Thoughts on AI and Customer Service

I’ve started to work on integrating AI into Herodesk. Using the OpenAI SDK (software development kit) isn’t that difficult, but as I’ve said many times before:

It isn’t difficult to build a chatbot. It’s difficult to build a chatbot that doesn’t give useless replies.?

In all my work with customer service and building Herodesk as a customer service platform, I always start with the same goal: Make it easy!?

It should be easy for the customer to get the help they need.

If they need to get in touch with a human being, that should be easy.?

Likewise, it should be easy for my customers to sign up for- and get started with Herodesk.?

Great customer service is about making it easy for the customer.

Regardless of whether it’s my customers using Herodesk or my customers’ customers who need help.

Therefore, I don’t mind making decisions on behalf of my customers about setting directions for how customer service should be run or how Herodesks should work. I know it’ll make it easier for them to give good customer service (read: make it easier for their customers).?

Now, integrating AI into customer service can definitely make things easier—if done right. If not, it’s just plain annoying.?

When integrating AI into Herodesk, we’ll be doing it in three stages:

Stage one: Text manipulation

The first stage will contain tools to make it easier for my customers to give great service their customers. It’ll be a set of tools that, using AI, will improve the functionality of Herodesk by analysing and manipulating text.?

This will include functions such as:

• Translation from and to foreign languages. If a customer contacts you in Italian, you’ll see the text in English (or whatever your preferred language is). When you reply in English, your text will be translated into Italian, so the customer receives a reply in their native language.
• Text manipulation. Change your text's tone, length or temperature before sending it to the customer.?
• Text analysis. Use AI to analyse an inbound message. Fx. If the customer seems angry, automatically assign the conversation to Bob and mark it as “important”.

These are just three examples, but I hope you catch the drift.?

Functions like these will enhance Herodesk’s functionality and make it easier for you to help your customers.

Stage two: Suggested replies

Now, things start to get a bit more interesting.?

By training an AI model on various data sources (your website, ToC, FAQ, help-centre articles and all previous replies in your Herodesk), this model will prepare a suggested reply for you when you open a conversation.

It won’t send it—it’ll just be prepared so you can make the required changes before sending it.?

To do this, some new stuff is needed…

• Helpcenter / FAQ support. Yup, we’re gonna make that. You’ll be able to create your own knowledge bases (note the plural - you will be able to make public and/or private collections of articles) and FAQs in Herodesk and display them publicly and/or embed them onto your website.
• Significantly more advanced implementation of AI, including configuration and training of the model.?

Training the model is essential. This is what ensures that replies are sensible and actually helps the customer. There’ll be ways to do this in Herodesk, but by noting the difference between the suggested replies and the actual replies, we’ll be training the model “behind the scenes” as you reply to customers (again: make it easy).?

Stage three: Chatbot

By now, we have all the prerequisites to release the full power of AI in Herodesk and offer a fully automatic and autonomous chatbot!?

This will be implemented into our existing Live chat.?

Before I proceed, let’s do a quick raise of hands: How often have you been greeted by an AI chatbot and instantly started to type “agent”, “human”, or similar to try to find the keyword that skips the bot and sends you to a human? I know I have… Why? Because those bots were utterly useless.?

We don’t want that. That’s not easy. That’s annoying. So, instead, we’ll be doing this:?

When a customer starts a new conversation in the Live chat, the chatbot will attempt to answer the question.?

After the first reply, the customer will get the option to have their conversation moved to a human for a reply (with a warning about the expected reply time).?

At any time during the conversation, the customer can opt to switch from the chatbot to a human. Make it easy.?

When a conversation ends, the customer will be prompted to tell whether their initial problem was solved. This information will be used for training.?

Also, Herodesk admins and Managers will have access to all completed chatbot interactions to review them and make changes and improvements to train their chatbots.?

These are very down-to-earth examples and descriptions, but I hope it gives an idea about the overall direction we want to go with AI.

Timeline, you ask? I expect that we’ll have stages one and two ready this year.?

These are my initial thoughts on AI and customer service. They’ll most likely change as work progresses, features are launched, and feedback starts to come in. Honestly, I have no idea how this ends up, but I'm super excited to start working on it. And in relation to that, I’d also like to hear what you think.?

How do you think AI can make customer service easier? Both as a company giving the support and as a customer who needs it.