Let's Talk About Virtual Agents and Assistants (Part 1) / by Jon Meyer Meyer

...Aaaand We’re Back

Happy New Year! I hope you all had a great and safe holiday season. I did...which turned my weekly-blog into a monthly-blog in December. Oh, friends and family, you so crazy!

So let’s get back to it.

If you work in customer service, you undoubtedly get a 3-1000 webinar invites a week for the latest tools, how to improve, etc., etc. It wasn’t until the past couple of years that I saw real uptick in virtual agent (aka vAgents, chat bots) pitches. This isn’t new technology as I have used two different platforms over the past 10 years with great success and some admissions of things that could have been done better. So I’m going to try and explain what they are and what the best scenarios are for your self-service objectives.

What is a Virtual Agent?

A virtual agent (vAgent) is basically a natural language search knowledge base in the form of a chat window (hence ‘chat bot’). These are the predecessors of Apple’s SIRI as they do the exact same thing: answer a question just as a person would.  So, yes, they are designed to be somewhat of an artificial intelligence (AI), but pretty far from Samantha. If you read this old Bloomberg article, you’d think that we’d have vAgent’s rattling down call center aisles terminating staff with their own high powered Nerf® guns - but that simply isn’t the case. 

What vAgents do really, really well is that they can convey a customer friendly support experience beyond a giant search box with bulleted FAQ links floating around it. It’s more a guide to a solution than a list of results. Also, the chat format encourages people to type sentences of their queries rather than abbreviated keywords or phrases - which can help with first contact resolution (FCR) by having more detailed keywords and phrase strings.

Some companies go the anthropomorphic route with ‘personalities’ that give the user a more personal experience. Such as Ikea’s Anna:

However, most companies today simply call them virtual agents/assistants so customers don’t confuse them with real people (as I have seen many times in the ones I've administered). Some examples to play with are Coca-Cola, Lenovo, Snapfish, and Autodesk.

There are quite a few players in the vAgent space now, and some popular companies that offer virtual agents are:

How Do They Work?

I have to be careful here, so I’ll keep it very high level on the real nuts and bolts. Basically, the customer experience works similar to a human chat agent: a customer initiates a session that includes one or more questions and responses that will hopefully deliver a correct answer. If the vAgent fails to answer the question(s), then an escalation path is offered to a live agent channel. 

Because they use natural language search (NLS), that means the secret sauce of a vAgent’s success is how well vAgent’s ‘vocabulary’ and ‘voice’ is programmed by the admin to return the proper response. It sounds a lot like knowledge base parameters, but this differs with phrases and omission words are also factored in. 

Here’s an end to end example:  a customer goes to a furniture store’s website and asks, “I want a chocolate davenport”, even a well programmed KB will likely return a null result unless that terminology is in the article and the keyword algorithm has the option to weight words or terms that are attached to synonyms.

vAgents have CSAT correlations to sessions (or conversations) where user input data is harvested for successful and failed sessions. This data harvest/correlation can be used to see what user phrases were used in unsuccessful conversations and likely escalated to a live agent. So a great vAgent admin will see unsuccessful or escalated sessions, review the conversation, and realize that the customer was looking for a “brown couch”. From there, they will add that phrase or key words into a synonym list (“Brown = chocolate, coffee, espresso”, “couch = davenport, sofa”) and omit the unimportant words (“I” “want” “a”). The admin may also have the option for keyword tweaking so certain words (and their attached solutions) will be pushed higher or lower in the results. 

Phew. Give me a second to wipe this trickle of blood from my nose after navigating how these things work without telling *exactly* how they work. 

To expand on their analytics, this type of tool has to have very powerful and detailed information to ensure CSAT. As mentioned above, trending of searches, clicks, navigation, promotions, session completion and abandons, searches, and much more. They are a data junky's dream in a tight little package that can handle unlimited customers at a time (concurrency).

For Part Two, I'll address use cases, scenarios, and KPIs.