#339 Modern Analytics with Mike Palmer, CEO at Sigma Computing

Generative AI is transforming industries at an unprecedented pace. But as AI changes how you or your team work, one thing is clear. Your skills also need to evolve. At datacamp, we offer everything you or your team need to adapt and thrive with AI. Whether it's business users looking to get the Most out of ChatGPT and Copilot, or developers and data scientists looking to fine tune models, you can learn the entire AI skills spectrum on Datacamp. Power your AI transformation today. Start learning@datacamp.com. My suspicion is over the next couple of years we'll start to see more of these aha moments where people think this is not 30% better than what I'm doing. It's a hundred times better than what I'm doing. If you try to put a 15% improvement in front of somebody, typically just the cost of change is too high. Yeah, it's 15% better. But I have to learn something new. I have to work to bring this system in. I'm not going to do that. Like it's just not worth it for the 15%. So I don't know, it's so much recommending a great place to get started. As I come back to, my other answer is like where's the biggest difference going to be made? And just do that. Don't waste your efforts, don't waste your time for the just the sake of trying AI stuff, apply it to the thing that would have the biggest possible outcome for you so that your efforts are going to pay off. Predictivity shows up when you act on it. And apps AI apps for Sigma are now the ability for that same end user who because of language and because of spread spreadsheets can get direct access to billions of rows of data at like literally the row level. So they can do the same sort of discovery and analysis, come up with a perspective, they can reconcile that perspective and they can say under this condition, I want this thing to happen. That was always something that would happen elsewhere, right? They would swivel chair into a different system and then they would take an action. They can do that live on warehouse data today. So we've had 5,000 applications at this point built in literally the last 15 months.

Welcome to Data Framed. This is Richie. In the last year generative AI for analytics has shifted from oh interesting demo to huh, that actually works. This has huge implications for the analytics tech stack, for analytics workflows and for analytics careers. It also tantalizingly brings us closer to some long standing dreams. Will we finally get self service analytics? Will dashboards be replaced with something better. Today's guest is ideally placed to help us navigate these new waters. Mike Palmer is CEO at next generation BI platform Sigma. He came to BI from a background in data security, having been Chief Product Officer at Druva and Veritas. He's also had stints at Verizon and Seagate. Mike has a fascinating vision for the future of BI as AI powered analytics apps. I'm keen to hear all about it. Hi Mike, welcome to the show.

Thank you very much for having me. Glad to be here.

Yeah, great to have you. So, to begin, I'd like to talk about self service analytics. It's been a dream that's been around for a few years now. How close are we?

Well, I mean, I think when we started on the path here at Sigma, one of the things that we tried to make happen was recognizing that the skill people had was spreadsheets. And that prior to spreadsheets you had to write SQL, uh, you know, obviously more recently had to write Python. You had to know how to use complex tools. And we figured if you could make a spreadsheet work, you could reach, you know, a billion people. While we're not quite to a billion people, we actually had great success there. And I would have told you we were in general as an industry, I think, making good progress. I think in the advent of AI, we have accelerated that in a really sort of market way, you know, that everyone, we thought everyone could use a spreadsheet, but we're pretty, pretty sure everyone can actually speak language. And that has really lowered the bar to complexity. So I would say that kind of the, the long standing goal of democratization, at least in terms of accessing data, is very attainable in the near future.

Okay, that's kind of cool. It's like you want to do something that is better than spreadsheets, but also has that same level of accessibility.

I think the two things work together, honestly. You know, I think for data discovery, asking a question, especially when you have a billion records sitting in the back end, I think it works as the best method. Would I really want to tell an AI system to, you know, increment over a scenario model? I think that probably is still better, better done in, in a, in a format like a spreadsheet. So I think they'll work together.

Okay, all right. Actually, do you want to talk me through the limits and like, like where does AI work and where doesn't it work right now? Like, how complex can you get?

That's a very broad question. You know, I, I think that you know, one of the areas that AI and then those of us that use AI that we struggle with really are two things. You know, one is, you know, the very fancy words that we use around things being stochastic and deterministic. And then we struggle with accuracy. You know, stochastic, of course, being things don't happen the same way every time versus enterprises love deterministic things when they. They want to know that when they press A, they get B. And if, you know, they should do it a thousand times, they will always get B. When we think about structured data, we like right answers. Is the second area right. So if I'm building things using a model, you tend to get random type of outcomes. Now, the cost of building is very low, so you might be willing to suffer through that or so to speak. But we have to get used to things not being exactly the same and very behaving exactly the same every time we engage them. And we've seen in models that things that we do, minute one, literally 15 minutes later, don't work. And on the other hand, you know, we go back to this idea of structured data where if I work for an investment firm and ask the question, you know, what's the revenue for Sigma? There's really only one answer. And models don't do a great job of that either. Because you ask the question differently, it accesses the data differently. There are a lot of reasons why that it will come up with different answers. And so we're either back to providing models with deterministic things, in which case it's not really a model anymore, or we accept lower accuracy rates and answers, or as Sigma's taking the approach, both where we give customers the ability to ask great questions and add additional conversation and context, but then use tools that are a little bit more, if you will, traditional, so that they can go check those answers to the extent that they need to know them, to a hundred percent accuracy level. So we're in this transition phase. I think there are some things that AI does exceptionally well. Some things not so much.

Yeah, certainly the idea of like something working and then I guess it's like you show your boss and that's the point where it stops working.

Right? Of course. Or the live demo. It's always in the live demo.

Yeah, absolutely. So, I mean, this idea of stochastic things, things being a bit random, this is something that I think data analyst, data scientists deal with a lot of. But it's very new to software teams where these two things being deterministic. Do you want to talk me through, like what's the skill set that needed to be able to deal with these stochastic processes?

This is great, great question. You know, one of the fun parts, you know, working here at the company is just watching how different types of skills are coming into play and having huge importance inside the company as we run it. People that, or skills that I wouldn't even have thought of a year ago. You know, we, we are able to do things because of AI that would have taken an enormous amount of human power in the past. So think about going through, you know, thousands of transcripts every week to understand what, what conversations are going on with customers, for example, or what sort of competitors are we facing. Those are the sort of questions we can ask on demand today because we've created prompt generated systems that we've distributed to the entire company that people can now ask some really important questions that were never even possible before. And they were done by folks who don't have 10 years of software development. Right. They are learning how to, they have a good knowledge of data, they're learning how to prompt engineer in the system and they're creating very usable assets that are having significant impact. We had a very particular example of a thing that was being done in our, you know, our stage two opportunities. Right. In sales, if you're in stage two, it's like the discovery call where if this one thing got mentioned, the conversion rate to booked one was 350% higher. That's not a marginal difference. Right. If you were a salesperson and you realized that you could get one thing done in this call that you would have a 350% higher chance of closing. This opportunity is life changing. Right. It's the sort of thing that you immediately start doing that all the time. It's the only thing that you focus on. But it was only discovered because of some work that someone on the team had done. Just kind of looking through trends I guess is the way that I best put that and how people were engaging on these calls. So massive impact. Completely different than a year ago. Done at scale and done with folks that are in effect training themselves on these systems on the fly.

Yeah, I love that. I mean, 350%. Sorry. Yeah, it's like just a huge, it's, that's the difference between like I'm getting fired and I'm the top performer in the sales team. So yeah, uh, I guess just, yeah, being able to identify this is amazing. But it's not the sort of thing that sales teams would traditionally do because Going through call transcripts and then turning that into some kind of data analysis. It's, it was hard technical stuff. All right, so I've note like what the format is. So I mean, I guess traditionally business intelligence, it's dashboards, it's spreadsheets, it's reports. You're talking about the idea of analytics apps. What does that mean in practice? Can you give me an example of some of these?

It's a good question, you know, and I, I don't very much talk about VI anymore and, and largely because as you just described it, I think the industry suffers from a bit of a almost 30 year old definition of dashboarding, which was like a proxy for using data. And it was a proxy because giving users direct access to row level data was very difficult. It was difficult either because the infrastructure was difficult. If you remember the days of VPNing into your system and getting, you know, networked to get to certain types of databases and oh by the way, you really need to know SQL to do that. This was not something the average person could do. So they got a dashboard and most people today still think about reporting in that way. Unfortunately. The reality of the workflow though was they wanted to manipulate data. So they would take this dashboard, they would download it into a spreadsheet and this is where they would add their personal touch. Right. That was either I'm going to reformat the data to try to view it in the way that I want to view it. You know, pivot table is a great example here. But they would also add data they would want to write their own scenario in or they would want to reconcile the data in some way and then they would create a completely different view than was available in the dashboard. The dashboard of course also at times obscuring like the interesting parts of the data in favor of showing you kind of aggregate trends. But as I tell folks, is like looking at historical data and even forecasting data is not where productivity shows up. Productivity shows up when you act on it. And apps, AI apps for Sigma are now the ability for that same end user who because of language and because of spreadsheets can get direct access to billions of rows of data at like literally the row level. So they can do the same sort of discovery and analysis and come up with a perspective, they can reconcile that perspective and they can say under this condition, having known all of this data, under this condition, I want this thing to happen. That was always something that would happen elsewhere, right? They would swivel chair into a different system and then they would take an action they can do that live on warehouse data today. So we've had 5,000 applications at this point built in literally the last 15 months. Customers who are building inventory management applications, they're building HCM applications, they're building CRM applications, they're doing novel applications in investment management. Just amazing different workflows that range from standard departmental workflows all the way down to highly proprietary workflows that only their company does. But they're building it as business users and they're taking super timely action on data. They know the best. And so this idea, now that we've closed the loop, we've looked at historical data, we've reconciled it and forecast it, and now under the condition we are acting on it, this is the full life cycle of data. And so for us, apps is the always been the part of the category that should have been there but wasn't possible until today.

This is absolutely fascinating stuff. And it seems like there are two kinds of, I think about like dashboards. You have like the kind of disposable analyses where it's like, okay, I'm doing one off thing for me and then you've got like the, like, I'm analyzing like podcast listening data or something and you've got the permanent. This is like our company's revenue. It's a very different beast. Where do these apps fit in? Like, are they good? If you're building it yourself, is it like, is it supposed to be a disposable app? Or is this like an app you trust with your top level company data?

And the answer is yes. Answer is yes. Okay. And that's why. And that's what the beauty of this system is. So number one, the data that you're consuming is all in the warehouse. So from a security and governance and lineage and accuracy and reliability and trustability or so to speak, this is enterprise data. So there is nothing, you know, that, that you would consider like informal about it or inaccurate or unauthorized. Right? This is the company data. But it does turn out that people write apps and do analysis on data across the spectrum of permanence. So you asked really interesting question. It's been a couple of months since I looked at this exact figure and I always tend to try to be very accurate. So I always qualify here. But it was a couple of months I looked at this and we had some hundred. It was like 400 applications at that time that had no activity on them. And I thought to myself at what your immediate reaction is, oh, that's terrible. You know, they built Things and they're not using them and they must not getting value that we looked into it and it was the exact opposite. The cost of writing the application was zero. So the bar to wanting an application could also be just above zero. So they could write applications that for example only lived for four days. Think about going tactical. Example, like they wanted to build a registration page and bring in user demographics and they want to connect that to other parts of their site software stack, if you will. But the event was going to happen and be done. They could do that in Sigma with no cost because they're building right on top of the live warehouse data. They don't have to migrate the data or do anything exceptional to it. They're business users who now have UI tools and AI that can help write the workflow. And the utility of the one day event was sufficient to justify the work to put into it. And then they just stopped using it. And so that anchors you on the one end. On the other end we have customers large like Blackstone, large asset managers that are doing very significant financial workflows that include closing their books that the public markets rely on from them. So you do have the whole spectrum of critical to company financial performance all the way to throwaway work that because it had zero cost could now all of a sudden be made in an automated way.

Okay. I mean, yeah, certainly think about whenever a company has like a sale or maybe a marketing campaign, these things do tend to be like they last a week or two weeks or whatever and you want to analyze the performance after that, you don't care. So there are a lot of these sort of short term requirements like needs for analyses. Okay, all right. So I guess there was a sort of hidden thing at the start where you said, okay, this is all enterprise data is all governed. So it sounds like the governance is the hard bit. Like what do you need to put in place to make sure that you can build analytics apps sensibly?

And this by the way, is where both bi and just, you know, data analytics products help us a ton because you still have to have great data models. You don't want people using data that you don't trust. You still want to have metrics, you want them calling on formulas and definitions of things. You mentioned the revenue dashboard, right? You don't want people making up the definition for revenue for your company. So you want to give them those assets. You don't want to have them made up on the fly. And I think those are super important things for IT folks that really do care about Lineage and telemetry. They want to know who's doing what, when did they change things, what data did they use? These are all things that are built into the administration part of the system. You of course want those because actually the more you build on them, the more, the more the risk increases. And if you don't understand what those underpinning assets are so inheriting all of that governance and structure, I could keep going with audit logging and all kinds of other tools that are necessary to make you. If you want to democratize things, you have to have even a more solid underpinning of administration capability. Those still have to be there. And the better you are at it, I think the more leverage you'll get creating, you will create out of democratization. Because when the 4,000th person at your company writes a workflow, you'll feel good that that workflow was using data that you were confident in. You know who, you didn't have to know them, but you can still see what they built via the telemetry. You know who they've invited to that application because you can also see that in the logging. You know, you can do all your work on the back end, know that, you know that the company is still safe and operating efficiently. So we, we tell folks it's not so much a matter of the user interface or the, whether you can build things, whether they're analytics or applications. It is the moat, you know, for, for many of either the software providers or the enterprises is really is in the foundational data modeling, the security, the telemetry, the permissions management, all of the things that provide the guardrails around the freedoms that the end users then get afterwards.

Okay, so that's interesting. So in order to make sure that everyone can be free to create the analytics applications they want, you need to make sure that you've got all that infrastructure in place to make sure they don't build stupid things or they use the wrong kind of numbers, the wrong metrics.

This is no different than, you know, driving on the road. Right? You know, we, our cars could go anywhere, but we still create lanes, we have stoplights. You know, we, we want to organize everyone on the road to make sure that they get home safely, you know, and I think, you know, we still need that structure and data.

That's a great analogy.

I like that.

Okay, so you mentioned before that a lot of the motivation for this is dashboards tend to be pretty static for every single user. People want to personalize them. Can you give me some examples of like how different Users might want to personalize a dashboard or how they consume analytics.

I mean, I think the most simple example and BI has always been things like filtering. You know, people don't want to see the data, the everybody data. They want to see it for my region, they want to see it for my store, they want to see it. They want certain sort of conditional formatting. They don't want their dashboard to be complicated with five other elements on it. They just want to look at this one thing. And in traditional bi, every time you made a change, you made a copy. And this kind of, this created massive sprawl. And the outcome for a lot of technical organizations was there's hoarding a bunch of assets that they knew weren't really being used and were marginally different. It also created the need to service requests from all these people, like, hey, can you modify this for me? And then create this new copy? The desires are still there. Everyone wants to personalize. They want to see things that are the easiest way for them to get their jobs done most effectively. So you can't stop that. In fact, you need to service it better over time. There are basic things that you can do in Sigma, for example, to address this, you know, we allow a feature that our customers love called bookmarking. I can build them a central asset. It can have, we'll call it 20 elements of things in there. There could be charts, there could be tables, there could be dynamic text, there could be all kinds of things. But in my job, it might say in a person, number two's job, that person could say, like, I don't even, I only need three of those elements. And they can go into that workbook and delete them. But they're not being deleted from the workbook, they're just being deleted from their view. And then they save that bookmark called number two's bookmark. And every time they log in, they will just see that. And in fact, they can share that with you. So you will also be able to see their, their dashboard. Even though the backend workbook's exactly the same workbook, it still has the 20 elements in it. So this ability to create personalized views that create efficiencies, time efficiencies, because you don't have to recreate that view every time that you want to get to the thing that you wanted to understand. These are again the sort of things that modern products that are operating live on the warehouses. These are the things that I think make a big difference for people in their day to day work.

Ah, this Is interesting. So I guess if I understood this correctly, you've got an application which is for a sort of broad group of users, and then you have views which are going to be personalized for individuals. Does the level of personalization affect, like, how big an audience you want? I think often dashboards are built with, with a sort of a very common audience in mind, a broad audience, because it's, it's got to appeal to everyone because you only get one of them. In this case, if you've got different views of different people, like, how do you think about how big the audience should be for a particular application?

That's a great question. And it's almost. You don't have to answer that question because the an. The answer is you can create things for NN of one and then. And the. Actually the idea here is that no one's creating things for anyone else anymore. Right? I am able to create this for myself and I am the n of 1. I don't have to worry about whether the time that I spent is going to service 50 other people, because as long as I think it's valuable for myself, I can put the time in to do it. So, you know, just like you would never think, how many searches can I afford to do on Google today? You know, the answer for Google is limitless. You don't have to prioritize the number of searches. How many different views do I want to create is up to me as an individual. That would make it useful for me. And it's not pressuring the system any further on the back end. So it's not a noisy neighbor problem is what we tend to call this, where you have multiple people going after the same resources. You don't have to worry about the time of an administrator who has to create an asset for you and has a long list of priorities. You just do these things for yourself. So I think architecture has alleviated some of the traditional questions about prioritization and resource bottlenecking, you know, simply by allocating the work to the. To the person who actually wants it done.

Okay, so build stuff yourself. And then if it's cheap to build stuff, you know, you build whatever. Okay.

If the cost is zero, you'll build more.

That sounds very true, yeah. Okay, so what's the new workflow then? So suppose you start with, okay, I've got. I've got some business questions you want to answer. It's going to involve some numbers. How do you go about solving stuff?

It really depends on how. How the beginning of that cycle is you could have a question in your mind, but not know where the data is. You don't even know if we have the data. Right. So the first thing you're going to do is you're going to go into Sigma and you're going to ask a natural language question and you're going to say, you know how many people you know put bets on my favorite football team yesterday, and it's going to return information about not just an answer, but where it got the data from. So now you're going to see, oh, wait a minute, there's data on the customers we had that place bets. We have data on the teams that we're playing. We probably have some additional things. So, okay, great, I've got an answer. But I probably have now three or four more questions based on the data that I've seen. So I'm going to drag that in and I'm going to say, well, you know what, I'd really like to do a scenario based on next week's game. So who are they playing? I'm going to bring in some new data. I am going to build a scenario. So now I'm going to do like a forecast. I have a variable or maybe a set of variables. So I'm going to add that data. Sigma is unique in its ability to do right back to these systems. So I don't just have to leverage the data that's there. I can add new data. I can even upload a spreadsheet of data that I have my laptop that it created relative to what I think goes on in these games. So I can upload that, I can join it to the table. This is called an input table. I could join this to the existing enterprise data set. So now I've written back my conditions, I've added new data, I've taken the existing data, I've created this amazing forecast. But I said, you know what, I didn't do this for myself. I did this because I want our marketing person to know what might happen next week. So what I'm going to do now is I'm going to invite them into this workbook and they can collaborate and live edit with me. And we do that together and we agree under this conclusion for next week. If we have this number of people that are starting to log into our system, let's say to place a bet that we want to notify our cfo. So we're going to put a conditional action on this, the CFO is going to get notified under this particular situation and they're going to come back into the system after that notification. They're going to add some information about the cost of marketing this for us now that the user base has gotten to be a bit bigger. And then we're going to say, okay, great, we've got new financing coming in. Our CFO understands what we're doing now. We want to place. We want to take the users that we've seen in the system and we want to now move them into our, into our payment system because they're going to be net new users and we want them to be able to transact. So we're now going to make a, an API call or via an MCP server on the other side and we're going to say, hey, we're going to take this data and we're going to send it to that system with the instruction to add that to the new user table for the payment system. You could do all of that from one interface. Nice.

So it's really like there's. It's the full workflow from, okay, where's my data through to like, let's build stuff to having something that's actually usable by the end users.

I didn't even give you the part of that where I call a or AI over the top warehouse and model services where it can recommend things to me, it can give me additional services while I'm doing my work.

All right, nice. So this all sounds amazing, but I guess there's been challenges where, okay, AI, generative AI doesn't always work. It doesn't give you the right answer, and then you stuck with debugging stuff. What happens when AI goes wrong?

It's a great question. And it's funny that we have this. Everyone uses this term, human in the loop, and I don't use that term. I always say it's AI in the loop because I think human in the loop sort of presupposes that AI is doing most of the work and that it's asking a human for small interventions. And I think where we still are today mostly is humans doing the work where AI is adding value at points in time, but we're still dependent on the human in almost through the entirety of the workflow. So we see our customers leveraging AI. We don't really face the question of AI failed and now we're in trouble because again, in the example I gave you, humans are calling services along the way. And if I'm calling a service, for example, to do analysis for me, and I can get that as an overlay and I Can look at it and think, hmm, uh, that's interesting. I'm not sure I believe that. Let me go test that myself. It's an asset to me, but it wasn't integral, you know, to being able to accomplish my output. It's an assist in the case of building an application in Sigma which you can leverage a front end interface for Sigma, you can say, build me an expense management application. It will go to the LLM, it'll come back and say, I think an expense management application has the following components. I'm going to build this and it's going to leverage tools that Sigma gives the model to on the fly. Build that application. If it fails, no harm done. If it gets the job 20% done, you can finish 80% with tools. If it gets the job 95% done, you can use tools for the last 5%. So again, we're still in this world where we're getting a big accelerator from AI, but we're not giving it the nuclear launch code. So we don't really see like too much of the. I'm going to blow things up if AI gets it wrong.

Okay, yeah. I mean certainly if you're drawing charts or things like that, then it's usually you've got like a visual indicator when things are going wrong. I guess it's when you have things like maybe a number is slightly off and it's not clear whether it's go wrong. So are the tools around like to help you surface any potential issues like how do you uncover what might have gone wrong?

It's a great question. And, and the, and even a bigger question is how would you even know that things might not be right? Part of that is what we call just chain of thought. Like we want to understand when we get an answer from an AI system where it came from. We want to know what data set was used, we want to understand what filters or formulas that were applied and then we want to pressure test that. So for example, in Sigma, I can get that answer from the AI system, but I can also see the places it went and the steps it took and I can manipulate that data so I can take a, a six step chain of thought and I can change any one of the steps and it will recalculate that number on the fly. So as a human I can get to understand the relationships of the decisions that it made relative to the outcome. And I think we're going to be in that, or at least to a certain degree we're going to be in that. I want to trust but verify, you know, what I'm getting from the system for quite some time. Okay.

So yeah, certainly it seems like having the chain of thought, there's going to be a big boost for explainability. Are there any kind of like testing tools or other things, ways of like evaluating quality?

I don't know that we have a one size fits all way of thinking about enterprise way of testing quality to each and every user across each and every vertical. Not sure that's a problem we're tackling right now.

Okay, I was wondering whether, like, is it, is that an area of research or is it just like not a necessary thing for Dashboard for analytics applications?

A good question. You know, I think right now we're relying on the user's testability for, for that sort of work. And of course we see plenty of observability tools just trying to understand the quality of the data over time. I don't know that I, I see right now a need for that, partially because the folks that are engaged in the data in Sigma are the ones that know the data the best. And I think we've struggled for some time with centralized data teams having to serve data to users. And in a world where if I'm in marketing, I tend to understand my marketing data better than a data engineer does and the more I can have control over that data, the better. Now I may not understand our revenue data and by the way, in marketing, I may want to create metrics around marketing data for other people to consume because I don't want them trying to create on the fly analysis of marketing data without some guardrails that I understand best. I do trust people, you know, I think that people engaged in data are going to create great outcomes. I think we see that every day.

Okay, yeah, certainly it seems like at least traditionally you've always had to have like the data skills and the tool skills and then you also had to have the domain knowledge. And it's been quite rare for it to be one person with all three of these.

That's right.

Okay, so actually it sounds like this is a very big change to workflows to like the data analyst role itself. Talk me through, like, how do you think being a data analyst as a career is changing and how do you change your skillset to keep up?

I think it's a great question. Data analysts are going to be more valuable in the future than even they are today. And they're already not only a hugely valuable, but a hugely constrained skill set in most enterprises. Number one, if we just start with AI, it's because we're down to the same age old problem of garbage in, garbage out. So we need to have highly centralized, good quality data. Data has to have great semantics. You want to have more metadata, not less. Right. So we're still in a world where more data is better than less and where modeling that data for quality and consistency is going to be super important because the more you get that, the more the system can train on the sort of data and the data outcomes that will create more reliable answers. So fundamental to AI is a solid underpinning data strategy. I would say that there's a whole new area coming toward data engineers, though. As I mentioned earlier, there are, I think there's a revolution coming in software. I think that if we take a little bit of a historical context and we moved compute and storage through this highly fungible cloud thing, and we've now consolidated a lot of what were formerly disparate databases, these cloud data warehouse platforms, we are now going to see a dismantling of the silos of the software products that we then brought to cloud in the form of SaaS. And as those highly specific, typically departmental software applications get dismantled, there's going to be a whole new skill and focus area for data teams to work with business users on, you know, which is, well, wait a minute, I didn't need to buy an app to create that workflow. I could, I could create it here. So let's think about the underpinning schema that we want. What tables do we have available? We're going to create workflows that the software we purchased could never deliver because their schemas were fixed, but ours is not. So we can create workflows that formerly would have been three or four different applications, but now are just tables in the warehouse. So we can work together on creating far more valuable, far more customized applications. We'll have to think about all the usual things. How do I make sure that those apps are available? How do I think about sdlc? How do I think about permissioning? Now these things are, you know, but now not just within the context of my analytics system. I'm going to be applying all those concepts all the way through these thousands of applications that are being built. And I think that's going to really expand the, the, the remit, if you will, of the, of the data team. And instead of being the team that looks at the output of applications dumped into the, into the database, they're now going to be part of building those workflows. That's what we see here. At Sigma, we have ourselves on the order of 80 applications that we've built in Sigma that run our daily activities from finance to sales operations to expense management and other things that we've built, all with our data teams and our departmental users. So we know this is possible. And the savings for these companies are off the charts. You think about the millions of dollars that we're spending in what we call system of records or departmental applications that could be done for a fraction of the cost in platforms like Sigma. I think the people that can build those systems and understand workflow without having to go to buy something off the shelf, I think they're going to become hugely valuable in the future.

That's fascinating. There's a kind of key phrasing that you said, dismantling all these SaaS applications so you can get rid of them.

That's right.

So I think a lot of, well, probably almost every business now just built on top of like dozens and dozens of these SaaS applications. So which ones do you think, like organizations are going to be wanting to get rid of? I'm sure we've got a few chief financial officers listening, going, oh, we can save some money here. Yeah, what can you get rid of now, do you think?

Oh, I mean, the last meeting I was at even coming forward to do this podcast was the progress we're making on our own, what we call our long range plan and our OPEX modeling and our fpa. So, you know, think about the anaplans of the world. You know, no one loves that product. It's too hard to use, it's expensive, deployments often fail, and they're not really very customized to the finance flows that an individual company has. Those are the sort of products that are going to go away. Think about the highly focused applications that do things like sales forecasting, or they do things like territory management, or they do things like commissions planning. You know, if I'm a sales ops person, I have three, four, five licenses to do my job. And if my job description requires five different software products, this is a terrible thing. It's complicated. It's expensive doing all of that in Sigma today. So those are the products. I do think that eventually we go after the, you know, the, the biggest of the big software applications, the ones that do things like, you know, human capital management, the ones that do the full sales pipeline management. We've already placed a number of inventory management systems. This is a five year plus trajectory that we're on. But for the customers that I mentioned that are already built these 5,000 applications. They've already benefited from not buying third party software. They've already benefited from automating things that were too costly to get a developer to do for them inside their company. And you know, third party software is going to have a couple of renewal cycles ahead of it, but not without, but they're not going to have growth, you know, and I think that AI is going to be a big fuel for this because the folks that understand the workflow are going to be very participant, if you will, in the, in the ability to deploy these applications, which means that the risk and the accuracy, if you will, of what they get on and the outcome is much, is much better. Wow.

I mean that'll be a huge shift in terms of like the enterprise software stack if all these kind of big name platforms are going away. Yeah, I mean there's a ton of like enterprise software that's clunky. You need to have like people who are just there. The whole job is just configuring it for your own company. So yeah, if we can get rid.

Of some of that, then customizing it, ETL ing its data, applying security principles to it. I mean the cost of software is much higher than even just the cost of the license of the software. It's the, all of the other things that you have to do to make sure that that thing lives in your environment successfully. Those are all opportunity for cost savings. Wow.

Okay. All right, so suppose I'm convinced for this. What's a good first analytics application to build? Like what's a simple hello world type thing?

Wow, that's a great question. You know, I think people only build, build things that they feel like are have the best ROI for their particular business, you know, and the first thing you know is what, in what vertical context am I living? If I'm, if I'm a retailer, the answer is almost always inventory. I'm going to build something that gives me a more accurate version of inventory. I won't name this customer. We have a particular customer. They've got chains of food that they sell and they, they ship inventory to them basically on a daily basis. And they want to know things like what activity around this chain store is going to increase or decrease the foot traffic. I want to be able to pull in third party data sources to understand that they want to understand things like I thought I shipped 30 cases, but I only can see 25 of them that were sold. Where did the other five go? Oh, they were dropped on the floor during the day because of the activity level. I'm not able to track that data. I want a component of live data updates that come from the fryer to the centralized system so that I know I should ship more food to them the next day because the last thing we want is for them to run out. Those are the sort of applications that get people out of their chair. Because it's not give me a better dashboard. It's like, holy cow, this is what my business is. It is selling food. And my understanding of how much food I have is far less than it should be, and it could be much better. So I have less spoilage, I have less stockouts, I have fewer stockouts. Like, these are, you know, this changes the very bottom line performance. So I, I advocate to folks like, I don't think you should start with something that makes you sort of feel good in terms of the I accomplished an app. I think you go right after the heart of where there will be the biggest roi, because where there's big roi, the momentum to keep building is going to be much higher.

Okay, yeah, I love that. Just giving greater visibility into like what your business problems are you're trying to solve so you can make better decisions. That seems absolutely essential. Do you want to expand on that? Like, suppose you go all in on this, improving your analytics capabilities. Like, what does success look like? How do you even measure it?

Great question. I think that we have a very, very simple way of looking at what the future should be. I think that a few years from now, customers are going to have maybe 30% as much purchase software. I think they're going to have like 300% as much what I refer to as proprietary software. I think they're going to be building at an increased rate. For me, what Sigma is going to look like three years from now is an AI apps platform that has a very solid underpinning in data access and governance. So that as those workflows are built, they're built on solid understandings of accurate data. But I think success in the end should always be based on productivity. And productivity is either I made people more efficient or I reduced risk, I reduced cost, or I found a new way to engage with my customers. Obviously on the revenue side, that almost always happens through applications. I don't think the world has changed. I think in the end, software is still the means for most companies to grow more efficient over time. AI is going to be a big part of creating that software, and AI is going to be a big part of adding value in the workflow process where AI understands patterns. For example, that a Human might not easily pick up on at the moment that they would most benefit from that data. So I think success in the end is we've all built our own workflows ourselves, that we're able to do that for zero cost or near zero cost and that the work that we're creating is highly customized to create competitive differentiation from what I do from what you do.

Okay, yeah. I love the idea that like at the end of it you go back to like top level business metrics like have you increased productivity, have you saved some costs, have you increased your revenue, have you improved your customer satisfaction, all that kind of stuff.

That's right.

It does feel like the buying software is the easy part. Getting people to change their processes, that's going to be tricky. Do you have any advice on like how do you go about changing to new analytics workflows?

Great question. I do agree with you by the way. I would say that like technology changes first, products change second and people change third. And you have to have all three. Right. Because I think we've seen a massive change in hardware which is driven has made AI possible. Right. AI was always, has been around for many decades and the theories behind AI but it's really in the hardware that we got. Performance building products that leverage AI effectively. I think think is we're kind of in the early stages of that today and therefore we're seeing a bunch of exploratory use cases. The obviously the, the best known right now are things like code generation. There are some others in things like call center, but everyone's trying, you know. And my suspicion is over the next couple of years we'll start to see more of these aha moments where people think this is not 30% better than what I'm doing, it's a hundred times better than what I'm doing. And then you won't have to convince them because people recognize 100x improvements. If you try to put a 15% improvement in front of somebody, typically just the cost of change is too high. Like why? Yeah, it's 15% better. But I have to learn something new. I have to work to bring this system in. I'm not going to do that. Like it's just not worth it for the 15. So I don't know, it's so much, you know, recommending a great place to get started. As I come back to my other answer is like where's the biggest difference going to be made? And, and just do that. Don't waste your efforts, don't waste your time for the Just the sake of trying AI stuff. Apply it to the thing that would have the biggest possible outcome for you so that your efforts are going to pay off.

I love that. It's just thinking about, like, what's a cool use case? Like, what business problems am I really trying to solve? Where am I wasting time? And then, yeah, go for it there. Okay, so just to wrap up, I always want more people to follow. So whose work are you most interested in right now?

In. In the tech space? Yeah. Yeah, that's a really great question. You know, I. If you're not fascinated by the model providers, then, you know, I think you're not paying attention. And I, I find it a very interesting industry because of the neck and neck race that we seem to constantly be running out there. You know, you've an OpenAI model update comes out, you have a Gemini model comes out like it, and it's just advancing. And I love that because they're, they have to be so laser focused on what they're doing, it'll extract the best out of them. I also think it means that those of us that are building on those products as platforms have a great opportunity to find differentiated value, to take them to end users. I have to watch them the most because it's those capabilities that give us capabilities that benefit our customers.

Yeah, definitely. Fierce competition is a wonderful thing, especially when it's like someone else's industry, not your own industry.

Well, trust me, we're all competing every single day. Adam Smith would be very happy with us. Wonderful.

Yeah, there's a lot of exciting stuff going on. So anyway, thank you for your time, Mike.

It was an absolute pleasure. Thanks for the conversation.