Why AI Customer Service Fails—And How Smartrole.ai Fixes It

What we talked about.

In this episode, Sunny Ray sits down with Tue, CEO of Smartrole.ai, an AI platform for customer service training and onboarding. Tue shares his twenty year path through customer service, from frontline agent to manager, then as early employee at Danish startup Dixa, where he helped raise 165 million dollars and scale the company to 340 people before it was acquired. He explains why he pivoted to build Smartrole, a platform that flips the usual chatbot model by having AI play the customer so human agents can practice, get feedback, and become experts faster in a safe environment. The conversation covers how AI hallucinations affect customer facing bots versus training tools, why Tue and his co-founders chose to build a sustainable lifestyle business instead of chasing venture capital, and lessons learned from an Antler accelerator pitch. They also discuss the future of AI in workplace training beyond customer service, skepticism around AGI hype, and why most companies build chatbots to cut headcount rather than improve experience.

Smartrole.ai's CEO explains how flipping the chatbot script trains better customer service agents instead of replacing them.

The questions, and the answers.

Can you tell us about your background before Smartrole.ai?

I've worked in customer service for 20 years, starting as a frontline agent at Elco, then moving into a credit card fraud and authorization department where I became a manager. After that I managed customer service for an online bookstore for three and a half years, then joined Dixa, a Danish customer service platform, as employee number six before there was any revenue.

What was your experience at Dixa like?

I stayed with Dixa for about seven years. We raised roughly 165 million dollars in funding, acquired three companies, and grew to 340 people at one point. It was a venture backed company with everything that comes with that, both the good and the bad.

What problem does Smartrole.ai actually solve?

Customer service training is still done in a traditional way with classroom sessions and manual side by side reviews, so agents don't know when they're ready or what they'll be measured on. We built an inverted chatbot where the AI plays the customer instead of the agent, across phone, chat, and email, and gives agents specific feedback so they know exactly what to improve.

Where did the core insight for Smartrole.ai come from?

We didn't start with AI, we started with the problem of customer service agent training and happened to use AI to solve it. AI is really strong when it's empowering humans to become better versions of themselves, rather than trying to be a silver bullet that solves everything in customer service.

What challenges have you faced building on top of large language models?

These models can be a black box. Sometimes during a training conversation the AI playing the customer will suddenly flip and start acting like a customer service agent instead, which is a strange hallucination. We put in as many guardrails as possible, and thankfully our users aren't trying to break the model since they're just trying to get better at their job.

Why did you choose not to pursue venture capital for Smartrole.ai?

We're three co-founders who aren't in our twenties anymore, so we're realistic about what can and can't be done. During an Antler accelerator pitch I realized our own deck projected five million after three years, which isn't what VCs want to see. We'd rather stay true to the size of business we actually want to build than doctor numbers to chase funding.

Do you think AGI is close, or is it mostly hype?

I think it's mostly marketing fluff. A good test is asking whether chatbot conversations are significantly better now than before ChatGPT, and in most cases they aren't. Few people design conversational experiences well, and organizations are slow to build the self-service endpoints customers actually need.

Where can people learn more about Smartrole.ai?

The best place is smartroll.ai, where we also have a simulator so you can put in your company's website and see what the training experience would look like for your brand. I'm also active on LinkedIn sharing content about quality assurance and customer experience.