MicroStrategy CEO Phong Le on Generative AI and Business Intelligence

hi I'm James Maguire and on today's e speaks we're talking about the combination of generative Ai and business intelligence a combination that offers enormous potential but also all sorts of challenges to discuss that I'm joined by a major industry leader with me is Fong Lee chief executive officer at micro strategy Fong absolutely thrilled to have you with us today Jim I'm excited to be here and uh obviously this is a topic that that I and micro you're passionate about so I'm looking forward to the discussion great so I think the combination of generative Ai and business intelligence offers a lot of potential we know that but it feels like almost the marriage of uh opposites a little bit in that business intelligence has always been so steady and people rely on it for data analysis whereas generative AI is new it it could be prone to hallucinations a lot of potential vast potential so what is what is the current Enterprise state of of this emerging combined sector um it's early right right it's nent in many cases micr strategy has been doing business intelligence to your point for over 30 years now right we invented the sector as you know it and there haven't been that many major Innovations in this space especially in the last 10 years in fact I I like to think business intelligence lost its way a little bit going into the data visualization World which isn't really what I think of as business intelligence getting true insights uh and action able insights in your business and and gen has uh in my opinion breathed new life into the bi space uh and redefine what's important right with geni what do you care about right it's it's it's often said now it's really the underlying data data used to train your llm data used to produce your answers but to date geni has largely dealt with unstructured words right and so if you ask it questions it can answer you verbally very simple questions because it predicts the answer using words and that's useful in a consumer context is useful in certain business contexts but when you get into sort of the core of what businesses need they need the answers to numerical data I'm an ex CFO of micr strategy like what I want to know when I ask a question to anybody an analyst or eventually a geni bot my forecast for Revenue or my product sales in this particular country for the next six months I don't want a predictive answer I want the answer sure and that's what gen AI plus bi starts to do and solves a problem that isn't solved today via just uh llms who are predicting verbal you know Word answers well what about the issue of trust you know sometimes generative AI comes up with some pretty wacky answers whereas we can't allow that in a bi environment yeah and then that's why this combination is so important what gen gen AI gives you is it gives you the ability to ask a question in natural language it gives you an ability to get the answer back in natural language and what it doesn't give you is numerical trustworthy answers so if you combine this idea of something that is uh that that's smart right like that can understand what you're saying right and start to give you a predictive answer and very nice natural language with the Precision of bi and the data and the numbers and frankly the security layers that bi has built be that that intersection is what we're focused on very difficult engineering especially prompt engineering but when you get those two things together Jim when you get sort of this really nice easy to use interface combined with the ability to give you trustworthy secure accurate answers that's magical well I want to ask ask you about uh what a business should look for if if they're shopping for a solution like this but I want to back up even before that question and ask you do you think businesses are are ready for this I mean they think they think about the combination of geni and bi and go yeah we're ready or it's like oh we not yet maybe well I mean what do you look for something that works right and so and and when will you be ready when you find something that you ask the question and you get a answer that is uh right 95% of the time what happens today when somebody wants to ask a data question right like best case maybe they're going to a spreadsheet or a dashboard they're doing some analysis it takes them two days maybe longer two weeks to get a good answer right worst case they call up the person who's their data analyst that's you know they've been working with for the last five years he or she has immediate recall very fast ability to do math in their head and says yeah I think it's 42% right how accurate are they who knows but they're very credible now what if you knew that you had a data analyst bot what if you had a financial what even HR analyst bot in Microsoft teams or in slack and you can go there in there you can say what do you expect my voluntary attrition in the technology department to be for the next three months so not having to build a dashboard and take two weeks not having to call the data analyst or HR analyst so you don't really know if they know the answer right and just go to the bot and it tells you well your attrition for the last three months has been 4% but based on seasonality and some information we're getting from our uh Employee Engagement survey we think it's going to be 6% for the next three months and getting that answer embedded in Microsoft teams underlying you know it's the micr strategy bi engine plus some llm giving you the answer wouldn't that be a beautiful thing and it gives you the answer back in eight seconds do you think we're in a world where using a business intelligence platform is going to require a prompt engineer I mean not not for those easier questions but for the more advanced queries do we will we need a prompt engineer well and and so that's what we are working on we're not you we're not building an llm right micr strategy doesn't need to because the large hyperscalers out there whether it be Amazon uh whether it be Microsoft whether it be Google anybody else they're building the LMS what we are doing is the prompt engineering right so that we can take somebody's question turn it into a reasonable prompt to the llm to give us a query to run on the data right so uh yes a prompt engineer is a good skill to have but just like you don't want the person who is the best data analyst Excel modeler to be able to to to to to be required you don't want to always have to have a prompt engineer you just want the answer to the question right so I guess what we're doing where an llm might be replaced in um you know a a data scientist what we're trying to replace is the need to have a prompt engineer and is that through natural language processing what is a little bit down to the weeds how do you do that yeah so so think about it this way and and for those who not familiar with how micro strategies work for the last 30 years right we're a business layer on top of a data warehouse or on top of your transactional databases Etc and in that business layer we Define the definitions and objects and metrics and attri attributes of a corpor of a company right and and usually somebody can through drag and drop or through queries then use that business layer to ask questions of the data right or query and pull data that's what micro micr strategy has been doing for the last 30 years right so now when we add gen AI into that what you're doing is let's say you ask a question of micr strategy and use our ai ai bi capabilities you're actually asking the question natural language what do you expect my attrition in the engineering department to be for the next three months right the first thing that we have to do there is interpret what that question actually means and that's where the llm is great we send the question to the llm along with the metadata the structure of the underlying data and the llm comes back says this is what they meant this is the query you should run against the micr strategy data and then we go run the query like we always would do which gives us the performance it's against structured data it's against secure data and then the answer comes back and then we send the answer to the LM the LM comes back says here's how you answer the question in in you know in a friendly tone and so we've just called the llm it gave us a query we ran the query we gave the answer back to the LM the LM comes back says this is what you say and after two round trips we give the answer back to you that's what we're doing essentially so we're doing the prompt Engineering in between and using the llm for natural language capabilities makes perfect sense so all right let's drill down into the micr strategy bi gen AI offering uh you've got a product called micro strategy one what distinguishes this from competitors uh a few things one we're first to Market we launched in September of last year and we release our software every month so we have now had nine cycles of improvement upon the software most of our competitors do not have a ga product today right they're still working off the beta product and they're trying to give work the second piece is because we have the underlying call it the semantic layer that business layer right which our competitors don't have right it allows us to again sort of what I just described we're still running queries against our underlying data which is performant which can be trustworthy which is secure that's huge right because you know we started this conversation of well if you don't trust the data you can't really trust the output of the AI sure well here we start with trusted data right and therefore you can trust the output so we're first to Market we have the underlying semantic layer uh and and and together with those things we're in market and people can try it and it works right everything else you're doing I've had folks try beta versions of competitive tools and they look nice on the surface but you don't get an answer that you can actually run your business with and I'll say the third thing that we do is because we're an open ecosystem right like over time you can use any llm can use on any Cloud you can embed the output not just in a dashboard but uh you can build your own bot that you can embed in a web page as an example on any application you have can you talk at all about the some of the Growing Pains of developing the product it seems like the technology combination is so new that you may have encountered some challenges or or maybe the clients have are there any challenges involved in the process yeah of course I I mean we measure ourselves on two dimensions of output one is speed and the second is accuracy right and you know we're striving for when someone ask a question of the data that it give an answer right greater than 90% of the time right and generally speaking if I could ever get 90% accuracy from anybody in the company when I ask you a question I would be darn happy right even even a simple question so we we feel pretty good about the 90% uh on the speed piece right uh in some cases we're riant on the underlying speed of the LM so as we went from gpt3 to GPT 4 we saw a a 2X Improvement in speed so uh we're starting to get better and better and and you know generally we're giving answers in 8 to 10 seconds people are very happy to wait 8 to 10 seconds versus you know 8 to 10 days to get an answer or even eight to 10 hours to get answer 10 minutes so that's not too bad um back back back on the the accuracy of the answer the other challenge we've had is understanding the nomenclature of a company right right like so you know is revenue and sales and the same thing or turnover right and and that's just general you know uh company agnostic terms imagine if you're working for a pharmaceutical company right they they've had they've renamed certain uh C certain Pharmaceuticals every two years sure C or you talk about uh geography in a multinational right like different countries different cities different names of business unit so uh we now have the capability to for for folks just to basically upload a data dictionary set of summaries so that we can include that in what we tell the chatbot and that's fixed that but you know if you ask a question and you use different terms that mean the same thing a chatot wouldn't normally just know the the language of that particular Enterprise so we' learned a lot about that in our training chap out on that and that's when it gets really cool is the personality that Chapa is not just a financial analyst but it's a financial analyst at micr strategy or it's a financial analyst at you know ex compan that's when it gets really neat that's really interesting actually it's fascinating way to cure that problem um I think the big question is what do you see as the future of business intelligence in G of AI I mean obviously there's a lot of potential here but some challenges as well yeah so I'm excited about the future like like I said at the beginning this is this is the the biggest change to the biggest additive change to business intelligence that we've seen in decades right and and look geni has done that for for every sector of software but bi especially uh so I I'm really excited about what the future prends I I think as we move over as what's happened in the last 10 15 years well go what's happened in the last 30 years in the bi space is we've tried to give information out to the fingertips at the edge of an organization right and that came in the form of dashboards and applications uh and what I think people have found is the farther you go away from corporate the farther you go away from core the less likely people are to be comfortable consuming that information in a traditional B dashboard or a traditional grid report when when you say core you mean core data so I interrupt sorry that the core of the organization right core corporate versus The Edge being a field worker imagine like a a finance person in corporate versus a store manager in a retail store got and if many would argue this getting the data to for the to the store manager in the retail store to take the right action much more important than the corporate finance person having the data to take the right action because he or she will find a way to get to that data whereas a store manager is going to use Instinct rather than consume the data in a grid report right um so what's happened what we tried to do over the last three years how does how does those Edge workers right how do those folks how do we get them to use bi more and more and so we went from traditional grid report showing up in your email or print it out it's was like well okay forget print out let's give them an email well they're not really reading their emails let's give it to them in the web where they can Self Serve well they don't really go to the web let's give it to them in a mobile device okay now we're moving somewhere and then we went backwards said oh why don't we let them build their own visualizations we know they're not doing that right what gen does is is it it solves that last mile problem where that retail store worker rather than all of these consumption paradigms that they don't use now just let them ask a question natural language you know what's my inventory right at this particular Point into a mobile device into a headphone into a zebra device whatever ever it is now now we now we're solving this final mile problem last mile problem of the employee at the front line using bi and data to make decisions and I think gen AI solves that last mile problem well so what you're really talking about is the democratization of generative AI so more workers can use it yeah it's de the democratization of data I don't like that term because it gets used and weaponized into sort of everybody FR should get have access to everything right that's not really the answer the answer is people get access to what they need when they need it the precise information great F I think you said it a lot of good stuff I certainly learned a lot uh thank you so much for sharing your expertise today thank you for your time uh super exciting time for bi and AI very exciting time for micr strategy too so I appreciate your interest

This transcript was generated automatically from the video's captions and may contain errors.