Watch all the Transform 2020 sessions on-demand here.
Last month, IBM General Manager of Data and Watson AI, Rob Thomas, told VentureBeat that IBM was the only major enterprise provider in the red-hot area of virtual agents.
Virtual agents are software that can chat with customers through text, voice, or web chat. “There really are no big players, except for us,” Thomas said at the time. He called the rest of the virtual agent providers “fireflies,” because they are small and there are so many of them.
After we published our interview with Thomas, we asked a few of the so-called “fireflies” what they thought of his assessment. We heard back from Zor Gorelov, CEO of Kasisto, and Ryan Lester, an exec at LogMeIn’s Bold360 unit, which builds virtual agents for enterprise companies. Here’s what they had to say.
Zor Gorelov, CEO, Kasisto:
Here are some interesting and specific examples of where we feel Rob is just missing the point.
June 5th: The AI Audit in NYC
Join us next week in NYC to engage with top executive leaders, delving into strategies for auditing AI models to ensure fairness, optimal performance, and ethical compliance across diverse organizations. Secure your attendance for this exclusive invite-only event.
Domain knowledge and depth is important.
IBM does not have domain specificity required to be effective in financial services. Natural language understanding (NLU) is an essential part of making an effective virtual agent. And the agent has to be well trained for the domain it will be working in. At Kasisto we believe we are able to build the most effective virtual agents for the financial services industry because we have data from 30 million+ utterances (growing at millions per month) collected from real users interacting with our virtual agents, making our virtual agents smarter every day.
Our platform, KAI, started its life as an advanced R&D project at SRI International, creators of Siri. Some of the most innovative and advanced AI technologies have come out of SRI over the decades. Rob might define Kasisto as a firefly, but we have the AI team and technology pedigree that many in the industry know, respect, and even admire. Eighty strong, our employees work long hours to deliver a conversational AI experience that financial institutions all over the world are adopting, with contracts extending 3-5 years and virtual agents deployed to millions of banking customers in North America, Europe, Asia, and the Middle East.
The assumption that chatbots are built on 1990s rules-based technology is factually incorrect.
Rob said: “I would distinguish that from chatbots, which are mostly rules-based engines. That’s not what we do with Watson Assistant. At the core of it is a model for intent classification.” KAI is built on advanced intent classification using state of the art NLU engines that leverage many of the same underlying technology and algorithms that Watson uses, and has been trained to understand precisely what customers are asking. As a matter of fact, often our customers experience 80%+ conversation containment rate (meaning that KAI completely serviced the conversation without any human intervention). This could never be achieved with a rules-based chatbot.
IBM isn’t the only company that can do “feature engineering.”
Rob said: “Any competitor can do hyper parameter optimization, but nobody other than us can do feature engineering. With something called AutoAI, we can automate feature engineering that cuts down 80% of the data science work.”
Feature engineering has been around for quite some time. It’s a method our industry has been using to help simplify how new AI models are created. IBM is certainly not alone in using automation for these methods. With that said, feature engineering itself has actually become somewhat antiquated and been replaced by more advanced methods enabled by the large amounts of compute power, data, and deep neural network algorithms. So where Rob feels this is state of the art for Watson, it’s no longer state of the art in the industry.
It’s not true that startups can’t handle large numbers of intents.
Rob said: “Most of the fireflies will serve, you know, 10 questions that they can teach the assistant to answer. But what happens when 10 questions becomes 500 questions? That’s when you need us.”
Some of Kasisto’s largest deployments have 2000+ intents and are being used by millions of users across multiple geographies, countries, and languages. So again, call Kasisto a firefly, but we are serving larger and more complicated customer deployments than Rob is probably aware of.
Ryan Lester, Senior Director, Customer Engagement Technologies, LogMeIn
Major players like IBM, Microsoft, Google, and Amazon have all made announcements related to virtual agents — assistants that go beyond rules-based chatbots to provide more free-form interactions. And numerous smaller companies, including my own, are releasing virtual agents, too. This trend is only going to accelerate as we go into 2020.
While many of these large platform investments are exciting, they are often out of reach for companies that are not in the Fortune 1000 and lack the technical resources to build on top of these platforms. So powerful but smaller, more nimble ‘fireflies’ have a major role to fill here.
Companies should be thoughtful about how and where they use solutions from the large tech providers as they often require significant development and integration work.
And that leads me to a second trend: the expansion of access to virtual agents.
Conversational AI and virtual agents are no longer just for the enterprise. Even with investment from the largest tech companies, much of the virtual agent technology to date has been out of reach for the mid-market and smaller business, due to a lack of technical talent, insufficient data to train the systems, and the cost and time of implementation. The good news is that is no longer the case. There are numerous companies working to make AI-powered virtual agents more accessible and easier to implement, even for non-profit companies.
These projects are no longer monumental lifts, and anyone looking to implement a virtual agent should see a return on investment within a year and an implementation that takes just a few months. Things can certainly get more complex over time, and therefore more transformational to a business, but in general there are a lot of easy wins that any size business can tackle.
The latest generation of tools, many of them coming from “fireflies,” is expanding access to building and managing virtual agents to a much broader audience of users beyond developers and data scientists. They do this via three channels:
- Better NLP tools that don’t require months of data training and set up
- Simplified user interfaces that don’t require code writing and have pre-built connectors for third-party data and content
- Analytics tools that help business and subject matter experts better understand how well the solution is working and where to focus next.
Last year was a year full of announcements that are pushing the virtual agent and chatbot industry forward, both in foundational technology and in business applications that will drive value for any size business.
The large technology platforms are delivering new functionality, but they can also eat up valuable internal development resources and can take time to build and implement. There’s often a bias to standardize a single AI technology platform for all projects, but virtual agents may be better designed on more purpose-built applications. The entire industry is making it easier to create and manage virtual agents and chatbots, so companies need to consider how customized they need their solution to be as they plan for 2020.
