Am I typing to a person or a bot?

Dear Bot, may I speak to your manager?

It seems like almost every company with a large volume of customer requests is looking into AI assistance to help them deflect (in lay terms, avoid) initial conversations from customers so that they can do more with less.

Using AI, the thinking goes, will make the experience better for the customer by presenting relevant resources to them and better for the company by waiting until the last minute to drop the team member into a chat with the customer. Ideally, that team member will have excellent context due to AI helping them understand the summary of the customer’s request and their past experience. It might even hint that the customer is kind of annoyed given a sentiment analysis of the text and the comparative speed of the keystrokes that are happening.

But are we missing the opportunity to think at the same time about improving the customer experience of typing to (and perhaps, in future, speaking to, or seeing a virtual) bot?

From a product perspective, this falls into three broad questions to answer:

1. Should a bot answer truthfully when asked if it is a bot?
2. What’s the benefit of offering a way for a human to ask for escalation?
3. How will we measure the success of using an automated bot to handle initial requests?

It also raises the question: is there a competitive moat that can be created by not using bots? (Spoiler: Yes.) Humans like to speak to other humans.

Satisfaction = Perception - Expectation

When you ask a question and you get a middling to poor or evasive answer, how do you feel? You’re probably frustrated or wondering how you can find a better answer, or might want to escalate to speak to a manager.

Compare this to getting a comprehensive answer that’s different than what you expected or wanted. You might feel frustrated, but less likely to escalate because you are pretty sure nothing will happen.

Finally, when you get a pretty good answer with details and it solves your problem, your emotions range from satisfied to pleased.

All of these results are related to your expectation – how you thought the situation would go – and the perception of how it actually went.

I’d argue that all Conversation Bots should have a feature that you could enable that would clearly enable that bot to show that it is a bot. Whether it’s using a ?? emoji or stating that it’s automated, bots need to make it easy for humans to know they are automated.

There are two primary reasons for making it clear that a bot is doing the answering:

1. Bots can do so much more than humans, but humans need to know how to talk to bots to get the right answer. (Skeptical people might say this sounds like a phone tree, but I think it can be better with generative AI tools that allow you to use plain language to ask what you want.)
2. Humans can emote much more than bots, so it’s best to flag that there is a human on the other end of the conversation who has more emotional capability than a bot.

It's important to know whether the conversation is automated or a human - because it changes the way you expect to communicate and also sets the tone for a better outcome.

Why offer an option for a human to request a non-bot?

The Toyota Product Method profoundly changed manufacturing, infusing methods of quality deeply into the production process. One of the key tenets of that system is the idea of an andon cord, a way for plant workers to stop the line by pulling on a cord and prevent defects from moving later into production when they are spotted.

Prospects and customers need an andon cord equivalent to stop a bad bot conversation. When the results or the tone of the conversation are wrong, we need to be able to ask for a human, or route the conversation to a human automatically so that the prospect is less frustrated.

Is this less efficient? Probably on the individual conversation level, but it’s key to retaining brand equity. Brands that offer a more human experience will differentiate themselves against brands that offer only bots.

How do you measure bot success in a CX context?

For the people who are going to use bots in CX – primarily to deflect initial contacts and to save money – using a bot in a CX context is valuable if it does the following:

• answers more questions faster and more consistently at a higher quality and satisfaction rate
• helps team members to answer more complicated questions in an effective manner
• doesn’t lower the overall customer satisfaction

This means you’re going to be watching the service level (during a time frame, the number of questions answered / number of questions offered), the CSAT (are prospects and customers satisfied with the experience), escalation count (how many times do people pull the virtual andon cord), and sentiment (are there generally the same proportion of happy, neutral, and unhappy messages).

Does all of this tell you whether or not to hire a bot to answer your inbound CX messages? Nope. You’ll need to decide whether you want your brand to human or robotic. By the way, another way to consider this is to think about how to help your team behind the scenes. Bots are amazing at classifying and summarizing information - this can be a bigger lift to the front line than simply handing off the answering part.

What’s the takeaway??

It’s likely that you will think about using AI to save costs and improve your CX process. But simply placing a bot on the front lines to answer questions might give you a lousy customer experience.

Why not make it possible for customers to ask for a human, or simply prioritize the humans while providing AI help and summarization for them behind the scenes?