Ashley Garcia closed deals at such a high rate that she became the top performing salesperson for the Phoenix Suns. The basketball franchise’s sales department responded by celebrating Garcia, which boosted morale for the close-knit team.
There’s only one problem. Garcia isn’t a real woman, but a conversational agent created by an AI company on behalf of the Phoenix Suns to converse with fans and, ultimately, sell tickets and other merchandise.
The fact that Garcia is a chatbot doesn’t detract from her considerable achievements with the Suns, according to Jim Kaskade, the CEO of Conversica, the San Mateo, California company that developed the agent for the Suns.
“Some people really embrace it, like sport teams,” Kaskade says. “They really embrace this idea of a digital employee by celebrating their anniversaries and stuff.”
Conversica was founded in 2007 by Ben Brigham to develop conversational agents. Back then, natural language processing (NLP) technology was nowhere near as good as it is now. But when Kaskade took the reins of the company in late 2019, he had an inkling something big was on the horizon.
“I went on a walk with the head of AI at Google in CTO office, because they’re a friend of mine, and he talked about this,” he says. “He said, ‘Jim, you don’t realize what’s coming.’”
In early 2020, Conversica used APIs to tap into one of the first publicly accessible large language models (LLM), OpenAI’s GPT-2, to help it generate more natural responses to questions. It was better than what Conversica had before, but still not up to par. So the company set out trying to build its own LLM.
“We failed miserably, because it requires a massive amount of data, a lot of compute, and a sh*#-ton of money, and we kind of hit the wall on that,” Kaskade tells Datanami. “We tried everything. We were so far ahead of people in terms of engineering the possible.”
Today, Conversica’s agents run on a combination of models, including LLMs from OpenAI but also smaller language models that are custom trained to understand the words and phrases in each client’s industry.
“That allows us to answer 99% of the questions based on our teams’ data,” he says, “and the 1% of the questions that aren’t team specific, like, ‘Do you surf? Who is the president? What’s it like to go to Paris.’ Whatever the question is…that’s not domain specific…we go out to the public LLM to answer that.”
Conversica has attracted 1,100 customers over the years, including about 40 professional sports teams across 11 leagues. Sports has become one of Conversica’s most popular verticals, but it also counts customers like AT&T, Visa, and Oracle, so it has a well-rounded roster of blue chips.
Conversica specializes in chatbots for sales and marketing organizations, and that focus has allowed the company to build chatbots that are highly tuned to those tasks, which don’t change much from company to company, despite the differences in industry vertical, Kaskade says.
“We’re automating almost every task you can think of trying to make money and selling to somebody, whether it’s a prospect or an existing customer,” he says. “Our agents can talk to you in email, SMS, messaging apps. You name it, we can have a conversation through that digital medium.”
Conversica has done its best to inject personality into the agents, too (which might be why customers like the Phoenix Suns give them names).
“You can actually have some fun with it. That’s the beauty of this generative AI technology era is it interests you and it understands your intent,” Kaskade says. “It can mirror your sentiment. If you’re pissed, it’s not going to have the same jovial [approach] ‘Oh, you know, things are great!’”
Kaskade says his chatbots are so good that you can’t tell if they’re real or not. In fact, Kaskade challenged this reporter to tell the difference between one of his chatbots and a real person. (For the record, this reporter has never been stumped in that particular department.)
Turing Tests aside, the thing that really matters is the bottom line, and in that department, the Ashey Garcias of the world are doing quite well, thank you.
“The economics are phenomenally over the top,” Kaskade says. “We find, across our thousands of customers, because we’ve been in business so long, we get a 24% engagement rate, when clients are like in the 1% to 2% range and trying to get that to 5%.”
The conversion rate, or converting a lead to a sale, is not as good: 12%. But that’s much better than the industry standard, which is less than 1%, says Kaskade, who guarantees a 10x return on investment.
“Once we get them running–and we get them going quickly–it just works,” he says. “We’re printing money for them.”