Tuesday, September 24, 2013

The Confidential, Top Secret Interview: How to build a Virtual Agent

noHold, Inc. took some time this week to sit down with one of our very own Knowledge Platform Specialist, Tom Newby. He is part of a team here at noHold, who actually creates Virtual Agents for our customers. Check out the interview to gain insight to how Virtual Agents, like Siri or IBM's Watson are built. 

Preface: Today I’m speaking with Tom Newby, Knowledge Platform Specialist at noHold. With a title like that, there’s definitely a need to break it down. Tom, please tell me what does that mean? Are you like the “Dr. Frankenstein” of Virtual Agents?   
Q: What does the title mean? 
A: Being a KPS requires me to know my subject matter (which can come from multiple sources) and     intended audience pretty well so I can bring the two together with a single unified tool, a Virtual Agent.  So I'm more like a Cupid of Virtual Agents.

Q: I understand that we are under Non Disclosure Agreement (NDA) with our customers, so we cannot share specific company names however, can you tell me about some of your customers? What are they looking for when you ‘create’ a Virtual Agent that knows all about their products? 
A: Most of my customers want a virtual agent that shortens the amount of time it takes a user to get help.  Our goal is to take some of the load off their traditional help options, such as email, phone and live chat assistance.  You can only have so many human operators. We usually have to wait in a queue for the next available person to help.  But with a Virtual Agent, the connection is nearly instant, and the number of simultaneous connections is only limited by the amount of bandwidth the servers can handle.

One customer placed a Virtual Agent on the same screen as their live chat agent.  While waiting for the live operator to get to them (because their call is important to them), users are given an opportunity to try the Virtual Agent.  This resulted in a 13% drop in chat volume because users were able to find their answers on the Virtual Agent instead of waiting for the live operator.

Q: Over the years, you have learned tricks of the trade. Can you tell us about your ‘secret weapon’ for maintaining a ‘well behaved’ Virtual Agent?  
A: People are different, and many have different ways of describing the same thing.  Knowing the "slang" or alternate terms for things gives us "synonyms" that expand the agent's vocabulary.  It can be a "Monitor" to one person, a "screen" to another.  Then you have "Display", "Picture" and "that big square lit-up thing."

Logical organization of the content is also important so users can navigate easily. It’s a balance to cater to the hardcore computer expert who knows exactly what they need and the casual computer user who might be having a monitor problem.

Q: All the media talk about Virtual Assistants like Siri has made an impact on the population. People are wondering how they are built? How does a Virtual Agent answer a question, and then take an action?  

A: First you have to anticipate their questions, and then have the answers ready.  This is critical; you must have content that covers all aspects of the subject.  This can include content for troubleshooting, how-to's, and even glossary definitions.   Any virtual agent could answer Samuel L. Jackson's question about gazpacho as long as the agent wrangler puts in the question and answer(s).  Those answers can be a recipe, a link to a YouTube video, a picture, or anything else you want associated with gazpacho.  I had to look it up.  It’s a cold tomato soup.

Q: Just the other day, I told an acquaintance about noHold. With eyes enlarged, she said “Robots scare me!” Then she proceeded to tell me she asked her phone’s Virtual Assistant if her questions are being recorded. Do Virtual Agents for the Enterprise record our conversations? And … do you think your phone’s Virtual Assistant records your questions? Haha.  

A: Be afraid.  Be very afraid.  Seriously, robots are here to help, and they can't do anything we don't tell them how to do, and virtual agents don’t have robotic murder claws.

Customer sessions are recorded strictly for research (non-NSA!) purposes.  Just like a casino pit boss watching over the blackjack dealers, surveying customer experiences tell us if the Virtual Agent is doing its job.

Q: I’d love to know how you give a Virtual Agent a personality. Tell me some of the steps you take to name, choose face and colors, as well as personalize a Virtual Agent? 

A: Virtual Agent names and personality are mostly chosen at the customer level.  These are usually used to set the tone and expectation for the Virtual Agent.  For example, a Virtual Agent for a bank might want to be softer, more instructive to relate to customers:

“How can I help you today?”

A Virtual Agent for gaming computers might want a more "direct-and-to-the-point" experience, because gamers might be savvier and just want their answers without a lot of fluff, because they want to get back to blasting aliens:

“Select your option below:”

Notice how the two very different tones can be applied to the same Virtual Agent, but the content and intended audience should dictate which one is applied.

Q: Since noHold creates Virtual Agents, tell our audience about the company’s environment. That is, do we work in a laboratory or a clean room? How are Virtual Agents created? What tools do we use? 

A: Here at noHold the Virtual Agents are grown fresh on the vine, nurtured with care using only the finest natural ingredients.  No, wait, that's cranberry juice. 

We use SICURA™, a proprietary tool to build and maintain our Virtual Agents.  It provides us with tools for creating and storing content, as well as metrics reporting. We don't need a clean room, but I would love one of those foil clean suits. I would wear mine every day.

Q: Since you have access to reporting on interactions, can you think of a crazy question someone asked the VA?  
A: Our VA's have been asked out on dates and even gotten marriage proposals.  I like to have responses to questions like these just so we have an answer, and we can let them down easy.  I don't know how serious they were, but I like to think our Virtual Agents are that attractive. But if you must send flowers, please address them to me. I’ll see that the Virtual Agent get them.

Q: What is the most important “metric” to most companies? 
A: Bottom line, I think a return on investment is the most important metric to customers. Virtual Agents take some of the workload off live operators, frees up resources and saves everyone time. Some Virtual Agents also collect surveys of the user’s experiences, so of course we all like getting positive feedback.

Q: Can you give the audience a typical step by step guide to creating the best VA? 
A: In the beginning the Virtual Agent works through "growing pains".  But even as content becomes more and more complete, we still watch user experiences.  Are questions being answered correctly?  Are they getting that answer quick enough?  Are they getting enough information?  Are they happy with the Virtual Agent experience?

A Virtual Agent should be a living, growing, evolving entity.  Robot or not, an Agent's job is to assist people, save time and save our customers money.

World domination comes later.

Thanks so much for your time, Tom. If our audience has questions about building a Virtual Agent, please leave a comment below and we will respond within 24 hours.