41 | Supercharge Your Decision Making with AI's Data Superpowers with Anthony Alcaraz, Chief AI Officer at Aldecis

0:00Hello, and welcome to Leveraging AI, the podcast that shares practical ethical ways to leverage AI, to improve efficiency, grow your business, and advance your career. This is Isar Maitis, your host. And as you probably know, if you've been listening to this podcast, I'm a huge believer in data and data analysis. And it enables us to make data-driven decisions, making it allows us to make smarter decisions based on data that we have. And that's one of the benefits that we gain from moving to the digital era. However, if we look a little deeper, you can see that we became very good, and some people say addicted, to quantitative data analysis.

0:40We know exactly how many people visit every page on our website and how many clients are paying on time versus paying late and how many people applied for the recent job posting we opened and what's the percentages or proposals we win, et cetera, et cetera. These are all quantitative insights that we get basically from every system we have in our company that gives us these kinds of reporting in almost every department. What we do very little of is qualitative data analysis. And the reason we do less of qualitative data analysis is just a lot of work. So things like interviewing the companies who are not paying on time to see if there's a reason that maybe we can help them resolve.

1:24Looking at and interviewing at clients that have left us and not use our service or our product anymore and trying to understand why. Aggregating customer reviews and trying to find the silver lining of what people like very much and what people do not like so we can improve on both aspects and so on and so forth. The sad thing about not doing a lot of qualitative data analysis is that while quantitative data analysis tells us what happened, qualitative data analysis tells us why it happened. And the why is what really what makes you help, what helps you make better decisions. If you know why something happened, you can change it and then get better results.

2:13And as I mentioned, the reason we're not doing more of that is just a lot of time. If we take one of those examples that I gave, reviewing all of our customer service reviews, people have to go and read every single review and then define them in different buckets of topics and then define if they're positive or negative and then dive a layer deeper of what are the things within that topic that people are saying. It's a lot of work and it requires a lot of manpower. That was true until AI showed up. So now AI enables us to do better quantitative data analysis, but also overlay qualitative data analysis at scale without investing huge resources.

2:54Our guest today, Anthony Elkaraz, is the chief AI officer of Aldesis, which is a company that specializes in exactly this, assessing data, including qualitative data, to help companies make the right decisions based on both quantitative and qualitative data, because because I think this is extremely critical for successful companies, I'm really excited to welcome Anthony to the show. Anthony, welcome to Leveraging AI. In the next few years, AI technology will change our world dramatically. Whether you are a business executive trying to catapult your business forward, or just somebody who refuses to be left behind and want to advance your career, this is the show for you.

3:43I'm your host Isar Maitis, a serial entrepreneur and an AI enthusiast. You'll hear invaluable practical tips from innovative business leaders, AI practitioners, and some of the brightest AI minds in our world today on how you can leverage AI in ethical ways to advance your career and grow your business.

4:06Thank you for having me. Anthony, let's really start with the problem that you're trying to solve, right? Because you're in a company and you have a problem that you're addressing. Can you define it better than I did? Yes. What is the problem and why does it still exist in the year of almost 2024? We're recording this in November. I think there is a problem in both sides, in quantitative and qualitative. For quantitative, there is a huge issue currently with a reporting system that businesses have implemented, is that they are mostly doing what we call consolidated analysis. Is that they are aggregating data to have an overview of their situation.

4:52The problem with consolidating, with consolidated analysis, is that it hides anomalies. It hides, there is, there are compensation effects between values. If you take an average, you don't, you can't see. The outliers. Yeah, outliers. Exactly. And it is, it seems simple, but to get the right granular level of analysis, let's say the correct view of your data is very complicated. And so at my company, we developed a systematic method to analyze data at the right level. And thanks to this, we get seven times more insight than consolidated view. And it is important to do this because you can see problems at the consolidated view.

5:52But the moment that you see the problem, it already happened because consolidation is aggregation over time. And if a problem has appeared and it is so important that it is like a rooted branch. So it has disseminated, it has contaminated the consolidated view. So by only looking at consolidated data, you only see 50 % of current insights, current anomalies, where the value is. Because as a businessman, you want to act on anomalies, on opportunities and risks. Because anomalies can be positive, but also negatives, like a problem on a client or a business unit. Okay. So at the quantitative level, there is a huge issue with business intelligence reporting.

6:49Let me pause you for one second, because you keep going to something that I think is very critical, which is getting insights. So let's define what's an insight, because you said you lose 50 % of the insights by looking at consolidated data. You said it by not looking at it, you're getting seven times more insights. Let's define what is an insight, because this, I agree with you, is the most critical measure that at the end when you're looking at data. Yes. So basically, there is many definitions of insights. Currently, there is no scientific definition that would say an insight is this. But papers that have been worked on, that have been written on this, said that an insight is something that will change your view of a situation or business.

7:42So it's more a subjective approach. What we have taken at Aldesis is a more business-related approach. For example, our app works with symbolic artificial intelligence and we define thresholds. For example, our current use case is for corporate data analysis. And we know that for analyzing corporate data, current experts use certain thresholds of alerts. For example, you can have, I will tell you a quick example, you can have a gross revenue increasing while the net revenue is decreasing. This is an example of threshold that our app is capping. And we have hundreds of those thresholds that we are tracking.

8:37Okay. There might be insights, mathematical insights, a value that is an outlier. And you can have different types of threshold for outlier. But most of the time, that will be outlier that are, I don't know, 25 % exterior of the Of the average. Of the average, yes, the mean, et cetera. But you can have also a logical business insight, something that is weird, but it's not functioning correctly. Like I said, just for the gross revenue and net revenue. What we do, because currently we work with our clients, we work on corporate analysis, corporate analysis. We have defined insights based on expert analysis.

9:24And we use symbolic AI to do that in a systematic way. Yes. This is our definition of insight. To sum it up, an insight is something that is not normal within the data. Yes. That is connected to a business outcome, right? This business outcome can be something that is numeric. It could be something that is behavioral. So it could be like saying an increase or a decrease in a number, but it could also be too many clients are not renewing their service, right? That's not a number. It's a behavior thing. So it could be, like we said in the beginning, either qualitative or quantitative. And it's basically a red flag saying you should probably either take a look at this or take an action on this thing because something that is beyond business as usual is happening.

10:20And I would add something. It's perfect what you said, but I would add that there is no performance without comparisons. So to define an insight, you need to compare products, clients over time that are comparable. For example, our app is comparing business units that are similar or the same business unit with itself over time. So you need a reference to compare and to get this insight. So it would be an evolution. You need enough data in order to even get started because otherwise you don't know what the baseline is. Exactly. You need a baseline, a standard with which you will compare and detect an anomaly, a threshold that has been crossed, etc.

11:08And to do this, to go back to consolidation, if you do this at a very average, a very top-down, what we call a top-down approach, you won't track many insights. Okay, so quick summary so far. We talked about aggregated data being something that is the norm today, but then you're losing a lot of those insights. We said that an insight is something that is not business as usual, that should hint that you should take an action or at least pay attention to something. And I stopped you before when you were about to talk about qualitative analysis. So let's continue really with that. How do you approach the qualitative data in your view, in your business?

11:54Okay. So we talked quantitative, but there is much more, a lot more to say about quantitative. But for the qualitative, so our approach is this. Once we detect insights, quantitative insights, what we are doing, what we are building is to leverage generative AI that are impersonating stakeholders, personas, to look at rational data, qualitative data that are already present within the companies. So we work with large companies. most of the time they don't leverage all their text data all their emails all i don't know reports analysis etc so we use generative ai so let's say a lot of people right now knows llms and chatbot gpt we use that generative ai combined with a way to to to give to those ai corresponding data to find the root cause of the insight so we start with an insight a quantitative insight.

13:01So it is an alert, an alert of symptoms. We do that at the most granular level in a multidimensional analysis. We can talk about after a multidimensional analysis. Once we spot this insight, we have a module called Deep Root that helps the analyst find the root cause of this insight, to act on it. Because what we think, what we believe, is that the where question of business intelligence is no longer relevant for humans because the where question is much more dealt with artificial intelligence in a more efficient way, in a more systematic way. We think that analysts know their role with, for example, our application is a more qualitative job, is a more complex job because it is a complex, you can be in a complex setting, business setting, where are complex, a lot of factors, co-founders, factors, etc.

14:07And it needs to be helped to find the root cause of an insight. So what we do is that we automate insight detection in the data warehouse for corporate finance right now, because this is our current use case, but it can be, it will be a new place in the company. I think any use case is possible because it is a new way to do reporting. And after that, the user is helped with artificial intelligence. We generate TVI. Generate TVI will ask questions. We leverage methodologies. We'll give him access to data. We'll tell him where there is maybe information missing. So we give our artificial intelligence via techniques like knowledge graph a way to help intelligently and with methods analysts.

14:58Yes. So this is a more qualitative analysis. Let me pause you for a second just to do a quick summary because I think it's very important. And I think it's critical for people to understand the power of the systems that we have access to today. So you defined two different steps that are both critical to actually getting to an actionable outcome, right? The first step is identifying something happened. And you said that even this wasn't very easy with systems we had today because in order to handle the huge amount of data that we had to aggregate the data. Once you aggregate the data, you lose some of the details.

15:40So every data analysis platform out there today works that way. I'll add something on top of that that was the subtext of what we're talking about. every business today has multiple systems that are tracking data. And they're not connected. So they're usually in silos and it's very hard to analyze information from your marketing system versus your CRM versus your ERP versus your customer service platform. And in many cases, they don't talk to one another and each one has some clues on what might be the reason for something happening. which leads to your step two. So step one is I can now better identify when something that shouldn't happen is happening.

16:29Step two is now I can actually look at all these sources of data together. And it's funny. I talk to a lot of CEOs. I teach courses. I speak on conferences and I consult to businesses. So I get to talk to a lot of business leaders about this thing. And a lot of people, especially in smaller companies saying, we don't have a lot of data. And the reality is, I'm like, you have tons of data. Just look at all your emails with all your clients and your potential clients. Look at all the marketing presentation that you give. Look at the recordings of every sales call or every support call that you have ever recorded.

17:04All these kind of things are pieces of information that you have in every company. Now, the reason people don't look back at them is because it's just a lot of data and it's very hard to analyze, right? Yes. If you had, and we do now, you're literally telling me that these tools exist right now. You're developing one of those that can look across all these unique silos of data, connect the dots and say, oh, in this email and in this call and in this customer service review and in this channel, I can see a pattern that can explain why this outcome that we're seeing that is an outlier is actually happening.

17:48And this was the hardest part of data scientists is to look through all these different silos and somehow connect them together to get better outcomes. Yes, exactly. Currently, our approach is this. You have finance data and operational data. Finance data is often called lag data, lag analysis, because finance happens every month. every month you got finance data it is an obligation for irs auditors shareholders all of this there is a lot of regulation it is regulatory to get finance data and even with those rules you get data quality issues with finance data so our approach for our first module to detect insights the quantitative approach is to look at finance data okay we are currently also integrating logistic data, but we are not there yet.

18:52Okay, so we take an approach of data quality. So we take our finance data. Our approach allows us to detect very quickly data quality issues. Because with our deconsolidated data analysis, we have a super prioritization algorithm. Because one problem with our approach is that you get many insights. So you need some way to prioritize them. So we build up a prioritization algorithm. And with this, we get a lot of insights. And at first, if a company has data quality issues, you will get insights that are explained by data quality anomalies. You will get absurd numbers, but they are related to wrong numbers put by someone, data quality issues explained by many reasons.

19:44Okay, this is the first step. The second step is that we use operational data. So we simulate stakeholders with GenerativeEI, with a RAG framework. RAG is retrieval augmented generation. So we give to the GenerativeEI TINN data to help the analyst find the root cause of the problem. What we do is completely reverse of the current trend. Lots of BI providers right now have automated data up to the reports. Once you get your reports with average data, consolidated data, it is to the human, to the analyst to find the problem. So what we call currently with data breaks, etc. is to data on demand. Okay.

20:38We think that this approach is not efficient because it will multiply the needs for reports. If you stay at the average level, the problem that you will see, it will be too late. because by detecting earlier problems, by detecting at the right granular level, you detect problem much more earlier that if you wait for it, that it attains the consolidated level. You won't see many insights, so you will need to dig in, to drill down. But drilling down is very approximate. It's very hazardous. You don't know where to look at. Maybe you don't have the resources to look at systematically for every insight.

21:24So I think currently it's very complicated to get all your anomalies, all your things to get down in a systematic way with current tool. I want to pause you just for one second because you touched on two very important points. One is, and it's counterintuitive, but once you think about it, it's very obvious. Once you go to automated systems, because they can look at a lot more data and don't miss anything, you're going to get a lot more alerts because now it's a lot easier for an AI to find where something is not the way it's supposed to be. Exactly. Then you get to the business logic behind it of saying, okay, is this important for the business?

22:10How important it is on a scale of whatever, one to five, one to 10, one to 100, it doesn't matter. And then you really want to dive into only the stuff that makes a big difference in your business, right? So you don't want to spend the time and now have analysts and decision makers and meetings around problems that touch 1 % of your revenue. But then if there's problems who touch 20 % of your revenue or 30 % of your operational cost, if you're looking at the expense side, you definitely want to look at these first before you start diving into the smaller things that are less relevant. So the critical aspect here is that building these systems, yes, AI is capable of doing this today.

22:53And you can even do some of it with tools like Claude and ChatGPT and definitely APIs that connect to these tools, right? You can do this in your business right now, especially if you're a smaller business. You can use these tools. The things you have to remember, and that's the big take from this last segment that we talked about, is that you have to prioritize the things you're looking at based on its impact on business results. Because otherwise, you will have enough problems to take care of every single day, every single hour, instead of actually running the business, because they're not going to be prioritized to the stuff that's actually going to move the needle from the success of the business.

23:32With our methods, it is humanly impossible to deal with all insights. So we get so much insight. And what I haven't talked about just yet is that we systematize multidimensional analysis. In most reporting today, you do up to three dimensions at the same time. So let's confirm what that means. So give me the three dimensions. Yes. Yes. For example, you can have clients, product, region at the same time. Got it. So list of clients, list of regions, list of projects. That's an example of three dimensions you're talking about. It is very hard to do it. Three dimensions in the reporting system, it's hard to visualize.

24:24And it's hard to do it manually. Yes. Okay. But you have much more dimension that you can do at the same time. You can include temporal dimensions, so have many years. You can include multiple scenarios, budget versus real, etc. Right now, our system is able to do as much as dimension that is needed. Yeah, so it goes beyond the limitation of a human to… understand and present more than three dimensions because it's an AI and it doesn't care. Like from its perspective, it's another piece of data that knows how to connect. It does it in a systematic way on the most deconsolidated granular level.

25:17So right here, I am showing Tab when you see the importance of doing multidimensional analysis. Because if you take only one dimension, the other dimension that will be in line in your report will be consolidated. They will be average. Okay? If you combine two dimensions, you will see inside that you don't see with one dimension. Because the lines are consolidated. Yeah. Okay? So you need to systematize a multidimensional analysis. We have some sort of a way to measure the maturity level of a system that is an insight detection tool. And we think that the first step to have a maturity in your insight detection is a multidimensional analysis, systematic multidimensional analysis.

26:12And the only capable system of doing that in a systematic way are artificial intelligence with massively parallel processing. So I want to clarify this for people who are less technical and explain what we're talking about and why it's so important. So let's say I want to compare the success of a new product that I've launched across several different clients. Right. So that's already two dimensions because I can look at a success of each client and I can look at the success of the new product. If I want to combine the two, I'm already looking at two dimensions. But now let's say I want to compare this in different regions.

26:53So now I want to look at North America versus Europe versus LATAM versus Oceania. Okay, so now I have another dimension that I need to look at. And I can keep adding these levels. So as an example, I now want to compare this to the previous product launch that I've done. So I've done a similar product launch a year ago. How does the two compare to one another? The more I add these layers of data, the harder it becomes to, A, even do the analysis. But once you do the analysis, to really understand what's going on. So it's very easy to take one client and look at one client on last year's product launch and this year's product launch.

27:32You go to two clients, it becomes more complicated. If you look at 300, it's a whole different story. Now you look at three, four product lines. So once you start adding these additional layers, the ability to A, set up the data correctly, and B, really understand what's going on becomes very hard unless you're using AI tools like today that can find the needles in the haystack, that can really identify patterns in this crazy amount of data that goes across multiple levels and layers and really give you the right business. insights. Because at the end of the day, everything that we're talking about, that some of it is really technical, is about one thing, which is making the right business decisions in order to have a more successful business outcome.

28:22Exactly. And quicker and better than your competitors. And better than what you had before. So even if I speak to too many businesses who say, we don't have real competition, like everybody does. But even compared to yourself, you want to grow this year more than last year. And in order to do that, you need to do a better job than you do. You too, Shane, extremely important points, because even better than yourself and with deconsolidation, even your best business unit has some dimension, some subcategories to improve. But even your best department that seem better than every other department, if you deconsolidate, if you go at the most granular level, you will find ways to improve it.

29:11I give an example to make people laugh. Even Michael Phelps, the best Olympic champion, tried to improve himself by improving in swimming, by improving his plunging, by improving every part of his sub routine, every part of the sum of his sport. okay it's the same for businesses so by looking only by average by consolidation you are not doing that you are masking these sub anomalies these subs area of improvement and there is a tragedy because we only work at my company with companies or large companies that are invested in a multidimensional database okay so they are able to do this they are able to do multidimensional analysis because they are invested in data warehouse that is able to do this.

30:07But at the reporting level, they are not doing this. Okay. So they have the means to do it, but they don't have the tool to do in a systematic way at the reporting level. So currently, most multidimensional OLAP cubes are useless. And this is why ALDCs, we built it. Because ALDCs, we have a 20-year of experience of building those multidimensional data warehouse. And that's why we come up with this idea to use artificial intelligence to systematize a multidimensional analysis. So I want to ask you two very practical business questions. Yes. One is, who are the people in the organization? Because now you need less technical people.

30:58because if the system is doing its work correctly, it comes out with a business insight. Yes. Here's a problem. And I want to go back to the steps because I think your steps are very critical. Here's a problem. Here's the root cause. Here's a recommendation on how to solve it. Does this go straight to the decision-making level? How does this change business processes? That's basically what I'm asking. Like in the companies you're working with, does this go straight to the VP or the director in order to make a decision based on something? What we do is that we are managing the reporting world with the project management world.

31:43Okay. So what we're building is a tool to communicate between those two worlds that are not communicating right today. So we automate most of the reporting world, like I just explained. And our second module that are looking for the root cause and proposing action plan and writing with the analyst business action plan is a bridge between our tool and a tool like ServiceNow or Microsoft Project, etc. So we are preparing the meeting with stakeholders that will be choosing between different business plan, etc. So this is where we are at.