Getting Emotional Online

You can learn something from a Chatterbot

If you traded an e-mail or two with Tom Rearick, Director of Self-Service Products and creator of eGain Assistant, you wouldn’t know that he’s a music aficionado. But get into a discussion with him about online customer service and a music lesson gets thrown in for free.

“Like mathematics and physics, music from the Classical period was an expression of disembodied truth. Classical music gave listeners a glimpse of perfection. At the end of the Renaissance, a group of Florentine academics promoted a “Doctrine of Affections”. It was a radical idea that music could include an embodiment of ‘Affekt’ or an emotional component. For example, ‘sadness’ might be represented by a lamento bass whereas joy might be represented by a rapidly rising sequence of thirds. The result was the Baroque period. Baroque would never be as sensual as the Romantic period to follow, but it was the first and music has not been the same ever since.”

He then he’s off on a discussion of how Bill and Larry (Gates and Ellison) may be the classical composers of the Information Age. Where’s the emotion in a Pentium chip? Why, he asks, are we trying to improve the human/computer interface when it’s clear that it’s been inhuman from the start. When coupled with our current condition of customer relationship mismanagement, it’s heartening to think that Rearick is actively working to make the Data Epoch a little more… romantic.

To find out if my shipment will be late, first I have to know that there is a tracking feature I can use. Then I have to figure out how to use it. In the past, using a command line interface, I had to know the specific command required to get the specific information I was after. As software became more and more complex, my ability to remember was thwarted by an ever-expanding lexicon. Next came the graphical user interface which allowed me to shift from knowing the command to finding the right button; a shift from memory to recognition. Over time, these interfaces have become more versatile.

Ask Jeeves and the companies who license it; Ask Maxwell (Microsoft), Ask IRIS (Toshiba) and Ask Martha (Martha Stewart), will reply with the questions it knows the answers to. That is, it will try to match your question with the nearest thing in its database. If one of their questions is a match, they can point you to the answer.

But these click-and-pray and type-and-hope interfaces leave one unsatisfied. With, what Rearick refers to as a *social* interface, I type in my question and rather than the typical computer responses (Order #12345: shipment pending) I am told by an attentive – although imaginary – customer service representative, “Your order, number 12345, is pending shipment.” Instead of being stymied, wondering what “shipment pending” means,  with the eGain Assistant, I can ask.

At the eGain Web site, you have the opportunity  to chat with Lisa the Klone, who can answer questions and solve problems automagically. Beyond the initial intrigue of communicating with a chatterbot, it’s engaging. Why is this interface so compelling? To start with, it’s the first step toward making the computer think like a human, rather than making the humans think like a computer.

When I ask Lisa the meaning of the term “shipment pending,” she (Yes, *she*. See how this works?) can ask me follow-on questions that will help her get to the root of my problem. The result is that I am more willing to reveal information about myself. It’s not a database query, it’s a conversation.

Rearick is quick to point out that people using Lisa generally finish their conversation by typing in, “Thank you.” Even if there were such a button on an automatic teller machine, why would you press it? But interacting with Lisa for even a short time makes you understand why he calls the interface “social”.

If Lisa seems to care about me it’s because Rearick has carefully crafted the eGain Assistant to do just that. She can textually and *visually* communicate as many emotions as you please. This is the icing on the cake; Lisa has a face. That means Lisa can communicate emotions as well as data.

For Lisa to do exhibit the appropriate feelings, you’ve got to assign an emotion to the answer to each question you expect customers to ask on your Web site – and then construct that emotion. In order to express a feeling of delight, as in, “I won the lottery,” the photograph of Lisa’s face would show a big smile, with her teeth showing and her tongue visible. How important is that tongue? You’d be surprised.

If Lisa only had three or four faces, interacting with her might be informative, but it wouldn’t be engaging. We need to differentiate a delighted face with one that is merely very pleased. To express very pleased Lisa needs the great big smile, but we limit the exposure to teeth alone. Just-sort-of-pleased would be communicated with the lips together.

Here are a handful of examples to give you a clue just how specific one has to get to create an captivating chatterbot:

Emotion: Neutral
Description: Paying attention, waiting, explaining
Expression: Head tipped forward and to one side, raised eyebrows
Example: The conference room is down the hall on the right

Emotion: Alternative Neutral
Description: Paying attention, waiting, explaining
Expression: Mouth relaxed, slightly parted, eyes slightly wide, slightest of smiles
Example: Yes, I might be able to help you

Emotion: Remorse
Description: A combination of shame and distress
Expression: Arched and knitted eyebrows, frown,
Example: I’m sorry that I let you down.

As the creator of eGain Assistant, Rearick understands how to split hairs. Look at the subtle, yet important differences between the following two pairs:

Emotion: Perplexed1
Description: Confused and worried
Expression: Knitted brow, looking downward or askance
Example: I can tell I’m not doing very well.

Emotion: Perplexed2
Description: Confused and frustrated
Expression: Lips pursed, knitted brow, impatient
Example: You’re confusing me.

Emotion: Gratification
Description: Pride and joy
Expression: Smile, head tipped up, eye partially closed
Example: Thank you for the flowers.

Emotion: Gratitude
Description: Admiration and joy
Expression: Smile, chin down, eyes wide
Example: I owe you big time.

So you think you’re ready to get a chatterbot to help you out? First you have to decide whom you want as the “face of the company.” What role do they play? Some companies have imaginary customer service reps. Some companies use real CSR’s. I’ve heard of one German company that has implemented Dave, the Virtual CEO who can answer questions about the firm’s fiscal situation including plans for an upcoming IPO. You might even want to chooses a cartoon character. Either way, Rearick has enough experience at this to have a few suggestions.

“Most of the advice is pretty straight forward,” he says. “Consider models with large, expressive facial features. Photographic portraits should be frontal facing only. Cartoon renderings should be three-quarter or frontal facing. Cartoon characters have a tradition of speaking to one side into a speech bubble but when using photographed talking heads, put the spoken text below in a box like subtitles.”

So what can we learn from a robotic, database driven, pretend customer service chatterbot? There are two important, but unintentional outcomes from the implementation of this sort of tool. First, the whole panoply of questions and answers that your CSR’s handle day in and day out is revealed, reviewed, and revised. When you’ve determined the best response to a particular frequently asked question, you can feed that answer back into the system for your human representatives. Face it; when was the last time you updated your boilerplate answers?

But much more interesting, is the exercise of assigning an appropriate emotion to every question/answer pair. Service reps may have been given the hour long speech on empathy when they first started, but nothing opens your eyes to what it’s like being a customer like discerning the definitive demeanor that should accompany each response.

As a new technology, a chatterbot like eGain’s Lisa starts out as a curiosity. She’s fun and she’s entertaining for three or four minutes. But customers can learn from her. Customers can confide in her. Customers can get questions answered and problems solved any time of day or night and without tying up your call center lines. If they *do* call the call center, your team  is just a little bit more sensitive. When it’s all over, when final strains of the concerto waft away on the evening breeze, your customers end up with better service, and applaud. Now *that’s* music to every customer service manager’s ears.