Q&A with Glenn Russell, Co-founder and CEO of UpSellit
How does traditional chat (no face) differ from avatar based chat?
UpSellit’s SMARTsales Agent does not appear as an avatar, but rather a chat box.
Most of the traditional chat (no face) out there is delivered by live chat agents in a customer service role. Not only is this a large expense for businesses, but also not the most cost-effective. Between chat agents’ poor spelling and grammar, sub-par reading comprehension skills and tendency to be slow or worse yet idle the experience is not ideal. Live agents also bring about issues of scalability: sites with slimmer margins may not be able to afford to handle hundreds of conversations a day when the average customer might only spend a few dollars on the site.
UpSellit tackles this problem by delivering online retailers an easy, effective and affordable way to connect with customers online to increase purchases and improve the purchasing process. Our proprietary virtual-chat technology initiates a chat session with shoppers at predetermined points of action resulting in increased sales, additional promotional opportunities and greater customer satisfaction. Using natural language parsing and approved company messaging, UpSellit’s SMARTagent engages with the shopper to determine the reason for the departure and offer them relevant incentives to complete the purchase. So far, we’ve saved our customers using UpSellit more than $75 million in revenue that would have otherwise been lost.
What is a typical amount of time it takes to set up the intelligence for an intelligent agent? What determines the length of time?
One of the greatest strengths of our company is our knowledge in a wide variety of verticals. From dating sites to e-commerce, from lead-gen to credit-card companies, we’ve worked with companies as big as fortune 500 to one-man yahoo-shopping cart sites. Using our patent-pending natural language processing engine our professional scripting team can quickly enter a complex set of regular expression and decision table rules based on the partner company’s FAQs and site copy. We leverage past experience with similar partners and can fairly accurately predict what will be asked and use that to aid in the script processes. Based on this, a fairly robust first edition of a script will take between 2-3 hours.
The main thing that can add some extra time to our script creation processes is a lengthy re-write process, typical due to the legal team on the partner’s side. Our scripting team draws as much as possible from the site’s own content, as their own text is most likely approved, but often answers need to be generated based on what we can glean from the site’s FAQs. A very picky legal team can force re-writes and stretch the processes.
Could you also describe a bit about one or two customers who are using intelligent chat?
GourmetStation – When GourmetStation, an online retailer that delivers gourmet dinners and other upscale food gifts, decided to implement UpSellit’s virtual chat technology it was important both to increase sales and also offer a personalized chat experience that reflected the company’s unique brand.
“Our virtual chat agent is named Susan. We designed her to embody all the best qualities of our employees. It is important that Susan is like our brand; elegant, sophisticated and helpful,” said Donna Lynes-Miller, founder of GourmetStation. “In just five months, Susan has converted enough shoppers into customers to deliver a 10 percent increase in our sales. Our customers love her, and so do we.”
What are the costs?
UpSellit works on a pay-per-performance model. There’s no set-up fee and our clients are only charged when the SMARTsales Agent closes a sale, similar to a sales agent at a store working off commission.
Some people might be concerned that an intelligent chat vs a human chat person could be a turn off. Any comments?
First off, when it comes to our use of intelligent chat vs. a human chat, we believe in complete transparency surrounding the chat experience. During any conversation with a SMARTsales Agent, you can ask them if they are real. Their response: “I’m a live virtual agent.”
In addition, for small businesses, the reality is that human powered chat (through call centers) is a very expensive option that doesn’t necessarily deliver consistent results. With SMARTsales Agent, you know exactly what you are getting, 24 hours a day, 7 days a week, 365 days a year. I don’t think the intelligent chat turns many people off, as our clients see a 12% increase in revenue as a result of the SMARTsales Agent and intelligent chats.
Furthermore, SMARTsales Agents are developed using a company’s FAQ and approved company messaging and branding. In this way, our clients can have complete control over the dialogue. Also, all the conversations are logged and collected so that a client can review and decide whether to adjust the conversation or change tone or product offerings.
How does one go about selecting the technology provider to use for chat?
Depending on a company’s needs, there are two important criteria to evaluate when looking for chat technology- the quality of the tech foundation powering the chat and the scalability and affordability of the service. For a growing business, it’s important that the chat is flexible and scalable, that you can go from operating say, 50 conversations a day to 50 conversations an hour.
We know firsthand how arduous the process is of perfecting the chat bot and how extremely important the quality of the conversation is to the success rate of the virtual agent. The UpSellit technology was years in development and tested on thousands of unwitting would-be internet addicts over AOL instant messenger. The technology was tweaked and evolved in such a way that it attempted to gleam true meaning out of the chatter on the other side’s input, not simply respond to a list of keywords.
The early versions of the technology that evolved into SMARTagent were
really no different than the other bots out there, including Jabberwacky,
Alice and a score of others that never found a real business use. Those
that did find a practical use tended to draw very little on the ability for
the program to converse with a human. For example, ConversAgent is little
more than a keyword driven interactive menu system (“You said ‘payment’.
Were you asking A) How do I make a payment, B) What payment types do you
accept or C) How much do I owe as payment this month?”).
What differentiates us from all the bots out there and makes us marketable
can be illustrated best in the cliché office joke “Are you working hard, or
hardly working?” The joke is based on the fact that the two words “working”
and “hard” mean exact opposite things when used in reverse. A bot would see
these words and be limited to one meaning. Natural language parsing
attempts to actually understand the sentence, not simply inspect the
This said, our technology is not flawless, nor can it understand a lot of
what people might say in day-to-day conversation. Certainly if you ran the
individual sentences of Steinbeck’s “Old Man and the Sea” through a
SMARTagent she would not be able to respond to a vast majority of the
sentences. And at the end of the chat log the SMARTagent would certainly
not understand the summary of the story it was just told.
Fortunately, the above is not what we’re trying to achieve. What we care
about is selling the product, offering a discount, and informing the
customer of the site’s FAQs (shipping, returns, pricing, etc). Many of the
questions and concerns about the product and the company selling the product
are similar or the same between chats. And those pieces of input that are
off topic from the sales pitch or FAQs — as most of the Steinbeck’s novel
would be — are categorized as “not understood” lines and responded to with
a default line. These default lines are sales/rebuttal/informational
general lines that try to bring the conversation back on point. The “not
understood” lines are later analyzed and the correct information is given to
the SMARTagent so the next time that question comes up she will have a
The marketability of the product and the function it can serve (handling
tens of thousands of chats on an extremely low, performance based model) is
fairly straight forward and simple. However, the complexity of the
underlying technology is what allows for this to be done in a way that won’t
provide a confusing or frustrating user experience.