of the channel.

How do I measure cross-channel context problems? 

Track repeat-explanation rate after a channel change, recognition rate at the start of a conversation, and abandonment at transition points.