From Call Centers to NLP: How a customer representative became an NLP trainer

Aldo Zuniga is a former customer service representative-turned chatbot trainer. Drawing from the numerous conversations he had with customers, he now trains machines to display the kind of empathy he shows to customers.

Here is his story.

Q. You spent close to 10 years at Teleperformance, which employs more than 10,000 employees for its call centers. What was that like?

A. After waiting tables and etc, my first official job was at a call center.

For close to 10 years, I took customer calls for companies such as FedEx, Comcast, Blackberry and Discover card. I worked my way up to a supervisor, trainer, then manager.

Our month-long training had to do with speaking style, as well as product knowledge. Aside from interacting with customers, we had to know the ins and outs of the knowledge base that had all the self-help articles.

Q. Did you have any software back then to help you?

We didn’t have anything like Zendesk at the time. We had to rely on our own search engine where we would type in a keyword like “damaged package” and get the answer that we translate to customer-friendly explanations.

Q. How does that decade-long period impact what you do now?

A lot of the work that I do as my job actually brings back a lot of memories of what I worked on when I was at the call center.

For a ticket management customer we have, I can think back to when I created tickets on behalf of the representatives I managed. If our wifi wasn’t working and they couldn’t do their jobs, I would create a ticket with our IT department and that IT department would create a ticket with the IT software vendor we’re contracted with. That vendor is the kind of customers we work with today.

Wizeline builds a management system that helps them triage all the tickets they get for their customers-- which vary across educational institutions, consumer goods companies.

Obviously, we now have the luxury of a self-help desks!

Q. So… now bots. How did you end up training bots to answer questions?

A. My first experience with bots was designing a conversation flow using Wit.ai. I was given an hour to put in copy at placeholders and publish a prototype. I designed a full conversation that finds a recipe for someone from start to finish. Since I had an hour, it wasn’t refined, but it had everything needed for functionality.

Q. Is training bots what you expected it to be?

I was told, ‘The position you’re gonna be placed in is as ambiguous as if I placed you in a dark room and told you to find the light switch.” It’s true. That’s how I started here with chatbots. That’s when I knew it was gonna be fun.

Q. You’ve also managed client projects as well. What has that been like?

Back at Teleperformance, I was a people manager. My job was to connect three stakeholders: the company I worked for, Teleperformance, the contractor we provided services to, like FedEx, and then the agent that I supervised. That helped me become skillful at time management and prioritizing.

Because of these experiences, I instinctively try to make the client feel understood and that their business is our business.

Q. So you must have had thousands of conversations with customers… how does that inform what you train the bot to respond to? I would imagine you have a good sense of how people ask for certain information, both directly and indirectly.

A. It’s true, those learnings can only come from human-to-human conversations.

When I covered calls coming from New York, I spoke to customers who came from various minority backgrounds, like African-American and Cuban. I would learn how they speak. I’m of Mexican origin, so if the customer was from a Hispanic background and spoke Spanglish, I would converse the same way. So it all comes down to empathy, understanding different people’s needs.

This is going to sound bad, but… I watched a lot of American sitcoms and movies, like Two and a Half men. I learned from Charlie Sheen’s way of speaking. But I promise I didn’t talk that way to customers!

I used Breaking Bad to get context into how people live in the Southwest. Will & Grace were in New York, friends were in New York, so the slang I learned from these helps with how I train bots to understand customers from various regions. They speak to bots the way they would talk to humans, so it makes sense to train the bot to understand the same things their local friends would, right?

And the last thing… Patience. Ask anyone who works on bots and they’ll tell you working with bots requires a lot of patience.

We are setting up a machine to mimic humans; training it to respond a certain way in one context and a different way in another truly takes a village.

It truly does take a village…. Of engineers, designers, and trainers like Aldo.

Klüg has a team of former journalists, customer service representatives, to take care of chatbots from the moment of ideation to full-on execution, then iteration. We have strong perspectives from domain experts for a holistic view into what kind of experience will delight your customers.

Questions? Reach us here.