Two numbers tell the whole story.
89% of customers say companies now push them into self-service before they can reach a person. Only 18% trust a chatbot to actually fix their problem.
Nearly every service journey now starts in a channel the customer has already given up on. People are not choosing the bot. They are passing through it on the way to what they actually wanted.
That experience is a side effect of how chatbot success gets measured. The most common scorecard gives the bot credit for ending a conversation, even when the customer leaves without a fix.
“Contained” was supposed to mean “solved”
Ask anyone in the contact center industry what containment means and you will hear something sensible. The customer’s issue was resolved without needing a person.
Now look at how it actually gets measured. A conversation counts as contained when the customer never reaches a person. That is the entire test.
Those are not the same thing. All of these count as a win.
- A customer who gives up in frustration
- A customer who types “agent” five times and closes the window
- A customer who abandons the chat and calls an hour later
This gap is where chatbots fail. Every unsolved conversation still counts as a win, so the dashboard keeps improving while more customers come back with the same problem.
A story you will recognize
Chris Karp, our Chief Growth Officer, shared a version of it at a recent seminar with CCW Digital.
A customer opens a chat to dispute a charge on their bill. The bot explains what the charge is for, and the customer leaves. The dashboard counts that conversation as contained. A success.
The customer did not want the charge explained. They wanted it removed. An hour later they call, wait in a queue, and tell the whole story again to an agent, who removes it in minutes.
The company paid for two conversations. The customer did the work twice. The scoreboard recorded one success.
Your agents already know
You do not need data to check whether this is happening in your operation. Ask the people who take the calls.
Ask how many customers open with “I already tried the chat.” They hear it all day. Every one of those calls starts in a hole, and the agent absorbs it.
The industry numbers say the same. 49% of customers do not see AI improving their experience, while company dashboards show automation improving every quarter. Both are true. They are measuring different things.
Find out for yourself in an afternoon
One month of data, one question. After the bot counted a conversation as a success, did the customer come back?
- Pull one month of conversations your bot counted as contained.
- Check whether those customers contacted you again within seven days, on any channel.
- Count how many came back about the same issue.
- The share who never came back is your real resolution rate. Expect it to sit below your containment rate.
- Read twenty of the failed conversations. The transcripts will teach you more than any report.
The same topics come up again and again. Billing disputes. Cancellations. Anything that needs a human judgment call. The bot was never going to solve these on its own, and it was never set up to pass them to a person quickly.
Most teams never run this audit, and the reason is telling. The data lives in different systems. Bot logs sit in one tool, calls in another, and nothing connects yesterday’s chat to today’s call from the same customer.
What the best teams do differently
The teams getting this right have not given up on self-service. They ask more of it.
- Give the bot work it can finish. Checking an order, moving an appointment, resetting a password. Customers happily skip the queue for these
- Design the exit like the greeting. When a conversation needs a person, the customer gets there fast, and everything they already said travels with them
- Follow the customer, not the conversation. A conversation is finished when the customer does not have to come back. That is a higher bar than any containment target.
Getting there takes two things. You need to see what happened after the bot said goodbye, across every channel, so a contained chat and the call it caused stop living in separate reports. That visibility is what Afiniti Intelligence provides. And you need automation judged on whether the problem got solved. Afiniti Agents is built to that standard, finishing the conversations it can and handing the rest to a person with the full story attached.
Your containment rate is probably fine. The question is what it is hiding. The executive brief covers the rest, the metrics worth pairing and the three things to check in every AI result.