AI-ready content: Governance, integration, outcomes featuring Deloitte

Key Takeaways

Hi, my name's Jenet Gessler from Box. I'm the vice president of Box Consulting go-to-market and partnerships. Today we're going to have a lovely conversation about data readiness in getting the AI outcomes that you'd like in the enterprise. And I'm joined today by our lovely panel. Could you please introduce yourself? >> Sure, Jenet. Mike Carlino from Deloitte. I'm a leader within our AI and data practice um focused on unstructured information. Hi, I'm Shivani Mayersky. Also from Deloitte Digital. I am a sector leader for our Salesforce practice in the US leading retail consumer products and travel and hospitality. >> And I'm I'm Nick Reed from Box. I'm a partner architect with our partnerships team. Well, thank you for joining me today. So we've learned a lot over the past couple of years about data in the enterprise and for customers to get the AI outcomes that they're looking for. We know that data is not clean. We know that it's not tagged properly, governed properly, there are proper policies in place. Mike, could I start with you? Could you help us understand what customers need to do to prepare for data readiness? >> Yeah, absolutely. It's definitely a a big area. We we we do call it data readiness. Um when you think of AI though, you think of all the the shiny shiny balls out there that it can it can just work, but it doesn't. There has to be a foundation. And the foundation includes both structured and unstructured information. A lot of the time when we we look at our POCs, it's just focused on a small subset of information. So you have success in those POCs. Making that crossover to production and a scalable solution, that's when you have to think about, okay, the whole corpus of knowledge within your enterprise. That's when it gets tricky. Mhm. So our approach is really maybe four or fivefold I'll say. But one is around just definitely the quality of data. And when you think of the quality of data, goes back to some concepts maybe 10, 20 years ago, really. I mean, when we think of garbage in, garbage out, GIGO. I don't know if you remember that. Um probably all of us do. Um And and that's so real today. I mean, it's if AI is going to be reading this information, it it has to be quality information and then on on a large scale, on a scalable platform. So we also have another term that we used to use a long time ago, which was a ROT. ROT ROT stands for redundant, obsolete, trivial. Yes. And when you think of today, people use it as knowledge rot, that they they coined the term. It's it's the same thing. And when we talk about knowledge rot, and you think of unstructured information, it's a little bit different than structured. The way that it exists, it has a whole life to its own. If you think of one document and if you are in an enterprise and you you want to send that document to a whole bunch of people, well, you could do it a couple ways. One is to create a link to the source document, which is the way you're supposed to do it. >> Mhm. But quite often that doesn't happen, right? You do an attachment and it goes out to maybe 10 people, 100 people, 1,000 people. Um that's that's a problem. >> floor edited. And floor edited and then he's going to make the changes. Um that's when you get into issues. Now AI can look at it and say, okay, well, maybe the version that is the most out there within the enterprise, that's the right version. Mhm. Is it? Don't know. Maybe it's the records the one that's being held by the records manager or legal has the real version or somebody in the business has a real version. That's where there has to be some governance around it and that's where we get into understanding what's out there. You know, rot within the enterprise represents 40 to 50% of all information. Wow. That's a lot, right? And if we think of unstructured information within the enterprise is 80% of all information, then you're getting into like a really large amount of information that has to be sifted out. And and then when you do that, then you're left with a good piece of workable information. But then we get into the other aspects of it. It's not just about cleaning it up, right? It's about making sure that the people that have access to that content should have and the people that shouldn't have shouldn't. So when you're getting into Agentyc, that's can become a a like what what's Agentyc doing with it? Are are they grabbing the stuff they are allowed to? Are they grabbing people's salaries? Are they grabbing personal private information? Um all of that needs to be considered. And when you unravel the onion if you will or take away the layers, >> Mhm. you approach it from that perspective, from a governance perspective, perhaps there's legal holds on this information. Maybe there's records management disposition rules around it. Um we we get into the whole aspect of that. And then we talked about, well, where does this information go from an integration perspective and interoperability perspective and how does it get matched up with structured information or or other unstructured information within a knowledge graph or or now we're talking about graphs for uh graph rags. So so having a common ontology that both structured and unstructured can talk to, that becomes another layer of cleaning up our foundation. So there's a lot to it. We've only scratched the surface most of our our our our clients. >> So you're saying that customers have to spend a lot of time in cleaning up and understanding what data they have and what they're doing with it and how it's being used and what's not needed and what may be trapped prior to making any progress on you know, their actual outcomes of what they're trying to do with it. >> Yeah, and and it's not just a one-time snapshot either. These documents have a living breathing aspect to them. They have a life cycle. So someone might be using it one way, one version and then all of a sudden through the life cycle it's being used another way from a few different groups. All of that has to be considered. So there's a lot to it. Very interesting. I did read the state of AI readiness which was just released from Deloitte in January. Super interesting. And I was reading about how most of the AI activity is centered around cost savings and not yet has not yet moved into revenue revenue generation. Shivani, hoping that you could maybe elaborate on what some customers are doing to get data integrated maybe into their, you know, other platforms like CRM or, you know, and any types of use cases that you could bring that you've seen customers make that transition. Yeah, I think we have a really great opportunity right now to leverage unstructured data in a way that we've never been able to do in the past. But with with that opportunity, I think there's also a great responsibility. So embedding governance is really important into those content workflows so that it's being tracked at every single point to to make sure that the right people are looking at the right data. It's not falling into the wrong hands. >> Yeah. I will say if you were integrating with a platform like Salesforce for instance, you know, that is a CRM that's based on relationships. Now you can access data to sort of arm those agents and those reps with real-time data so they can feel more empowered walking into a situation that they have the latest and greatest. So I think a huge opportunity there. Embedding governance and data quality is a huge piece as Mike mentioned as well as, you know, your output is only going to be good as good as what you put into it and and and making sure that your data is clean and structured from that perspective is really really important. Um I think there's opportunities more than cost savings. It's it's enabling, you know, faster deal closures and things like improved client satisfaction based on the fact that they can access the data that they have and they don't have to look for it in multiple places and that makes everybody go faster and be more efficient and you're delighting your clients just in a different way. Mhm. Mhm. And I mean, we really see it at Box as well internally like I mean, this lovely peanut butter and chocolate combo of unstructured data with structured data. Is it the PO information that I just quite frankly when we talk about our customers interacting with these things, is your job to copy paste information out of one form of data and turn it into a tabular format? Like all of these things can be just powered up to save everyone time to deliver better customer outcomes. How those teams can then get back to the work that they're actually supposed to be doing. The big thing that we'll see is these people on their crawl, walk, run, fly journeys at Box. Like when we kind of think about how customers want to come in for their first type of thing, what what's the adoption curve look like for you guys when you guys see it at Deloitte? Yeah, I'm glad you brought that up. I think the myth is if you build it, they will come and it doesn't work that good. Integrations can be disruptive. If it's not seamless and adding value right away, you can lose your audience. So I think adoption is extremely important to have that plan and that change management sort of piece figured out before you go into it. And creating awareness and creating excitement around what's coming, I think is the key part of that. So that whole communication plan goes along with that. I also think that a phased approach is really important. So when you're in discovery, you do that phase zero, you're looking at what are the risks that I'm going into? What is my change management plan? What are the three top use cases that I want to focus on that will bring a high impact? And then the next phase is sort of building out your pipelines, your integrations and ensuring that, you know, you're focused on those top use cases so everyone is very focused. And then get real-time data. Get get a feedback loop going so that you can figure out what's working, what's not, refine it as you go into sort of a pilot, not just a POC, but a true pilot where people can give you insight and then you go into production. So that that is risk adverse. It mitigates the risk to the point where at every phase you can look at it and say, is this really what I want to do or should we pivot and look at a different use case that's more relevant. Sounds so much more interesting than just doing another AI training. And it feels like I mean, again, it's all about tangibles at the end of the day. It's like getting your hands on with these really awesome new tools. Yeah. I I'd also add though you have to do it right and the risk mitigation is huge, right? So, planning out that risk and making sure that you've mitigated all of it. If If you have a bad implementation your first time, second time, third time in production and you've scaled up, you can do a lot of damage and that's not the way to start these programs. Mhm. Totally. Well, that leads me to something we've seen a lot in the market and that's just moving customers out of a pilot phase. So, there there seem to be a lot of interest in we have to do AI. And everybody's kind of trying everything out which is a compliance issue and governance issue in itself, but now there's like some fatigue both on the user adoption side, but also on just being ready to move from a POC out to a pilot or out to production. What kind of guidance would you would you give or that something you've seen to kind of move you know, be ready to to be to production. Going from POC to pilot to to production. I mean what we've seen is there's just a lot of POCs out there. We've we've spent the last 2 years doing a lot. Creating those those shiny objects. Wow, isn't this cool? Right. >> Well, doing it in an isolated way. I'm getting back to the data question too, right? You might get bored of me saying but the foundation of data is so important. Well, that's true. It's so important. So, like you can like I said, you can do a lot of damage if you take something from a POC just leveraging a data set that's really small and now scaling it up to production and relying on data which is literally garbage. You can run into a lot of problems. >> Yeah. So, so making sure that whatever use case you're starting with make sure that the relating content, the unstructured, the structured information that ties to that is clean and it's managed and it's governed. So, maybe that's the way that the best way to do it is just do it step by step, use case by use case, knowledge graph by knowledge graph, whatever you want to call it. But making sure that whatever it's talking to and relying on is reliable. I think that's that's really important. >> Mhm. I think in the beginning companies would take AI and say I'm already behind. What do I do? I just need to check the box. Now, I think companies are more focused on the strategy behind it. What is the true outcome you're trying to to get out of this? What is the ROI that you're looking for to really sell the case upstream? And so, if you build a POC as an accelerator that's already, you know, you're you're sort of hitting some of those stepping stones along the way, it's easier to get to pilot or production that way. >> I mean, it's such a common thing that we see in our customer base when I talk to our ecosystem of our customers, the the partners they work with, everyone just seems to be demoing for the board room. Like it's not one of these things that is grounded in practicality and I've loved working with Deloitte because it's just it it's bottom up where the a lot of these integrations and implementations start is how do we save 15 minutes on getting a PO through a process faster, legal contract reviews. I mean, Mike, I know you've worked for years in the unstructured and structured data worlds. Like do you guys have any places where you've seen this like go really right right now just out in the world? >> Yeah, and and you know, if you think of just the different places that we touch within both sectors and within each sector the businesses, you have inbound content, you have outbound content and then there's the stuff in between. If you break it out though from the inbound per part of it, the capture piece when you think of things like within insurance which is regulated, right? But there's there's a lot of back and forth with claims, policies and creating policies and that's a big area. Then when you get into banking, if you're going to do loan origination or if you're going to do if if you're going to do know your customer, things like that, that's a big area too. So, on that side if you look at like what's working really well, the the low hanging fruit it would be the stuff that again, getting back to the data and I apologize. But but it's it's very controlled. It's very controlled data. You know you're going to be getting information from your customers, from your clients and you're going to be looking at it. You know what to expect, what doc types to expect. And lever using the technology, you can say okay, based on these doc types I'm going to be looking for this information within the the doc types. You extract the information using the tools and then you automatically ingest it within these big systems. We're seeing that work the easiest, the best and we're seeing big savings like and and if I give a savings it might be like, oh, for us where we're expecting 20% or 50%. It's significant. I'll just say that because it depends on each client. But we're seeing significant savings. On the other side of things, we're seeing for example, oil and gas for example or anyone who's using engineering or assets or manufacturing, that kind of thing. You know, how how can we leverage all the documentation around maintaining an asset and and making sure that the life of the asset is being predictably managed so that you know when to have maintenance on it. You know that the life of the asset can be extended by that way. And we're seeing a lot of success there, too. And what is involved there though is a lot more combining the for your question about unstructured and structured, bringing it together. Having a common ontology across the board for both structured and unstructured so the two can talk together. Mhm. And so that when we're talking about all of the data around these pieces of assets within the manufacturing floor, engineering that you have both the structured data and then you have the documentation around it and everything else, the records, reports on the unstructured, bringing it all together. One common I'll call it knowledge graph, whatever you want to call it. Mhm. And making sure that the two talk together and pulling all that information, now you have 100% of the information. Not just 20% of the information and we've seen really strong success there, too. Especially when I talk to we both were kind of in the Salesforce ecosystem together for a while, but it's one of those things that when when I talk to any CRO or anyone that cares about the structured data platforms or what their teams are doing they're not hiring W-2 employees to go push paper around and move these things around. They're hiring them to have great relationships, deliver amazing customer success. So, in that world of having the right data to the right people at the right times, I mean, it shows up nowhere better than our Salesforce pieces and like how those systems integrate together. But I'm I'm interested of like where this starts to go in a like what the first types of adoptions look like and I think that's something that's really interesting to us is like we've gone through the agent hype cycles. Everyone had an agent. Everyone got agent buzzword overload and getting really back to the practical pieces of how do I just make myself a better service to my customer? How do I make sure that we're delivering better outcomes in our oil and gas processes? How do we make sure that our manufacturing times are hit with the right levels of compliance. All of that is so so important and I I think as we kind of come off the the highs of again super hype cycle buzzword fever paid by the time you say AI on a on a webinar, a call, anything like that we're really getting into how does the process actually work today? Mhm. Yeah, I think [clears throat] integration is a key piece of this, right? It's how you integrate. So, with a platform like Salesforce you don't want to embed the data directly. You want to ensure you're using a connector or an API. Which is going to be more secure and it's going to leverage all the governance that you're putting in place to keep it safe. The other piece is I think as important as the right data is, it's also the wrong data. It's being able to filter out the wrong data so that you're completely relevant. You're not coming to the table with something that happened 2 years ago. It's everything is moving so fast right now. Having that again at the at your fingertips is really really important to to make sure that you know, you have the great latest and greatest in front of you. Mhm. And then I mean, another thing that comes to mind is just I mean, Aaron Levie, our CEO, talks a lot about just trapped data. It's just you don't even know it's there and how to use it. Has that come up for you at all in in bringing that into the knowledge graph as you say? Yeah, it really it has and it it's exciting to see because we in the past it was a black box really. It was it was nothing that you could access. It was all in there and how do you how do you know what's in there? And now now we're saying, hey let's create some insights about this stuff and then get some real value out of it. We've always called it IP, but we've never been able to put value to it. Now we can. Yeah. And that's that's exciting and we actually had a client who was crying with joy because she found something that she had been searching for for I think it was like 2 months. And then we we found it within a day or well, actually minutes once the Amazing. system was put in place. I can't actually say that that that happens that often. You know, the a client of joy of with tears of joy, but that did and it made me it made my day. So, that was great. >> You got to be sassy. It's changing lives. That's your Get that client on stage. That's exactly the story. >> [laughter] >> It's like, yes, that's my goal. Our customers cry with joy. So, as much as AI can bring to the table, that human empathy is really important. That's what everyone resonates with I think and that's what really will will build adoption. And when you see the results emotionally on someone, especially on a stage I think it's there's nothing that's more compelling than that. Yeah, and then you get out of the oh my god, I'm going to lose my job day I to like I can do my job so much better and so much more efficiently and get better outcomes. And that's I think where where we're heading. But I think you always need humans, right? So, that's the balance. Let AI do what AI does best and then let humans bring the bring the emotions to the table. True. I mean, it it so true and it's one of those things like I talk about it as like you're giving pedal assist or like if you've ever ridden an e-bike, you're giving pedal assist to everyone like suddenly you're like, "Wow, I've got my Lance Armstrong legs. I can really go so much faster." I mean, we we talk and the funny thing is though like and you brought up fins and banking like I had a conversation with a big tier one bank and we got in there and we were like, "Oh, this is going to be they're going to have the most sophisticated processes." And we went in and we asked them like, "Well, how do you you guys must have this amazing way to do your loan processing piece?" And they were like, "Oh yeah, we offshore it to a team of people or we had W-2 employees transcribe the data into Salesforce for us." And we were like, "Why?" Like that that's how this that was the peak of sophistication. They were like, "Oh yeah, that's like that's because we don't have a way of unlocking that data in a reliable way that allows us to continue to keep that information flow going, that network graph going." It would and they just couldn't live without it. So for so long it's just been trapped and if you believe in anything like our CEO or CTO, Ben or Aaron talk about if this corpus of data is truly important, the most important thing kind of like the imperative for a lot of our customers is getting it to a system that makes sure that it can be as actionable as possible. Those insights are coming out, the process flows are kicking off. And you can get rid of that offshore team or hiring a bunch of W-2 employees to go and like transcribe data cuz I think all of those people can be better served doing something more fun, more exciting, more kind of impactful for their customers. Again, getting them to cry tears of joy like, "Oh my gosh, the the thing that I was looking for is here." Nick, you bring up a good point though because you know, some of our clients are more mature than others. And so some for for what you just said it would be like, well yeah like modernizing something that already works is great. But for someone else it's very manual today. So those that that maybe are a little bit more ahead on the maturity level, I'd say, you know, here's a great opportunity with new insights that we never had before from this this this area that we just didn't know what we had. And and now we can look at these processes totally different than we ever thought of before. So that is transformational, right? It's not just the quick low-hanging fruit. That's the transformational opportunity that we're seeing more and more. So it really depends where you are on the maturity level, but that's where we hope to be with all our clients is transformational. And I think as a consulting firm, it's really important that we meet our clients where they are. Not everyone is ready to go full force or a a a bandwagon, right? So you have to look at, you know, where are you? What are you looking to do? What is the outcome that you're looking for specifically and then tailor your approach to that client. And you know, I think that's where um we work alongside Box very well is coming to the table with the with a complete solution with that whole perspective in mind. Well, I found this conversation extremely engaging. Thank you so much for participating today. Where could we go to to discuss this more? Oh, I can help there. So we can be found at DC and New York World Tour for Salesforce events. I will also be at Connections obviously and Dreamforce and other tech events that are coming to city near you, I guess. Well, thank you for joining us today. I hope you gained some valuable insights from this conversation. We'll talk to you again next time.