Podcasts

Transcript

Pete: Welcome to ”Shared Insights,” the podcast from BA Insight. My name is Pete Wright and I am joined today once again by the long overdue BA Insight Solution Consultant and Project Lead, Sanjaya Paudel. Welcome back, Sanjaya.

Sanjaya: Thank you, Pete. Great to be back and thanks for having me.

Pete: Last time you were here, you gave us this fantastic sort of dissertation of a big customer implementation and that was back in episode 50. It was a great and illustrative conversation on how to do this. And today, you’re back to give us an intro into AI and machine learning. If you’ve ever found yourself struggling to determine how best to integrate these solutions into your enterprise search efforts and more, what solutions are coming from the big search organizations, this is the week for you. So, Sanjaya, how would you like to begin this conversation? I’m thinking maybe you could give us a little bit of an intro on how these technologies are being used today. We’ve all heard the language. How are they being used? And then we can get into how you approach this conversation on implementation into enterprise search.

Sanjaya: We can see AI and machine learning is everywhere. We’re using it in our day-to-day lives, our decision-making, and now with the availability of many AI-based services, even for unstructured data, companies are at a crossroads where they must make a choice and they want to know when and how to bring AI-related services, cognitive services to their enterprise search. If you look at it, in social networking or in your email, there is artificial intelligence playing a role. There’s a data mining going on, for example, to filter out spams or helping you tackle complexities of transportation. You know, as you are traveling from one city to another city, what’s the fastest route to get from place one to place two. The AI, working behind the scene and helping you with everything. So I think it’s time that we talk about it and we, you know, make up strategy for enterprise search.

Pete: I think one of the things that we have when we have our consumer hats on, we have all these sort of data brokers learning about us through our behavior and our activity online. And sometimes, you know, you hear these terms are associated with potentially negative connotations around, you know, social behavior tracking. But what we’re talking about is using these tools to really unlock your internal, your enterprise search and figure out how to make enterprise search work using some of these more advanced tools, right? How do you make life easier for your users?

Sanjaya: Yeah. There is so many things we can do and artificial intelligence can do for an enterprise search. So before we get there, I think this is one thing we need to understand is language, you know. Language is at the center of AI, right? Or let’s say language is the holy grail of AI, you know. And it is more so for enterprise search, because enterprise search is all about content, produced and written by human. So there is a language, you know. So language is the fundamental aspect of our behavior, right? So for AI to imitate human, it’s very important that we have a computational model for natural language processing. And that is what cognitive science is about.

And in the last two or three years, we have seen great NLP models from Microsoft, Google, even Amazon, which are going to revolutionalize the way computers are used and we can build a great platform using AI and use that in enterprise search. And there are different things we can do and, specifically, there are two types of applications that people can build using AI for enterprise search. One is text-based applications and another one is dialogue-based application, where you can use a smart assistant. And we can talk about each as we go.

Pete: Yeah. I think that would be very useful. I mean, you know, in terms of the transparency of interaction model with both of these sort of applications, I think it’s really important. Then let’s dig into that. Where do you look at the opportunities that come with implementing AI and machine learning into the enterprise search model?

Sanjaya: The thing about enterprise search is we have been in enterprise search market for, you know, 10 to 15 years now and we have understood what works and what doesn’t work, right? And if you look at it, the most successful enterprise applications have three things: metadata rich content, and actively-engaged search users and search administrators, because you have to have this process of giving feedback and improving over time. And then the third thing you must have is search analytics, which gives you an insight of how your search application is used on a daily basis, what’s working, what’s not working, how people are using it, you know, are they coming back for more, things like that.

In all of these three areas, now we can apply these AI services we have out there and improve in all three areas. And in terms of opportunities, there’s lots we can do here. When we are working with customers, the one question we always get is, “Why can’t I ask my enterprise search a question like I do with Google, Alexa or Cortana,” right? Now, we have opportunity to build that kind of application in our dialog-based smart assistant using artificial intelligence.

Pete: Just how practical is this right now? Like Monday morning, are you actively implementing these kinds of things right now? Are there people in enterprise, you know, client relationships that are talking to their, you know, Cortana and able to get search results now? Or is this still something that we’re looking at on the near horizon?

Sanjaya: So in enterprise search, not so much, but outside, yes, people are using it. You know, we’re using Google, Alexa, and other applications, every day almost, right, for different kinds of things. But coming to enterprise search, people don’t have it yet because they have not built a strategy or they are still exploring, right? And they have some challenges which we can’t get to. And that’s why it is very important to, you know, understand what AI can do and also they need to understand what’s not working for them. You know, before you bring AI to enterprise search, you really have to understand what’s not working, you know, and you have to get a real insight of what’s not working. And it’s very important to have search analytics provide that information to you so you can improve and you can fix the areas where AI can help.

Pete: Let’s take a step back then and talk about the three things that you mentioned. How do you see these AI tools applying to metadata-rich content? What are the opportunities to help us, you know, harvest better metadata around our content?

Sanjaya: Like I said before, with AI, there are two types of applications we can build, right? There’s text-based applications and dialogue-based applications. So to have text-based applications, we have to scan the content, analyze the content these enterprises have. And there are terabytes of unstructured data, you know, and legacy data that can be scanned or analyzed with this AI NLP services. I like to call them NLP services. NLP, which is Natural Language Processing, is an application of AI. So using NLP services, you know, we can extract key contents. We can extract key concepts and we can generate summary, which is like, you know, two to three lines or a two to three-paragraph summary of 100 pages document, and it exactly tells you what that document is about. You don’t have to go through the whole thing. And we can extract the languages. We can extract categories, you know, what are the documents about. So by extracting that information and labeling those documents upfront, now you have the opportunity to use that information on the UI side to read on the content the users are looking for.

Pete: How good is it, in your experience and you’re testing, how good are these summaries that are generated by the bots, for example?

Sanjaya: Really good. We find a document somebody’s generated by both Microsoft and Google AI Services, NLP services, very accurate. They accurately represent what the documents are about. So instead of reading the whole document, you can just read these four or five line-summary and understand what the documents are about. They are very accurate.

Pete: Okay. Next big opportunity is around bot conversations. Can you tell us what a bot conversation is in this context and how it’s gonna help you?

Sanjaya: So in that area, using artificial intelligence, we can build a question answering system. And one of the areas that’s really important to have a successful enterprise search application is actively-engaged users and administrators. You know, a way or process where users can provide feedback and for search administrators to know what’s working, what’s not working, right? Sometimes you go to a search application, you run a search and you don’t find anything. Or you’d like to ask a question, right, and you don’t get anything back or you don’t get any results back. So at that point, if you have a system or a smart assistant that can record your question, right, and search administrators can look at those questions and resolve them. And then when you are online again and when you’re accessing the application again, it reminds you about your questions you asked and then it tells you that, “Hey, you have an answer for your question,” right?

That improves the whole communication there and it’s kind of bridging the gap between search users and administrators. And that goes a long way because only through feedback system and this question answering system, you can know for sure that the application is, you know, whether helping users or not. And that’s where this bot conversations will help a lot.

Pete: Let’s talk about personalization. How does AI, how’s it going to help you personalize results for me, your user?

Sanjaya: That’s a great question. When it comes to personalization, historically, we have been able to return results based on who you are, your job title, department. And with NLP-generated rich metadata tags, we would be able to return documents that have your job title, department, interests as key concepts or as part of the document summary. And using confidence and sentiment score of concepts provides even further opportunity to surface more relevant content on top of the search results. We would also be able to connect this with user search behavior by mining analytics data and make results even more relevant. So we have so many possibilities around this.

Pete: You know, one of the things that we’ve talked about before on this show is how we’re able to embrace and extend text by applying other analytical tools for, you know, images, videos, PDFs, you know, taking the data beyond just text into…and sort of broadening our horizons. What’s the state of those sorts of tools and how they’re working with our AI image processing?

Sanjaya: One problem area for enterprise search has been poor or no metadata for images and video. All we had was file name for metadata. And now NLP services that do image and video processing will not only label or tag when and where these images or videos were taken, but also label them what these are about. Label them with all sorts of tags about the objects and people in the images and video, even what people are doing in those images and video. So enterprises will be able to quickly search and find the right images and video needed to do their job without having to click through hundreds of images and videos in company network drives. We can imagine how useful this would be for organizations primarily working with images and video every day.

Pete: I imagine that applies to one of our favorite conversations, expertise location, right, one of our key business tools, being able to find who does what when you need them to help you do it. Now, we’ve always been able to just search, you know, to a well-tagged LinkedIn page, but now we’re able, I imagine, to see work product tied to individuals and have it be more reliable than ever before. Am I thinking along the correct lines there?

Sanjaya: Sure. In terms of expertise locator, finding experts for a new project you’re taking on or, you know, you’d like to discuss a subject matter expert with someone, that’s a very common practice in enterprise. So with AI-based services, I see a possibility of building AI-enabled text-based applications that would return subject matter expert based on their experience, their work, key concepts related to their work, time billed, etc. The idea here is to find a true subject matter expert based on NLP-based knowledge mining and not based on his or her user profile. Also, like you asked, we could also find documents containing experts’ images, you know, a picture of their face and label them as such and use that information to determine who has the most subject matter expertise. Google Vision AI, Microsoft image processing and Amazon Rekognition AI will help tag documents containing experts’ images. We can build a really useful AI-based expertise locator for the enterprises.

Pete: That’s fantastic. And it’s amazing. And we have to reiterate that this is the result not of necessarily or specifically human data entry, right? We’re not just culling what people have entered about their work. This is the machine doing the work for you. And you are obviously bullish on these, you know, these tools and our AI future. But, you know, as we discussed earlier, we’re not there yet. Can you talk a little bit about some of the challenges that you’re running into and in terms of getting some big internal enterprise search projects launched and running and actually providing great data?

Sanjaya: I think there are far more opportunities than challenges, but then every new technology or opportunity certainly has its challenges. AI for enterprise search does it, too. We love challenges, don’t we? I think the first and foremost would be getting the right kind of education on AI, and sometimes people can easily get carried away with buzz around it. Most organizations are used to traditional way of searching and bringing AI-enabled, dialogue-based searching which require education, and in some cases may need to acquire new talent based on your goals, what you’re trying to achieve. We have also heard negative connotation like AI is dangerous, can be a threat to human existence, etc., which is far from truth and non-applicable even for artificial general intelligence. And AI for enterprise search is a special purpose AI to do certain tasks better than humans do, improving productivity.

Another challenge is some organizations do not know their current problems and challenges with their current enterprise search. And because they don’t have analytics data, they don’t know what AI is going to fix for them. So when you know what you are going to solve with enterprise search, it’s easier to plan your strategy. So having a tool that tells you what your problems and challenges are is very helpful. There is a notion that you have to learn a completely new application, build your own AI model for NLP, etc. That’s not true. You have very sophisticated pre-trained models available and they are available for free trials. Organizations can make use of that, test their sample data with those, understand what it does and what it can’t do. And in some cases, you can even bring AI-enabled in-app search to your existing applications and start on your learning path. AI-enabled search improves over time, although there are some challenges. Once you get behind it, you will start seeing the value out of it.

Pete: All of those issues notwithstanding, can you define what it looks like to be an organization who is ready for this? What are some characteristics of an organization that might say, “You know what? I think it’s time to start going through due diligence and figure out how we can expand our enterprise search tools to do this?” Would you look for a particular size of their document set? You know, distributed operations? Like, what does an organization look like that’s ready?

Sanjaya: That’s a great question. I’m sure most organizations will have these questions as to when and how they will be ready. There are no set of standards that makes you ready or not. I would say organizations that have identified problems with their current enterprise search, have tons of images and videos, generate tons of unstructured data every day should look at bringing AI-enabled search with natural language processing and a smart assistant to their enterprise search. Also, organizations that have fair understanding of AI-related offerings, cost, organizations that are looking to improve communication between search users and administrators, those are the organizations that are ready. It’s all about understanding the value these NLP services can bring to your organization and how they can help with your productivity. And because AI-based search will also have learning and training part and which improves over time, sooner you start, the better it is.

Pete: You are, as we said, you’re excited about the potential of these tools. But what does the AI future look like in terms of enterprise search? What’s next for you? You know, what’s just over the horizon that really has you excited?

Sanjaya: When it comes to future of AI and future of AI in enterprise search, I think it will be even more exciting. I already find it very exciting to be in enterprise search with so many AI-based services available there and the kind of opportunities we have. As we go, I think we will uncover more exciting possibilities. One thing we must understand is that AI-based systems improve over time. Therefore, getting behind AI-enabled enterprise search sooner will help organizations stay ahead of the curve. In near future, I think we would be able to develop smart assistant, even smarter, to anticipate your needs at a given time such as suggesting results for documents based on your upcoming meetings. So I’m very excited about the future.

Pete: Yeah. When you talk about like the summary services, being able to summarize not just documents but, you know, documents related to the day ahead of you.

Sanjaya: Exactly. We can bring AI-enabled search to your existing application such as ServiceNow or CRM. So right from where you are, right from the applications that you have been using, you can take advantage of these AI-enabled search and get the most relevant content for your, you know, day-to-day work.

Pete: Well, that leads to the last question I wanted to get your take on, which is where is BA Insight on this path in terms of helping customers? It sounds like you are actively…you’re working with customers on implementing these tools now.

Sanjaya: Yes, we are. So we had been working on AI and AI-related stuff for a couple of years now and we already have some customers using it and some customers evaluating it. And it’s going great. And, you know, sometimes we run into the customers when we, you know, deploy our tools, for example our smart assistant. They don’t always like it, right? Because people are not used to talking to bots or interact with bots. They find it, you know…it’s very overwhelming, you know. Like, “What do I ask?” you know, and, “Why is this thing popping up here? I’d rather talk to a real human,” things like that. But they need to understand that the smart assistant there has so many tools at his disposal than a human, you know, could ever have.

And one experience we had was, so we deployed our AI tool and customer said, ”Hey, I don’t like it. I wanna disable it,” you know. And what we said was, ”Please give it a week or so and then try and identify what works for you, what doesn’t work for you, what you don’t like about this system and what you like about it.” And then the customer went on for another two or three weeks and came back and said, ”Hey, I want to keep it.” And then was able to suggest some of the changes and we were able to make those changes. And then they’re using it and they love it, right? And it’s really helping them. So my point there is we run into all kinds of customers, you know, who are ready, who are not ready, and, “We are still thinking about it.” My suggestion is AI is the future. It’s here. We’re seeing it in our day to day lives and it’s time to, you know, make a strategy and make use of these tools for their enterprise search as well and take your enterprise search to the next level, basically.

Pete: You know, just in the spirit of not being left behind in terms of productivity and efficiency and giving your users a best in class search experience, the sooner you let the tools loose on your data, the sooner you will start being able to reap those benefits. Where would you suggest, that those who are interested in learning more about what BA Insight is doing with regard to AI and machine learning, where would you suggest they go for more information?

Sanjaya: I will say the best place would be to start with our site, bainsight.com. And also, you know, read online about some of these offerings from these great companies and what they can do. And that will give an idea of what’s already available and, you know, what’s coming. And at that point, see how that fits into your AI requirements and how you can integrate that into your enterprise search applications and your existing application, or you can even start with a new one. And sometimes, you can always do an AI-based application as a separate application and have some of the users start using it while you continue using your traditional applications. And once you have figured everything out and make a switch, that’s an easy way to go.

Pete: Links in the show notes, everybody. Thank you so much for downloading and listening to this show. And thank you, Sanjaya Paudel. It’s been too long. I hope you don’t keep us waiting so long before you come back next time.

Sanjaya: Sure, I like to come here. So thank you for having me.

Pete: Thanks everybody. On behalf of Sanjaya and the entire BA Insight team, I’m Pete Wright and we’ll catch you next time right here on ”Shared Insights,” the podcast from BA Insight.