Murali Subbarao is a founder and CEO of Parlo. Debu Chatterjee is senior director of platform engineering and is leading ServiceNow’s predictive intelligence product strategy. Debu was formerly the founder and CEO of DxContinuum which was acquired by ServiceNow back in January 2017.
I had a chance to sit down and talk with Murali Subbarao and Debu Chatterjee on our acquisition of Parlo announced on May 3, 2018. Here’s a summary of the key topics we discussed:
- Who is Parlo and why was it founded
- What is natural language understanding (NLU)
- The challenges facing enterprises in automating work with AI
- Why ServiceNow acquired Parlo and what customers can expect
- The market landscape for AI and natural language understanding
ROBERT: Murali, congratulations to you and your team on the acquisition. Tell us a little bit about Parlo and what natural language understanding or NLU is all about.
MURALI: Thank you, Robert. We are excited about this acquisition and joining the ServiceNow team. Our company name, “Parlo,” means, “I speak,” in Italian. We designed the company to enable enterprises to have conversations with their employees and their customers and enable it to happen in the most natural way that people interact.
Parlo has built a capability to understand language and expressions that people have in making requests and asking for information in the most natural way. Today, people are constrained by the limitations and frustrations of a computer application or a system interface.
ROBERT: Murali, what was the main problem you were trying to tackle when you started Parlo?
MURALI: We are engineers by background and we were looking at some of the cool technologies out there and the capabilities of artificial intelligence (AI) algorithms were very fascinating. One particular area we found most intriguing was natural language understanding. As we were looking through this we were trying to see all of the possible use cases, so we started talking to enterprises. Just as a little background, our DNA is in the enterprise space. My co-founders and I come from large companies and we worked in the enterprise as engineers.
What we heard from our conversations was that today, employees interact within the company or customers interact with the business and those interactions require human agents to be involved. People need to be involved to answer questions and follow up with requests. And time and again we heard that anywhere between 30-50% of these requests are typically mundane and repeatable with known answers. It turns out that these human agents get bogged down by these types of questions or requests and could not attend to things that required more of their time and attention. When we heard this, it looked like a very simple consistent problem whether it was related to HR, IT, procurement, finance, customer support, etc. We knew this was an area that could be attended to with the new technologies available and fundamentally having the ability to understand language.
Once you understand the way people express themselves we realized we can then transfer that understanding into a way machines can take action. This is possible. And this is a large problem we can solve...and that’s why we created Parlo.
ROBERT: I knew a little about natural language processing but I am not familiar with the concept of natural language understanding. Can you help me understand NLU a little better?
MURALI: Natural language understanding is a subset of natural language processing. In the artificial intelligence domain, NLU is a very hard problem. We as humans are used to speaking in our own way and still be understood. We can be grammatically incorrect in our expression, pronounce things incorrectly, or misuse words and still be understood by another human being. But getting a piece of software to do that is extremely hard. What is the meaning behind what is being expressed? When in a conversation, how do you relate to something that was said before? How do you remember the context of what is being said?
Just to give you a simple example, when you look at Google Search, each search stands on its own. You can search for something and then when you follow up with another search it’s a whole new search. But when you are trying to get work done, each time you speak you expect that what you said before is understood and you can relate to it and follow on to complete the action.
Natural language understanding is being able to understand the expressions made, the intent behind it, and all of the details pertaining to the conversation so you can act upon any request being made or question being asked.
Source: Understanding Natural Language Understanding, Bill MacCartney
ROBERT: Debu, what was it that you liked so much about Parlo?
DEBU: First of all, welcome aboard, Murali! We are really going to enjoy having you on our team and your technology on board with our Platform.
So, Robert, ServiceNow is embarking on a series of things that are on our AI roadmap that started with our DXContinuum acquisition, then our Qlue acquisition, and now Parlo. With Parlo we realized they bring us a deep understanding of AI and natural language.
Natural language understanding is extremely critical, specifically for ServiceNow because we deal with enterprises. Each enterprise has their own “enterprise language.” For example, if there is a communications company and they are talking about networks and data being moved from one place to another that involve routers, etc., that has a specific context in technology. Now compare that to a healthcare enterprise that also uses the term “network” where the meaning and context of the word is entirely different – for example, a “network” of healthcare providers. This means we have to understand how language is spoken by our customers inside and outside of their enterprise. We need to understand all of this in the context of the entities of these conversations. We need to understand this in a way that we can make sense of work requests and questions being asked so that the right things happen and the right outcomes achieved.
This technology from Parlo, which is primarily driven by deep learning, is another subclass of artificial intelligence and very valuable to ServiceNow and our customers. Natural language understanding is a very hard problem to solve and requires deep learning. With Parlo’s technology in place, we will re-platform it and make it work like all of our products: in our cloud, with our customer’s data, and make it useful for the problems our customers have.
And finally, let’s talk about the people and talent of Parlo. As you know, there is a talent war here in the Silicon Valley. Finding experts in artificial intelligence is not easy. Parlo brings us the people that have experience with machine learning and deep learning and the experience of solving real practical problems. We expect the Parlo team to contribute significantly to our product roadmap.
ROBERT: Debu, one of the obvious use cases I thought about with Parlo is our new Virtual Agent (chatbot) capabilities announced in the London release. How do you see our new NLU technology from Parlo working with Virtual Agent?
DEBU: Virtual Agent is actually helping us by understanding how our brains think and work. It involves getting input from multiple channels and with that input, it is trying to parse it and understand what needs to be done. With Parlo, we can actually help this get done better. Now we go from understanding one intent to understanding multiple intents when it comes to getting work done.
Let’s take an insurance example. Say a person want’s to know, “Do I have insurance coverage and do I also have death coverage?” Now those are two different intents in the request and both of the intents need to be parsed and understood. Having an NLU capability allows us to understand both intents and execute a specific action. The NLU technology will also give us the ability to determine what actions need to be called so that each of the entities can be understood. All of this takes our existing Virtual Agent technology and puts us way ahead of the competition. This is something our competitors cannot do “out-of-the-box” and we will be able to do this “out-of-the-box.” We will also be able to do things like answer, “what are the areas in which conversations are happening,” so we want to understand the topics and intent for building out these conversational chatbots. That’s the power of NLU.
In the future, all of our products across ITSM, HR, Customer Service, SecOps, and Intelligent Apps will be able to take advantage of this new technology to improve our customer's SLAs (service level agreements) and get issues, cases, and incidents resolved faster than ever before. This will deliver a more empathetic experience for employees and customers
ROBERT: If I’m a ServiceNow customer and I am sitting on a ton of knowledge base (KB) articles, how will I get value from this new NLU technology?
DEBU: Many of our customers have published numerous KBs and every time an incident or case comes in, one of the challenges is to make sure the fulfiller gets the right knowledge article to the requestor or customer. The fulfiller can do a search and that can help narrow things down. However, if you can really understand the context and summarized intent of the knowledge article and map that to the problem being stated, you can achieve the right customer outcome. Our new AI technology makes that mapping possible by improving the relevance of the KBs to request and resolve issues much faster.
The Parlo Team - Sunnyvale, CA
ROBERT: Debu, when can our ServiceNow customers expect for this to become available?
DEBU: As you know, we have two releases per year. The first thing we are going to do is to re-platform Parlo’s technology and our expectation is that our customers will see it in the New York release at the end of 2019.
ROBERT: Murali, can you talk a little bit about the relevant market landscape for NLU and AI?
MURALI: If you look at NLU and AI there are the big six: IBM, Microsoft, Google, Amazon, Facebook, and Apple. These are companies that are very active and you hear quite a bit about them in the news. Some of them deliver consumer-facing products that have AI capabilities. And these products address a broad range of consumer use cases. Parlo focuses on dealing with the enterprise use cases. How do we address the need for NLU in the enterprise?
There are four major differences that Parlo brings to the table for enterprise customers. One is in our ability to create what we call our “understanding model.” When you join a company, there is a lingo in that company. It could be domain specific, product specific, or even a geographic specific thing. We enable the creation of a language model that is specific to that enterprise. Secondly, having the ability to comprehend things in a rich way. Debu did a good job of explaining the complexity of intent and expressions. That is being able to create an understanding based on a natural way that humans express themselves. Humans are smart enough to downgrade themselves to the fifth-grade level that’s necessary based on whom they are conversing with by speaking in shorter bursts and using words that are more easily understood. Parlo differentiates itself by being able to elevate the level of understanding so that the comprehension is higher. Thirdly, is the refinement of expressions. The way we speak changes. In the enterprise use case, there are new products and new concepts so the vocabulary needs to expand. These are the shifts in enterprise conversation. If you don’t refine the understanding rapidly enough, the software you build or the agent technology you build gets out of sync and has blind spots of understanding. Parlo is really good at enabling that refinement in a continuous way. And the last one is the ability to deploy to the enterprise in a private instance where the data is securely managed by the enterprise and can stay within the enterprise for the model building and tweaking that needs to happen. So those are the four major differences that Parlo delivers from what is commonly available in the NLU space.
ROBERT: Gentleman, thank you for your time. This was a fascinating conversation!
DEBU: Thank you, Robert.
MURALI: Thank you, Robert. It’s great to be part of the ServiceNow family.
