[Title slide 1. Blue CAI company logo with tagline “We power the possible” appears in middle of screen. Company website www.cai.io appears at the bottom center of the screen]? [Title slide 2. Multi-color background with text centered in the middle of the screen that reads: “Virtual Event: The 3 Cs of Intelligent Automation: Using next-gen AI to resolve service desk issues in 7 seconds”. The white CAI company logo appears underneath of this text towards the bottom of the screen] [Two speakers appear on screen. Christina Kucek, CAI, is on the left and Christian Ventrigila, CAI is on the right.] 00:00:09 - 00:00:50 Christina Kucek Hello, and welcome to our latest session in a new series of 30-minute CAI learning sessions, the 3Cs of Intelligent Automation. Why do we call it 3Cs of Automation? Well, first, this series is brought to you by CAI. We're a global technology services company with a 40-year history of combining our dual strength of talent and technology to deliver lasting results to the public and commercial sectors. The second 2 Cs represent your host for the series. I'm Christina Kucek, executive director of intelligent automation at CAI. This is my colleague, Christian Ventriglia, UiPath Architect, and CAI UiPath MVP. 00:00:51 - 00:01:26 Christian Ventriglia The purpose of the learning series is to get under the covers on everything, intelligent automation, practical use cases for automation and technology advancements that drive efficiency and increased productivity. So sit back, grab a beverage, and learn how to hypercharge your automation teams with tips and tricks from our expert guests. If you have any questions, we encourage you to ask them in the chat. We will try to get to all your questions by the end of the session. If at any time you want to learn more, visit our website at cai.io for articles, client success stories or to set up a discussion with someone on our team. 00:01:27 - 00:01:58 Christina Welcome, everyone to today's Automation Learning session event, Improving Customer and Employee Experience at the Service Desk with Conversational AI. My name is Christina Kucek. Briefly about me, I'm passionate about assisting clients in their automation journey from building automation teams for RPA and document extraction to machine learning and artificial intelligence. Our solutions drive efficiency, cost savings, and a competitive advantage. With me is my co-host Christian Ventriglia. 00:01:59 - 00:02:30 Christian My RPA journey began nearly six years ago with a request from my manager to research an emerging technology called Robotic Process Automation. Fast-forward to today, and I'm a USN certified RPA Solution Architect with a passion for helping clients drive innovation using intelligent automation. I have a demonstrated history of delivering solutions across various business units, and I'm currently passionate about integrating machine learning and AI capabilities into the digital robotic workforce. 00:02:31 - 00:02:44 Christina All right, let's get started. In today's 30-minute discussion, we're going to talk improving customer and employee experiences at the service desk with conversational AI with Bhavin Shah. Let's go ahead and introduce him. [A third speaker, Bhavin Shah, Moveworks, appears on screen below Christina and Christian in the center of the screen.] 00:02:45 - 00:03:06 Christian Bhavin Shah is the CEO of Moveworks. He is an entrepreneur with over 25 years of experience taking companies from inception to scale. In 2016, Bhavin co-founded Moveworks, the generative conversational AI platform that unifies all enterprise systems. Bhavin began his career in educational toys at Leapfrog Enterprises. Welcome, Bhavin. 00:03:07 - 00:03:09 Christina Bhavin, thanks for- 00:03:10 Bhavin Shah (03:10): Thank you, Christian. 00:03:11 - 00:03:12 Christina Oh, I'm sorry. Thanks so much for joining us today. 00:03:12 - 00:03:27 Bhavin Yeah, well thank you Christina and Christian for the introduction, and thanks for having me on the show. I'm excited to share what we've learned and really to come together on how the world is thinking about this new category. 00:03:28 - 00:03:58 Christina Awesome. Well, I was hoping we could start by just dispelling some of the myths around chat technology. People easily get confused about chatbots versus virtual agents versus conversational AI powered by LLMs like Moveworks. I'd really like to start by level setting the audience on the differences between conversational AI and how it's different and it's taken a technological leap forward with this new technology. 00:03:59 - 00:04:54 Bhavin Yeah, it's a great question. There's so many words coming out we almost need a new dictionary to help everyone stay on top of it, but I think the history is going back, call it 15 years to just pick a point in time, people would call this sort of interaction model, this sort of technology, a chatbot. Really, what the goal was is to try and be able to provide basic answers to questions. As time went on, these systems were starting to be referred to as virtual agents. Largely when you started to see people connect these chatbots to systems of record, different platforms that were inside the enterprise, being able to pull out information, be able to pull out data, maybe even run a workflow or two, that was less about the technology, it was more about the utility that it was providing. 00:04:55 - 00:05:40 Bhavin Then if you fast-forward to today, you're hearing the word copilot everywhere, and I think Microsoft decided to start renaming their solutions and products into this new framework, which they call copilots. The idea there is, again, it's an evolution of the technology, and I'll talk about the big leaps in a bit, but the copilot is really meant to not just answer questions, but also to do work. You started to see that in the virtual agent era as people were seeing how these types of products could add utility to employees and their workloads. But copilots are really supposed to do more than just answering questions, right? 00:05:41 Christina Yes. 00:05:41 - 00:06:13 Bhavin We, I guess, pejoratively refer to bots that just answer questions as chitchat bots. I think the reality is people actually want work to be done on their behalf. They want to get access to systems, they want something reset, they want to be added to a list, et cetera. They want to be taken to a training. All that comes into picture when you think about a copilot because it's supposed to be that essentially assistant that can do the work on your behalf. 00:06:14 - 00:06:23 Christina Yeah, I love it. So just that confusion around chatbots, technically their question answer. I love your chitchat bot analogy, which I thought- 00:06:24 Christian Yeah, that's great. 00:06:25 - 00:06:42 Christina ... was great. Now people expect immediate resolution of their challenges, especially from IT service desk. Moveworks is delivering, I read it was seven seconds to resolve common IT support issues. Can you tell us how you're doing that? 00:06:43 - 00:07:23 Bhavin Yeah, so I'll give you a little bit of a context in terms of how we're doing that, but I think the history is important as well. So if you think about this whole space of AI and chatbots is that people were looking at this as a way to provide a more natural interface, but ultimately, what we discovered early on was that it was actually more of a language problem that needed to be solved. People describe what they need using words that are familiar to them. They ask for things they describe a place that they're getting stuck, maybe something, some sort of setback, and then they're asking for some help. 00:07:24 - 00:08:17 Bhavin I think that where we went starting to investigate this space, there were a lot of solutions that had come out which were designed to address some of this, but their approach that they took was really based on essentially a set of predetermined scripts, predetermined dialogues. So the job of the person adopting this technology was to anticipate all the questions that people would have and then code that up into these systems with if/and/or/then/they're/else types of logic gates enabling the bot to follow a certain logical pattern. The challenge with this was, of course, it's very fragile, it's very brittle, and it doesn't quite fully follow the user in their thinking. 00:08:18 - 00:09:28 Bhavin Instead, the user has to almost anticipate what are the options within the bot and then how do I actually speak to it in its format to be able to get something done, and that's never satisfying. That's a very friction-full experience that have a lot of people abandoned over the years and didn't really see that level of engagement. We saw those technologies and said, "Hey, look, there's got to be a better way to do this, because if at the core of this we need to understand the language and then take these actions, what best technologies can be used to deliver that?" When we formed this company, one of my co-founders came from Google where he was doing a lot of core machine learning work there, neural retrieval, semantic search, and the concepts of large language models and such were emerging at that time. Of course, the paper from Google, Attention is All you Need, came out in 2017, and it really gave us all a new perspective on what these machine learning models, these neural nets, deep learning models could actually achieve. 00:09:29 - 00:10:24 Bhavin In applying that, we found quite quickly that they're very good at understanding intent. They're very good at extracting any of these. They're very good at highlighting, they're very good at summarization, they're very good at also generating text as well. It's all of those tasks put together that enabled us to take this and to really deliver on this idea and promise that if someone has a question, well, we can get them to the answer, we can get them to the automation in just a few seconds. So when you think about the average time to resolve an issue, a lot of the focus over the last 10 years was about better workflow systems, about better tools for agents, about portals for self-service. The reality is people don't actually want to learn new systems. What they want to do is just ask the question like we do Google and say, "Just give me the sports scores. Give me the weather.- 00:10:25 Christina Right. 00:10:26 - 00:11:11 Bhavin Give me the tracking information. You figure out what context I'm in right now and find me the right answer." So for us, a lot of this has been enabling these dynamic queries and conversations so that we then can figure out exactly what the underlying system is and produce that answer. In doing this approach with large language models, it's very robust. There is no hard coded flow and you can switch context. You can try and confuse it with lots of different words and words it's never seen before in that sequence, and it still can extract and handle the task quite robustly. 00:11:12 - 00:11:39 Christina You mentioned context switching, which I love, because when I'm demoing to clients and perspective clients, it's something that's really dazzling. Being able to start with one thought and go, "Wait a second, I actually mean this other thing," and it answering that question saying, "Well, do you mean this, actually?" Like, "Yes, I do. You're right. Thank you. I did mean that." So it fulfills that request and then being able to remember, "Do you still need this other thing that you previously asked for?" 00:11:40 - 00:12:15 Christina So being able to manage that and my typos and figuring out what I actually want regardless of how I ask for it, it's just been such a game changer. That's what I want people to bring home when they read natural language understanding with this product's virtual agent or that product's virtual agent, that it's just completely a different experience and it's something that users really expect these days 'cause we walk around with these cell phones in our pockets or these devices we're asking to play music. Our kids are growing up with them, right? So the expectations have just completely changed. 00:12:16 - 00:13:13 Bhavin Yeah, I think it's the guidedness that these models can actually provide to help you actually discover the right answers and find what you're looking for that is so powerful where you may start with what we call an underspecified query. You're not giving enough information. You're not giving the context, but the system can then either ask you to clarify or use what context it has to say, "Hey, look, given this kind of problem, there's three areas. Are you trying to get access to it? Do you need troubleshooting help or is it something else?" It's that that opens your eyes to, "Okay, yes, it's actually one of those. Now I'm going to further explain what I'm looking for." That's something that people have been wanting to see for a long time, and it was really the transformer model, the large language models that are now enabling that to really come to fruition. 00:13:14 - 00:14:05 Christian I think that that guided experience helps instill confidence too, because if I'm interacting with some kind of AI or a chatbot and I ask it a question and the first thing, my first interaction with it is, "I'm not really sure what that means," or, "I'm not sure what you're asking," right away, I'm going to start to lose face. So having that guided experience really does help the user, "Hey, this is actually working. I'm going to use this more and more," and it really helps with adoption. So I think that that guided experience is key. I have a question actually for both of you. So seven seconds resolution time is obviously something that I'm sure many folks that are watching this are interested in. What can we expect from an implementation standpoint? Are we talking months and months to expect time-to-value, or is this something that can get spun up pretty quickly? Why don't we speak a little bit about that? 00:14:06 - 00:14:37 Bhavin Yeah. Well, our goals from early days was to overcome some of the shortcomings that we had seen in the industry where people and CIOs would tell me, they're like, "I bought this tool, we upgraded," whatever might be the scenario, "and we try to build something and it's been a year or it's been six months and one, it's not finished, and two, no one's really engaged with it." I think we were like, "Well, gosh, this just seems to be a repeating theme that we kept hearing over and over again." 00:14:38 - 00:15:22 Bhavin So the way we architected our system was every time we can learn something in terms of a new utterance, and new types of use case that accrues to all of our customers, which means that everyone is benefiting from the use of the product every single day, that is a true network effect that gives our customers a decreasing time-to-value. Today, if you come under our platform, out of the gate, it's doing thousands of different use cases, use cases that are IT use cases, HR use cases, many others that really give it the immediacy in terms of value. It's not predicated on your own creativity. 00:15:23 - 00:16:20 Bhavin It's not predicated on you putting a lot of engineers to go figure out how to build this. It just works. Now what's important is even if you go and you do the mapping and you try and build all of this, it still won't be as robust because, again, we have seen so much data, we've trained our models over time at a level of precision and recall that makes this extremely powerful and useful right out of the gate. So I think the average median implementation time is roughly around 10 weeks to go live, and that's including UAT testing. That's including all the different steps that people like to go through to fully, thoroughly validate all the pieces of the puzzle, integrations and so on and so forth, but much, much faster than I think what our customers had seen in other technologies. 00:16:21 - 00:17:17 Christina Absolutely. When we were implementing, I was so impressed, I was so impressed, we went live with so many different use cases that I use regularly, simple requests, password resets, VPN connection issues, software orders, things like that, that just opening a ticket, asking for knowledge articles, "How do I change my picture in Workday?" Things like that where we can serve up the knowledge articles in multiple languages. Can I tell you how delightful UAT was? As an implementation partner, we were truly evil. I don't know if you're aware of this, but we kicked the tires so hard, we asked the same question over and over again using different words just so we could judge you, 'cause if we're going to recommend Moveworks to our clients, we need to be really certain that we are sure that we understand the technology and we feel really confident and comfortable with it. 00:17:18 - 00:18:06 Christina So heads up, we really messed around doing UAT, but we've had a fantastic experience thanks to that, and it was, it measured in really weeks to implement, which I wasn't sure when we started our adventure together if that was going to be something that we've worked to deliver on. I think a lot of it's depended on us. I remember hearing them echo and it's something that I worked with my clients today on is, "How long do you think it's going to take you to help us get these connections ready on your end, because we can move as quickly as you are ready to go." So putting the right folks in place to be part of the project team I think is really a critical success factor. But once that's ready to go, yeah, exactly, weeks instead of months, which is exciting. 00:18:07 - 00:18:31 Bhavin I'll just add a little bit to what you're saying, which is I think that's the brilliance of this technology. We do this all the time where even we'll be impressed with what the system understood about a question that we had or words that we use. So we actually have a way of going back into the training set to see what the model was trained on. Based on your basic understanding of this technology, it doesn't have to have seen it to understand it, right? 00:18:32 Christina Mm-hmm. 00:18:33 - 00:18:36 Bhavin That's not how this works. This is not a memorization technology, right? 00:18:37 Christina Right. 00:18:38 - 00:19:31 Bhavin It is absorbing the characteristics of language using different features, different cuts and slices. So we'll try things and we'll say, "Yeah, let's do it. Go for it," and we'll go back to the training set, I'm like, "Yeah, that was never in the training data, yet the model interacted or responded in exactly the way that you would expect it to, like a human being." So I think that it's just a powerful technology that even the folks like us that are on the creation side of this are continuously impressed with what it can do. We have an internal channel here at Moveworks called Simply Magic, and it's whenever we see something just mind-blowing that our product can do that is almost emergent in terms of its capabilities than what we were originally set out or thinking that it would be good at doing. 00:19:32 - 00:19:49 Christina That's awesome. Love it. Now, Christian, didn't you have a question about different domains for Moveworks? They're not built for just IT service desk. You had asked about HR questions. 00:19:50 Christian Yes, yes, yes. 00:19:51 - 00:20:02 Christina I remember when we first rolled Out Moveworks, you were asking about pay periods and asking about benefits and payroll and things like that. So I guess- 00:20:03 Christian Right, yeah. 00:20:04 - 00:20:14 Christina ... the question that a lot of our clients have, "Is this a product that's just available for IT Service desk?" I guess that's a question for Bhavin since- 00:20:15 - 00:20:40 Bhavin Yeah. Yeah, for sure. As you know, our copilot works across the enterprise, so we're integrated into many different systems, right? We're not just based on one vendor or one platform because the reality is enterprises have many vendors and many platforms on which they use to run their organization and business. Of course, as you mentioned, IT being an area that we began with because it's fairly universal. 00:20:41 Christina Yeah. 00:20:41 - 00:21:50 Bhavin We see healthcare companies, hospitals have the same issues, IT issues as a gaming company, as a media company, as a semiconductor company. So that is an incredible amount of leverage that we can bring in terms of providing people access to software and approvals, information and knowledge, being able to reset passwords to update them on tickets, so on and so forth. This really is something that provides utility across all disciplines in all departments day one. Ultimately the goal is to provide this so that your service agents can work on more higher level, more complex tasks, also bring back productivity to the employees so that what would've taken a few days to get access or a few hours, it happens right away so they can stay on task and stay on track. So then HR is another big domain of where organizations have invested heavily in building support teams themselves to really be able to make sure employees get unstuck and can keep moving. 00:21:51 - 00:22:34 Bhavin But HR is many things, right? It's benefits and PTO, it's learning and development, it's performance management and annual review cycles. It's sometimes related to payroll and bonuses depending on how it's organized, and there's just a plan play of different use cases, so we're doing that. We're helping people do things like role changes. Apparently, people are asking to change their roles and titles all the time to HR professionals and HR professionals are like, "I wish the manager and the teams could just do this automatically." Okay, well, maybe a conversational interface would be useful for that. Looking at PTO seems like an obvious use case. Yes, that's available. Finding out benefit information, being nudged to finish an open enrollment. 00:22:35 - 00:23:32 Bhavin Being nudged to finish a manager training because you just got promoted and you need to finish that within a certain time period are all types of use cases that are both reactive and proactive, but it doesn't stop there. I think when we started to roll this out, the next department became jealous saying, "Hey, well wait a minute. We want a tool like this, and we get 50 questions a week about X or Y, how can we have the bot service that?" Or, "We have people coming to us wanting to have us look up some information about inventory levels or this or that. Why can't you have the system connect to that system so we can get out of this busy work, manual repetitive tasks and do something more valuable like this migration we're working on and this other thing that we're doing?" So we have this capability with our creator studio for people to go beyond those two primary domains, which are the most mature, by the way. They have teams that understand how to service and how to track and how to ticketing systems and so forth. 00:23:33 - 00:24:41 Bhavin But once you go beyond that, it turns out that in addition to the value of having these prebuilt, customizable, configurable solutions, we want to tackle the long tail and the setup use cases that are now emerging faster than ever. I've talked to CIOs who are like, "Ever since ChatGPT came out, I went from having 13 ideas on my desk in October of last year to 400 ideas [inaudible 00:24:06] team saying, 'We want to use a chatbot for this, chatbot for that, chatbot for this other thing.'" So I think the ability to extend while being aware of these domains is very important. It has to be aware for the company itself, 'cause sometimes this topic is a topic of finance domain. Sometimes it's handled by a different team, maybe legal affairs, maybe something else. So the bot not only has to have smarts that are generic across enterprise, but it has to have smarts that can apply specifically to the policies and the nature of that particular company. 00:24:42 - 00:25:34 Christina Yeah. I just wrote a tech target article about increasing employee engagement, and one of my topics was using conversational AI. My example was you have a new employee start, it's the weekend, they have an emergency, they don't have their benefits information handy, they can go on their phone. Oh, I love this chat through Teams or Slack. I love the meet them where they are factor. I don't want to have to go out to some other system, some other website. I like that copilot experience and the fact that I can do approvals on my phone while I'm traveling. I can do password resets from my bed at 5:00 in the morning which reminds me that I need a password reset or I'm going to get locked out. All of those examples really increase productivity overall of your employees and improve employee engagement overall. 00:25:35 - 00:26:17 Christian That happened to me actually today where I got the IM again reminding me, "Hey, your password's going to expire tomorrow, make sure you change it," so that was really helpful. That leads me into my next question, actually. So one of the things we love to say at CAI is we like to drink our own champagne. So we implemented Moveworks in house, and one of my favorite features personally is the ability to request software and hardware directly through an IM, because going into some kind of ticketing service, having to create a ticket, having to wait to see the response to my ticket, it's just a noisy headache. So now it's as easy as sending an IM, and it's been wonderful. I've procured a couple if different pieces of software via that. So I wanted to ask Christina, what's your favorite feature of Moveworks so far? 00:26:18 - 00:26:48 Christina I have many obviously, but my favorite recently, I've been working with a lot of our global clients that have to communicate primarily in English with IT service desk and HR, but they have a global presence. So there's people entering tickets, asking questions from all around the world in many different languages. So the ability to translate knowledge articles, I can keep all my knowledge articles in English and it will translate on the fly. I was at a conference last week and I pulled up my phone and I said, "Speak to me in Spanish.: 00:26:49 - 00:27:19 Christina Then I said, "How do I change my photo in Workday?" And it translated into Spanish right there on the fly. Then I switched back to English 'cause my Spanish vocabulary isn't what I wish it was, but it really knocks people's socks off because that capability alone is just tremendous in helping everyone really within your organization, even if English isn't their first language or if you have a global presence. So that's my top today. 00:27:20 Christian That's awesome. 00:27:21 - 00:28:19 Bhavin Well, I'll add to that. I think the topic of language translation is a big one. It really became front and center for us during COVID, one where companies recognized very quickly with instead of having five locations you have 5,000 locations with employees working from home, but more than that, they had to find ways for them to scale their operations more efficiently. They also had issues with agents resigning and the great resignation, the great reshuffle, and they started to hit these dependencies 'cause they had two people who were trilingual and can do these certain types of support calls, and then when one of them leaves or two of them leave, now you've got a major hole and a major gap. So language translation became front and center for us back then. There were a lot of technologies, lots of bot tools that were out there saying, "Yeah, we support 100+ languages." 00:28:20 - 00:29:00 Bhavin But really what they were giving you is a table that you have to put the English words and the French words and the Spanish words, and it was a very... I am surprised it didn't worked at all, but it was a very simplistic view of how to think about language where instead we said, we went straight to the large language model architecture and we say, "Hey, look, there are some pretty good models out there that do amazing work, that are lightweight, that are fast to run like Meta's M2M. What if we were to fine tune that model with our enriched enterprise data," which we collect and aggregate on a regular basis, "and then how can we apply that to provide real-time language translation in and out of our system?" This solved big issues. 00:29:01 - 00:29:42 Bhavin To your point, we have large customers, 60,000, 100,000 employees who are like, "Hey, we've over time consolidated our service desk to an English space desk who can now support Japanese users, Polish users, Italian users, because your system provides that interface." So they might respond to a ticket in English, the bot reaches out in Italian, the person responds in Italian. That comment gets added back to the ticket if it's going through a ticketing system in English, and it goes back and forth. At the same time, there's articles out there that might be sitting in English that need to be presented to the Italian user in Italian, and we do that on the fly. 00:29:43 - 00:30:44 Bhavin So what we've seen is just a tremendous lift in the business impact, in the business value because we can almost take language off the table. Now, one last thing I'll say is why language is hard, because we've had Google Translate, we've had these other services is because in this context, it has to be enterprise aware. It has to understand your enterprise graph because you will have words like power space point that shouldn't be translated into French because they still refer to it as PowerPoint. When you do that with a basic language translation model, a lot of things get goofed up. A lot of keywords and things that shouldn't be interpreted get interpreted. So we work in this a mixed world where some words people are using in English, but they're speaking fluently in another language, and your system has to know when and where to provide that translation. 00:30:45 Christina Right. 00:30:45 - 00:30:54 Bhavin I think that's something that we perfect every day, we're thinking about, we're iterating on, and our customers are getting those updates and benefiting from that. 00:30:55 Christina Fantastic. 00:30:56 - 00:31:06 Christian That's awesome. Unfortunately, our time for today is over. It's been an absolute pleasure chatting with all of you, especially Bhavin. I would like to also thank our audience for your attention and participation. 00:31:07 - 00:31:35 Christina We hope you'll take away these three key lessons for today. Moveworks and conversational AI isn't just for the IT service desk, they can help you fulfill requests in seven seconds or less. It's great for translation. This is a huge benefit of conversational AI in the marketplace, and the speed to market is measured in weeks instead of months, which makes it a really compelling solution for IT and other domains. 00:31:36 - 00:31:57 Christian Later, we will be sending everyone that attended a recording of this event to share with your colleagues. In the meantime, if we didn't get to your question and you're interested in learning more about CAI Intelligent Automation Solutions or someone that is, please visit our website at cai.io and fill out our contact form, or you can even contact one of our team members via LinkedIn. Thank you, and have a great rest of your day. 00:31:58 - 00:32:04 Christina Yes, thanks everyone for joining. Stay tuned for more details about our next event in the learning series coming soon. Thanks, Bhavin. Thanks, Christian. 00:32:05 Christian Thank you, Bhavin. 00:32:06 Bhavin Thanks, Christian. 00:32:07 Christian Thanks, Christina. 00:32:08 Bhavin Thanks, Christina. [Closing slide 1. Blue CAI company logo with tagline “We power the possible” appears in middle of screen. Company website www.cai.io appears at the bottom center of the screen]? 


Skip past transcript

Let's talk!

Interested in learning more? We'd love to connect and discuss the impact CAI could have on your organization.

All fields marked with * are required.

Please correct all errors below.
Please agree to our terms and conditions to continue.

For information about our collection and use of your personal information, our privacy and security practices and your data protection rights, please see our privacy policy and corresponding cookie policy.