Tobias Zwingmann: AI for Effective BI

Loris Marini: why we're here today? Today we are here to talk about AI in the context of business intelligence. The biggest outcome and probably number one outcome from this episode, or the message that we want to convey is that AI is not here to replace you anytime soon, rather is here to make you more effective.

So today, we'll learn how to leverage ai. To achieve more with less. And my guest today is Tobias. Tobias was featured on the podcast a couple, few sometime ago when we did a live event on habits for data engineers. Tobias is a managing partner of Rapid ai, a Germany based startup that helps businesses adopt AI and machine learning faster while delivering relevant business value.

Tobias is more than 15 years of experience. Working in the corporate world where he, his responsibilities included building data science use cases, digital B2B products and developing an enterprise-wide. Data strategy, which is why today this is super relevant because we're gonna focus on the business value that AI can bring.

Tobias is also the author of the rally book, AI Powered Business Intelligence and a data science mentor for Springboard, where he helps students land a career in data analytics. Super stok to have you on the podcast, Tobias. Welcome and thanks for taking the time.

Tobias Zwingmann: Yeah. Hi Loris. Thanks for having me. So stoked to be here. Thank you.

Loris Marini: Alright, so let gimme just a story, the story piece. How did the book come about? What prompted you to sit down and spend countless hours to, to put

Tobias Zwingmann: Yeah, . So I think it's very often in life, opportunity just comes. And in this case, the opportunity came to me when I was in contact with another O'Reilly author who brought me in contact with O'Reilly. And I got to know they were looking for that kind of book, which.

Bridge the gap between AI and bi somehow. It was not really clear how back then but this kind of really resonated with me a lot because I find that, especially in corporate worlds, there's so much potential, just left behind by just focusing on data science and even the whole bi part, which has been there for decades just unnoticed.

And I pitched the book proposal and that's how it all happened. And luckily got accepted and was so excited to write the book and got this opportunity.

Loris Marini: It couldn't be a more topical subject. This one, especially with what's happening these days with generative AI and chat BT and the likes. Notion just recently launched a better program and we started using their AI in, built in into the platform, which is incredibly convenient and it's a bit scary as well.

So before we, we really focus on the business outcomes, let's talk about the psychological stance around this advances in ai from your perspective, cuz you've been exposed to artificial intelligence for quite some time. I do have my beliefs and I know that my listeners know what I think about it and but I'm interested in your perspective.

Is it something, what do you see first, like in from by being on LinkedIn and listening to what people are saying, are we reacting in the right way to this transformation? And what are the biggest opportunities that lay ahead that maybe we are not even aware?

Tobias Zwingmann: Yeah, it's actually funny because in the very beginning of my book, I write about how people typically perceive ai. And this image of AI is typically influenced by pop culture. Think of Terminator or Skynet or Matrix, right? It's all about AI coming, to threaten humanity, right?

And to 

Loris Marini: Yeah, 

Tobias Zwingmann: take control over, over humanity. And that's why I think the first impulse of a lot of people is, to think, okay, if there's a technology like ai, how will it replace me? How will it threaten me? And honestly, I have no idea like where we'll be in 20 years.

So hopefully, no, there will be no takeover by ai, but at least for the time being, it's not AI that is coming like for you or for your job, but instead, that's also what I tell in my book. Imagine it's not ai, but imagine it's a colleague of yours who's coming that is using ai. Maybe they are replacing you or maybe they are challenging you because maybe they can do the things that you do today, just, in a fraction of the time.

And I think this is the current challenge. So they are really people who are leveraging AI to be like, much more productive. And they are people who just have never, ever heard about that. And I think this huge gap, right? That's really, the thing where people should, worry about 

Loris Marini: it's a fantastic way to put it. And there's always gonna be a resistance right when something is new, especially when it becomes really quickly subject of every conversation. A rapid AI went from zero to a million users in five days. Sorry, Rapid ai. The

Tobias Zwingmann: would be so amazing.

Loris Marini: you , I wish

Tobias Zwingmann: after that podcast, I hope. Yeah.

Loris Marini: Chat, G PT would open AI that is they went from zero to a million users in five days, which is incredible. Wasn't a hundred million. I don't know, like an insane number. And so that, that tells us that there's a lot of interest but also. There is a resistance and I found that resistance myself.

I don't wanna talk about that for a moment because I am in deep in the woods, right? That's what I do in my day job. That's what the podcast is all about. And yet it took me a good four weeks to, from the moment this was becoming hot on social. And I know that, open eyes been working on this for much longer, but the hotness came relatively recently and it took me quite some time to overcome that barrier.

Say, okay, I'm gonna prioritize this. Let's check out chat G B T, what can it actually do? And I think that has to do with just the amount of noise that there is, data ai and that pop culture effect affected you just mentioned. But then when you actually get exposed to it, it's quite powerful.

It's quite interesting in a new way. There's something fundamentally new, which is why then Microsoft invested 10 billion to. Integrate strategy between into their browser and teams and so much more. So this stuff is here to stay and is changing how we interact with machines.

Finally, to be honest, cause, and if you ask me, I I was hoping that this would come at least 10 years ago, but there's a whole bunch of people that are, don't know it. They see it as dis opaque, mysterious black box. And when we don't know something, we don't know how it works. Fear kicks in.

Fear I think is really a consequence of ignorance. So do we need to know how AI works to be able to chill out, relax feedback, and understand that this is a tool that we can use or we can just go ahead and use it and benefit from that top of spike. That comes from realizing that you just saved three hours a day.

Tobias Zwingmann: Yeah, I I think it depends, right? You should at least be able to to navigate that whole field because AI is such a broad term and depending on who you are, like people might understand different things on it. And it is really a, it makes a huge difference whether you train a supervised machine learning model, that has some, predicted outcomes that will be always the same, given some input, they provide you with some output.

Or if you work with a large language model like open ai, like chat bt where you know, you don't have the same outputs, which are not deterministic and. , very basic things like that to understand. I think this is really crucial. I think not everyone needs to know how the transformer architecture needs to work, right?

I think that's like too much, but at least being able to understand how an AI is trained, how these things work conceptually, and just to put that into different categories and also to learn how to interact with these things. Because, in the past, I think there has been a lot of emphasis on training people to build AI services or ML services for that sake, right?

There has been like ton of Udemy courses or tutorials on how to do supervised machine learning and Python or, anything like that. , but we haven't really trained people on how to use those systems. So we expect that, people will just can't figure it out. And I think now we are in the stage where we need to educate people on really how to use those services because there are like tons of AI services out there.

You mentioned chat bt, but like literally every large cloud vendor has their own suite of AI services, which you can just consume out of the box. And in many tools, AI services are just integrated without us knowing it. And I think for our, private life, that's okay if we just, take some photos and faces are recognized automatically on the latest iOS, we can just select some text and copy and paste there.

Most people aren't interested really how that works. It just works. But especially in a business setting, when you have people or teams building those models and other people who need. Take that model and turn it into a business value because, that's what it's all about. I think we need to really educate them and talk about how can we use those tools and how can we use them most effectively for the business.

And this really depends on the type of AI service you're working with, right? Very simple things like, you can't really ask chat g PT for any effects. It just, it will just make some things up. And likewise, if you if you work with a with a model that is predicting, I don't know, sales forecasts.

And so you should be aware that, the longer you predicting into the future, the more unreliable, those predictions will become like, per default. Like very simple things like that. And, sometimes it sounds trivial, but if you see how AI service are used out there in the world, I think okay, maybe we need some kind of like crash course for everyone on how to how to use those systems.

So that, that's why I think it's important right now.

Loris Marini: basic bare minimum data literacy to just be able to understand, can I trust what I'm seeing? Is this something that, and sometimes it's completely opaque, right? Text generation, for example, if you say, if you ask, write me a summary for book about topic X, right?

And chat GT goes off and writes here a little summary and, that could be too totally made up. It could be based on value information. That was in, just in the news yesterday, I believe that Google launched his own version of chat, G B T. And it was a massive, like stock value went down like

Tobias Zwingmann: A massive mess.

Loris Marini: that I don't know how many billion dollars they lost just with that announcement because they didn't check the validity and someone, I think the test question was about a fact in science that happened.

The the, I returned something that com that wasn't true. So anyways, , this is, we don't wanna go back into that rabbit hole, but it is not everything that is generated by an algorithm can be trusted or is valid. And it takes a bit of skill to try and understand, can I trust?

Can I go to the c e o with this or should I double check? What are the checks that I need to do? Because speeding up our work is okay, but getting fired because you recommending the wrong thing is not okay

Tobias Zwingmann: Yeah. 

Loris Marini: right? So there's that piece for sure. So we need to fill the gap.

But when I'm, when I think about that resistance and that fear, I see one thing that freaks me out is you go to the systems and you can type a query like, I don't know, Write me a, a programming ruby that, connects to this API and plots in, Plotly or Redash or whatever and it's incredible how much code the system can generate with commentary and debugging capabilities, which really, you know, from the software and the practitioner side, not so much the user people that are paid and they value themselves based on how good they can, how well they can code, how well they can debug and problem troubleshoot.

Having something that can do that job for you, of course it's not gonna replace you, but it is, it makes you think a little bit about what the hell is going on exactly here, with if my job can be, 80% of my job be replaced by a chat bot. Am I, is my job secure?

Am I gonna get paid? Will I repay my mortgage? This

Tobias Zwingmann: Yeah, there, there was this saying on Twitter, goes something like, the hottest programming language now is English . So it's how can you really, interact with those kind of systems. And honestly, I think, what really surprised a lot of people and including me, is that, The way AI is transforming, like jobs or industry is not as we thought, very repetitive, easy tasks.

It starts with the creative part, it starts with writing code. It starts with things where we thought, okay, maybe these are like the last, areas where AI would have tapped into. But it turns out like, these are the best use cases right now for chat G P T, right? Especially those use cases where there is no really single, true version of that.

Even if you write code, there are multiple ways to approach a problem. 

Loris Marini: I'm sorry. 

Tobias Zwingmann: And this is where chat G P T is pretty good. And the same applies to writing text or writing content. And I really think that, a lot of people, especially who are not native English speakers, like myself, this is such a leverage because you are able to really write text that sounds confident, that sounds good, and some people would have never been able to write that kind of text.

But at the same time that, puts a huge threat to Content being generated because everyone can sound confident, right? No matter if they actually have a clue about what they are, talk what they're talking about or not. And. Even being aware of that fact that we are now living in times where you can generate like really bullshit.

Sorry for that word in, in fractions of a

Loris Marini: Don't be sorry

Tobias Zwingmann: This is so important to understand. And and we are just at the beginning of it, right? It starts now with text. We have seen deep fake videos back then, but, who knows how the next wave will look like. I saw, today I saw David GTA posting some some song of him on TikTok where he imitated the voice of Eminem.

And, it was just like, he did a he did a co-production with Eminem, but then he said, okay, it's all generated by ai. So he just took an AI service and, he took Chei PT in order to write a verse in the, OR LX in the side of Eminem, then took another service to, they would take those verses and put them into that language of Eminem and just produce a song out of that.

So it's, and this is it, it left the tech world, right? So it's just entering all those spaces. But the next stage is you don't need David Geter to produce a song, you can just do that automatically. So where will that end? And just being aware of these things, I think this is really mind blowing and a lot of people are really not aware of that.

And I don't know where this will be in like two years or so from now yeah.

Loris Marini: Yeah, I dunno where we're going, but I know that having a trusted source of information for anything, whether it is, the news or education and mentorship in especially in in technical roles is gonna be more and more important because, otherwise we might as well just hire bots, right?

instead of human beings. That's one piece. But also, and I think that's what the part that excites me the most is the fact that we always talk about, on this podcast about bridging the gap between data and business. And for a quite some time we data practitioners focused more on technologies trying to.

Get up to speed and know the latest tools and make sure that you know the latest the latest technology. He's on our resume and we know how to use it. So the next interview, when we leave this ship and we jump to the next one we can sell ourselves better. We can market ourselves a little better and get paid a little bit more, which is totally fair, right?

That's the game we've been playing in a reality, in a world where it doesn't really matter because you, there's physically no way that you can follow. The evolution of something like a bot, like a, and we agree that takes a lot of energy to program these things. They hire developers to be able to teach tr how to write, code and how to debug it so it doesn't come for free.

There's a cost, but the thing is, once you've done it you only have to do it once. Then you deploy their capability and now everybody in the world with a URL can access that level of actionable intelligence. So there's no way that we can keep to, I think we should play another game here. Not try to catch up with these things, but try to use them knowing that we'll never fully understand them, but we need to understand them enough so that we can bring their value back to the business and invest in those human.

Skills to know, for example, how to have executive presence, how to speak the language of a stakeholders, understanding what their pain points are, what keeps them up at night and how to just shape everything we do and say so that it's optimally received by these people. No, it doesn't sound cool for us.

Maybe it's not something that we can go home and boast about, but it is effective because it's helping those stakeholders overcome some challenges. So it's a long way to introduce the first really book related question, in what way in practice have you seen ai generating business?

Tobias Zwingmann: Yeah, so there are many ways actually, and let's start with the most obvious one. When people think about AI in business, they think about, predicting things like, forecasting the future, predictive analytics, prescriptive analytics. And that's obviously, where AI has been used like for a lot of use cases.

But at the same time many businesses, if not most businesses, if they look at their analytical maturity, they think they are not there yet. And maybe if you remember, there's this data hierarchy or data pyramid of needs where it says, need to have like clean data and then transform data.

You have bi monitor, and then at the top there's AI somewhere. And honestly, I think, this is right when it comes to building your own AI service, but at the same time, it left a. People, just relax. Like we don't have to care about AI because like we have that kind of like crappy data, right?

So we will never get there to build our own AI systems. But this is actually wrong because the way to think about that is that AI can actually help you to achieve those steps. Like AI can literally be used everywhere. I give you a simple example because some people ask me, how can you use ai, for example, for descriptive analytics?

But then I say, you can use AI, for example, to process your PDF documents. If you have thousands of PDF documents, you can just throw that into an AI service and get insights from that. Or not, maybe not even insights in the first stage, but just get, extract the information in a tabular structured form in order to report that in your BI system.

So that's the very first step. And no one can say we have crap data. That doesn't work. Like it's just PDF documents and you can extract that from there. Or take another example. If you work with modern BI tools, for example, like Power bi, there are like plenty of tools built in that let you interact with.

Data in a much more seamless way. There's the key influencer tool, for example, I'm covering my book or the q and a visual, which let's like interact with your data using a like q and A interface, like natural language. And these things are there, right? So you can just use them. And that's where I think this is where a lot of, business value is already being realized.

But a lot of people, especially in larger corporates and especially people in the BI world, think of, ai, that's not for me, that's for the data scientists. That's the fancy stuff. And maybe we will get there like in five or 10 years, but I think this is really a dangerous. thinking because it's here, right?

It's here. You can paste your SQL schema from your SQL data warehouse into chat G B T and say write me a query to get the top-rated product and we'll write the query based on the schema that you provided. So this is and this is happening right now. And

Loris Marini: good is.

Tobias Zwingmann: yeah.

And I think more people need to be aware of that. I need to acknowledge that they can actually be more productive right now using AI. And yeah. And this is, I think where the real value lies, where you have those kind of like low hanging fruits in order to materialize those gains.

Loris Marini: Do you reckon that there are also, there could be some risks? By sharing, because a, if we talk about chat, G B T for example, is hosted somewhere in some servers that someone else controls. And our open AI has the keys to those servers. So when we, every query, we send every piece of information, it's gonna be stored somewhere.

Of course if we just shared the schema, there's no business value in just the schema, right? Because again, you can model a piece of data or business process in a billion different ways. Should we be careful about that?

Tobias Zwingmann: Yeah, definitely. That's what I mean with ai, educational AI literacy. So don't post any secret board meeting notes in there. Don't post any email conversations in there. Don't post any actual data in there. So you have to be like really careful about that and, but I think, honestly, I. , this will just be a matter of time until these things will be eventually solved.

Think back of the early days of cloud windows and cloud adoptions where people said, oh, we will never upload our data into the cloud. And now, like even the most heavy regulated industries are all using cloud services because it came with such an like regulatory framework that it's now possible to use that.

And just to be clear for enterprises it's not about, what happens actually with the data and where is it stored? What they need in the first place is they need the guarantee, they need the frameworks, the contracts, that data is processed in a certain way and stored, for example, in service within the eu.

I totally you believe that this will be coming chat, G p t or open AI services are integrated or will be integrated into Azure. And once they integrated into Azure, they'll, it'll be just like a switch of a button in order to, run everything in a hosted environment or just in a certain area of the world.

So that will be the natural next step. And then you will you will have some, like contracts and agreements, you just click and sign and then, you're basically done with that. But still, you have to live with the fact that, your data could be used for training an AI model, which you have never access to.

On the other hand, what's the solution then? I think most companies, even large, even really large companies 

Loris Marini: Mm-hmm. 

Tobias Zwingmann: will never be able to train these models from scratch, on the by, by themselves. I think this will never happen. There will be a small group of companies and you can probably count them on two hands who are able.

and who are focused on, building these large models and also deploying and maintaining them because building them is one thing. But, having the infrastructure to run these things also in a way that users can actually, interact with them really fast and in real time. That's a different kind of story.

And for even for a very large corporations, I think that's just on a good business case to, to make that investments and, hire and build the whole AI that is just doing that. So I think what we will be seeing is, there will be like large pre-train models. Companies are going to fine tune that on their own data and they will own that fine tuning process and, apply that to their business critical data.

Loris Marini: Exactly. So the asset becomes the ability of modeling the algorithm to your own use cases as opposed to owning the capability the algorithm itself.

Tobias Zwingmann: Yeah, exactly. Training the algorithm essentially with your own data, right on your own service, with your own platform. That will be a critical asset. And this is something you can do right now already. Just with regards to large language models, still very often that you need to send that to a remote API and do the fine tuning, for example, with open ai on their server.

But again, I think that's a matter of time until these things will be just enter the corporate world and enter the whole framework that comes with it. 

Loris Marini: For a long time I've been a skeptic about, the ability of generative AI to replace the role of taxonomies and ontologist. And we've done a few episodes here on the podcast with a number of subject matter experts on what does it take to create a knowledge graph. Like how do you turn a bunch of things in people's heads into a reliable source of information for the enterprise, the scales, and the moves at the speed of business.

And this is an open question, right? There's billions of dollars being invested to create data catalogs and data tagging systems and metadata management solutions, cuz that without context, right? And I always thought that context must come from human beings, someone that has that domain knowledge, that knows what matters and what doesn't.

So in other words, it's able to, or capable of prioritizing based on the context of that particular business. But then just recently I found I think something that potentially can change disbelief. And is the graph G p t, not chat g p t, but graph g p T. There's a GitHub repository that was created literally last week.

Is the 10th of February now that we are recording this one. And this is appeared a week ago. It's got already two and a half thousand stars in K dab. It's been forked a whole bunch of times, and what the author of this one is showing basically is just a way to convert natural language into.

Entities and relationships, which is what a knowledge graph , is. Now imagine if you're able to take all the unstructured texts that floats around the enterprise, throw it into a well governed, completely private environment where this stuff runs and get back a graph that literally you can visualize and navigate, explore that tells you what are the keywords that people use, what seems to be more important, what is related to, what, the relationships between entities.

That would be incredibly powerful. Of course, there should be a human oversight cuz we can't expect to be a hundred percent accurate. But, just the fact that these things are possible, it means that we are going to, we're going to save a whole hip of time,

Tobias Zwingmann: Yeah. And the large companies are working on that. The other day I heard a talk from a senior executive from sap, who said that they are evaluating or building a service using a large language model to model business processes because they have all the data for how business processes work.

They are business process experts. And if you could just dump your whole business process model into a large language model and. Please find redundancies, how can I improve my process model in organization? And it will just come out like, in fraction of a time. And before that would have taken you, I don't know, a 10 year project, right?

In order to look at all these things. And I think there are so many things, still around the corner, which we like would never think about, but being able to extract this kind of structured information from large corporates of text data, and this is a real game changer for so many applications.

And yeah, 

Loris Marini: Yeah. 

Tobias Zwingmann: honestly, I don't know what, I don't know what will be next? I don't know. The only thing I know, things will be really happening fast from now. I'm pretty sure about that. So there's so much pace right now in the industry and there's so much awareness on that. There's also a lot of capital right now because everyone has dumbed the Metaverse, and the nft.

So now capital's going to reinvest into ai but that's it. That's where the focus is right now. Four years ago people, we are not really clear what's the next big thing? Is it ai? Is it the metaverse? Is it NFTs? Right now it's clear, it's ai. The whole focus and attention is on that.

And yeah

and, 

Loris Marini: Yeah. Speaking of this, I just wanna remind our listeners that you do run a, an amazing newsletter. It's all you talk about there. So if you guys wanna know, the latest what's happening in the industry and get a, in an informed angle from someone that actually knows what it's talking about.

AI four bi.rocks. The four is the actual number. So AI four, the number four BI Rocks. We'll have those links in the show notes as well, but definitely recommend the subscribe to Tobias Newsletter. Sorry for the plug. 

Tobias Zwingmann: Thank Thank you

Loris Marini: I think it's something that is really worthwhile having in your inbox.

That's why sharing it. Tobias, back to the book. We talked about some of the ways AI is generating business value. What are the tools as opposed, or the platforms what's our first step? So say that I'm super keen, I'm like, okay, that's it, now I was a skeptical, now I'm converted.

Now I want to get from zero to proficient in this things. Of course, step number one is buy the book or get a copy or somehow read it. But for those that are listening to us right now, and they, what a quick, something quick that they can do right now in the next 10 minutes to get that document spike.

Where do you get your document spikes?

Tobias Zwingmann: Yeah, I think two components. The first component is AI services. So don't try to reinvent when deve, from scratch. So use services. This could be an like image recognition service. This could be an auto ML service, so tech services. And the second one is adopt the mindset of a prototyping mentality.

Especially with AI projects or machine learning projects, you never really know if the outcome will really, be the case if it'll really work out. So the worst thing you could do is you knows, Have a great use case in mind. I think it'll be amazing, set up a huge project plan for three years, hire a bunch of consultants, go ahead and after three years you find out that the data is crap also.

I don't know. And that's how you, not how you should not treat it. And the much better way is to really keep things small at the beginning and, build a first prototype and check feasibility and check if your data is actually in the right capacity for using those services. For example, if you have, if you want to run an auto mail task, you need to check if you have access to the data.

If you know you have enough data for training the models. If you want to use image recognition services, you need to check if, I don't know the images have the right size or if the objects you want to like, , tagged on those images can actually be detected by the AI service. And these are really things you can iterate over so quickly and quiz quickly.

Sometimes you can do that within hours days, but as long as you need to a couple of weeks to figure that out. And I think if you have that prototyping or iterative of mentality, then you are also able to quickly go through a list of use cases and just check them off and see what are things which you can really implement fast.

Because there's a lot of talking about. What's the feasibility of that use case versus the feasibility of Yeah, that use case or the impact of the, that use case and like their companies spending, I don't know, a month of, discussion around which use can, should we prioritize in order to, prepare those use cases for long-term projects.

But the way I think about it is, just forget about the moment for that really long-term planning. Try to realize what you can do right now. Because in the past years, the companies have been obsessed about their, like big AI use cases. They hired data science teams. They hired data engineers.

They built platforms in order to really facilitate those high value or high impact use cases, which also needed high investments. But the average company, at least that's my opinion or my experience there, is only, they can only really find or handle maybe a handful of those high value use cases.

But in the long tail, there are dozens, if not hundreds of use cases available, which maybe not have that huge impact, but you can. But in the sum, like they don't have that huge impact like on their own, but in the sum of them all, it is a huge impact. And these are use cases like just replacing that trend line forecast in your bi, which you hit by like right click and then, draw a trend line within.

ML model, for example, that is by producing a better forecast, which is maybe not so critical, but for the people using that, it might be critical in that case. Or for example take literally any business process where you want to get faster to insights or you went where you want to get a better prediction or a use case.

Like I covered that in my newsletter recently. Lead scoring for example, where you say you have a bunch of leads and your sales people need to call them. So the questions like, which order are they going to call them? And this is not a huge and complex use case, and for a lot of centralized data departments, this use case will always fall behind in the backlog because it's not relevant enough.

But if you have someone, for example, in the BI team who is able to, apply an automobile service and who has the data in the data warehouse anyway, they can just do that. And modern data platforms allow you to do those ML applications even in the data warehouse. If you use like platforms like BigQuery or so you can do that directly.

In the data warehouse with a SQL statement. So you don't even need software around that. But even if you use Power BI or anything else, like all these modern data tools, they have capabilities that let you integrate these these systems and services. And obviously in my newsletter I cover a lot the Microsoft stack, but there are plenty of other examples.

Where these things work out as well. There are tons of no-code tools available. So really technology is not an issue anymore. So no one can say, Hey, I don't know how to do Python programming, right? So ML is not for me. That's not how it works. It like, it's everywhere. And even if you don't want a programming can start, can get started with that.

But what you need, and I think this is really critical skill that you need to build over time, is you have to get that experience and as you say, the context around the data. . A lot of organizations are not aware that they have those kind of data experts in their own house, but it's, they're not aware.

Those people could actually be upskilled to do those more advanced analytical tasks, because they think, Hey, like this guy is just doing the reports, right? Or this guy is just doing the modeling on in the data warehouse, right? But now we need the data scientists in order to like, take that to the next level.

But no, that's not how it works. 

Loris Marini: really. 

Tobias Zwingmann: And this is why I really think, like looking for that talent that or people who have that long-term experience with data who know the pitfalls and who also know the connection to the business. I think this is really the best area to to look at, are the best people to focus on when it comes to realizing those like potentially low value use cases.

Because there are really plenty of them. And

Loris Marini: is a good message for a head of digital transformation. If there's someone in the audience now listening to this definitely take notes of that because I strongly believe that is the way to get the biggest bang for the back, for the business.

and they're not in opposition, right? You can have data scientists and you should probably plan to hire data team if you wanna get more serious and get better at these things. But the quickest thing you can do right now is leverage the domain expertise that is already in the organization.

Find those particularly curious individuals that wouldn't mind to spend a couple weeks on a bootcamp or like a crash course and get up to speed with this stuff and supercharge them, literally supercharge them.

Tobias Zwingmann: and it's a very simple calculation. If you have a data team, right? If a data scientist, I dunno, it costs you $150,000 per year and you have five people sitting there, you can't let them work on use cases that, bring the business $20,000 per year, right? It's just not no nori.

So that's why you need to look for those use cases where four percentage point increase give you millions of return and especially for large companies, there are those use cases, but again, these kind of use cases are limited. Overall amount. So what do you do with all those use cases that deliver maybe $50,000 per year or $40,000 per year, right?

Where you think, ah, that's not enough really, to set up on your whole project for that, and you have all the maintenance on the model. So that's the point where you really need to upskill people who are already working on those problems and just try to push the baselines of doing things right now a little bit further in order to, realize the value.

Loris Marini: Another point there to us is the value curve. I remember the Brenda Dykes is one that I remember, but there's plenty of these ones. But Brent Dykes was he. He's been here in the podcast in the past, we talked about data storytelling in one of his posts on LinkedIn, and he's a Forbes contributor.

So he constantly writes about, value and data strategy, and he's got that graph that, you have just two access on the vertical accesses business value. And all the horizontal access there is time and when you start with your foundational activities like data management, data access, setting up an environment just to be able to tap into the data in the first place, all of that is negative value, right?

You're investing, you're not really getting any, anything out of it. You're just preparing to be able then to leverage that asset. But then there's a crossover point on that time. And that's when you start interrogating the data. You start building your first reports. The actual real value is much according to that graph is much later down the track when you are able to do advanced analytics, advanced ai, machine learning, all that stuff.

Now what's happening is that we are compressing the second half of that graph. Instead of waiting for such a long time, we can get, we can deploy if the data is available, if it's, if we can use it and it's not siloed in a billion different places, we can deploy this stuff. So what I'm thinking is this could be part of the storytelling for, who is sitting in that strategic data role that is trying to motivate the business and get them interested in investing in this stuff.

Is that now the time to value is shrunk. We still need to do that foundational work because if, half the information is in an Oracle database and the other half in the SQL database and they don't talk to one another, there are, there's no way to even join those tables. And we always gonna have partial information, right?

So that's data solid. So we need to fix those problems. But, once we collect that data together into one place, from there to insights, we can do it so much

Tobias Zwingmann: And and this is where prototyping I'm completely agree, and this is where prototyping comes into place when you can prove. At, for example, bringing those data sources together back from, for example, taking a dump of them and just trying to bring them together and build something like, on the greenfield ex, like outside the existing infrastructure.

But if you can prove that you're able to integrate these sources and that you can also go the last mile of doing advanced analytics of, predicting things or prescribing things, if you can prove that in the prototyping phase you have a much better business case of all to, try convince people to really build a project and integrate those data sources.

Because otherwise, it's always very hard to describe what would be the outcome of those big initiatives of, integrating different silos, for example. But once you can prove the value, and once you have build a prototype, it's much easier to showcase that. So that, that's why I really think, Trying to get hands-on and trying to really build things, even if it's on a small scale.

This can be so valuable in different cases, right? It will upscale yourself. You will learn new things, but it will also make things much easier when you want to set up a bigger project in the organization.

Loris Marini: Yeah, the one suggestion I feel like giving to to the, those that are listening to us and thinking, oh, great idea. I should go to my database administrator and maybe buy them a beer or offer lunch and get them to do a data dump over the last 30 days so I can build a prototype. It is a good idea.

Just make sure you have a sponsor. Someone that it's not gonna be completely upset when they realize that you are doing something that they should have done. , it's just a small tip around organizational,

Tobias Zwingmann: it under the radar.

Loris Marini: Yeah, exactly. You won't definitely have a sponsor. It doesn't mean you need, you are, you're not asking for money.

You're asking for someone that can give you a green light and say, Hey, you know what? Yeah. If they, if you find resistance, forward an email to him to me and I'll I'll get in touch and I'll make sure you get a read only access to that database. Next step, dump into Excel. Connect the Power BI system.

Do some, really easy, analysis using all the AI you can use. And if you find an opportunity there, that's when you're gonna print it on it, one single slide with a compelling business need, and you're gonna pitch it, and that, that could be their way to get getting the business interested and getting even funding to make sure that you don't have to go to that friend and buy lunch to get your data.

But maybe you have a more, approved way of getting the data. . For sure. But let let's go back to the book because I think I'm diverging a little bit here, speaking more about my own challenges than

Tobias Zwingmann: Yeah, but these things are real, right? 

Loris Marini: 10. 

Tobias Zwingmann: These are things which are happening. So yeah, I can completely relate to that

Loris Marini: Yeah, yeah, absolutely. Yeah. every day. Making my job very interesting. But I was thinking that of all the chapters in your book, that's probably the one that caught my attention the most is the piece where you bring all the pieces together in chapter 10 you really guide the reader. Into building an AI powered customer analytics dashboard, and you make that story of that telecommunication company with the problem of churn.

Like what was happening there? Because I asked you this question because a lot of these talks can seem, disconnected. So I really wanna do, talk about the story there and the use case in a way that, and I'm sure a lot of people will resonate with that. But I wanna hear from you like what was happening in that company?

What was the problem that they were facing, and how did AI help them?

Tobias Zwingmann: Yeah, the problem Ross, really a to understand. , why churn is happening, B, what to do about that. And c also try to bring churn together with a business metric because the thing is like a churn number alone. It's very hard to quantify. Like what I did here in the book was to showcase how that company brought bi classical, bi data revenues, for example, from like from orders monthly subscription values together with a calculated churn risk.

So you could actually not only say what's the percentage that customers will churn, but what's the revenue that is at risk for the next month, for the next quarter. So to make management aware, The size of the problem because like 20% customer churn, like it might be bad, but it's just 20% customer churn in the sea customer segment, where people just pay like $9 a month and there will be new people coming in anyway.

It's not really that of a big problem, but if you have 20% customer churn and you know it, has an impact of your A and B customers, maybe, or, it's a six figure number we are looking at here. The problem obviously is gets much higher awareness. That, that was the first component of bringing like BI and AI together in that case.

And then also what we did here was analyzing information from from from user surveys where people, were asked about the company and there were a lot of free text entries and we used that to analyze the sentiment and find out what were things that comp that the customers were actually complaining about.

And in this particular case also relating that to BI data to see what were the. Kind of customer types and customer segments who complained the most actually. So that was the other component to understand like why churn is happening and really to understand why in order to what are the driving factors here in order to, mitigate those churn and risk.

So yeah, not only why in terms of because you could also say why people are chatting here. They just joined last month. Probability of the counseling is much higher if you joined last month than two years ago. But that's not really helpful for the business because that's just, always there.

Yeah. It's not actionable. But if, for example, that people complaining about not having the desired network quality that they have or that there's slow speed on their network, these things which are more actionable and these things are typically, things you don't find in your in your data, right?

So you have to ask customers and this feedback is typically unstructured. And yeah. It's funny actually because tomorrow I'll be covering you said how to analyze that feedback now with a large language model because back in the book I used the sentiment analysis service from Microsoft for that.

But now we have chat p t or open ai. We can basically just All the data and say, what are the most critical pain points here? Of course, like if this was a real data set for, that new study example, there would be like, tons of things going on. But let's imagine you are allowed to use that data and send it to a large language model.

It's much easier to get the main messages across and see what people are actually saying and complaining about, for example. Yeah, but in this case, here, in that dashboard, we really analyze, like each single entry individually and try to identify the sentiment score of that and figure out what the key challenges are that customers have and make them churn.

Loris Marini: Yeah, I love it. I love it. Your book is literally on the top or the top of the stack, and I just rearranged the priorities there because I have so many books that I wanna read. But, AI powered Business Intelligence is, I think it's a must-have. I must have anyone that has an interest in data, whether you are a data professional or not, you should read this book and actually follow the step to step examples.

Cause I was flicking just through the book, and you've got, apart from, the graphics and the schematics that are really nice to have and help you, follow the book, where you literally have screenshots of step by step, how to set up an environment, how to load data what, you taking people to read their by hand

Tobias Zwingmann: Yeah.

Loris Marini: so 

Tobias Zwingmann: Yeah. My goal was, my, my goal was just like, also people who read the book that they see how these things look like, because, oftentimes these things sound so complicated, right? Azure cognitive service for sentiment analysis. If you tell that to typical people they feel just oh, that sounds way too complicated.

But then if you see, okay, it's actually very simple, so you just have to copy that API key and paste it over there. That's it. And I think, look, we really, one of the key challenges is to make people understand what the data must look like in order to, be processed efficiently by those AI services.

So for example, in my case in the book, I'm introducing that concept, which is, Ultimately popular in the data world, which is called tidy data. You have one row, which is one observation and one column, which is one variable and each sells value. But to a lot of business users this is not so intuitive even for a lot of bi people who have different data models in their relational cubes.

This is not so intuitive. But like this is the average, or the standard way you should organize or arrange your data, especially if you want to work with those ML services, right? Because that's the format they expect. And like very basic things like that, can help you to get started with that much easier.

Loris Marini: Back to data management, baby

Tobias Zwingmann: back to data manager. Exactly. Yeah. But that's, and, but that's a critical part really, because if your data's not in the right shape, even the best automated service, will never be able to deal with that. So it really needs to be in the shape that these services.

Loris Marini: So this is really a fundamental point. It's worth reiterating. So in your explorative data analysis, yes, you can have all the fun in the world with your ai stuck. You might even be able to manually clean that data set to put it in a tabular format. Do a bit of a, prove a concept and pitch it to someone that can take big stakes decision in your organization and get approved for the new initiative, whatever it is.

Don't forget the data management part because the last thing you wanna do is promise the world show, show the dangle the sugar in front of them. Hey, this is a candy. Look at it's delicious. You want it? And they say yes. But then you can't deliver it a scale because you are, you're not ready.

And of, I'm not saying that you should stop and just clean data for the sake of cleaning data, but you should integrate that piece into your pitch and say, Hey, we want this. This is completely attainable. We can do it. Provide it that we have the data in the structure. Now getting the data into the structure is a work stream.

Let's plan the talent. Let's put some money in. Let's do it. You don't have to boil the ocean. You can do it step by step, use case by use case, but you've got to have that in your

Tobias Zwingmann: Yep, a hundred percent. You need to have that plan before

Loris Marini: yeah, cuz really I don't want to see you getting fired because maybe you were an extremely talented storyteller. You promised the moon and then you can't deliver because you totally forget about data management. So don't do that

Tobias Zwingmann: also, these things keep you busy, like even after production, especially after production, so that's also a lot of things like very often centralized data teams, faced with they have, started all those initiatives, prototypes, put things into production, but now they can't deliver anything more because they're just focused on these things, trying to keep them up.

But they have raised the expectations of being the data guys. And everyone can just like, throw things over the fence and they'll just care about that. But that does not scale, right? There is a point in the organization where this does not work again. Yeah, also make sure that you, who will be the owner and who will be in charge and who will be maintaining these things.

Because any ML service that you or any AI or ML service that you deploy needs to be taken care of. So there needs to be someone who does the maintenance and this is yeah, should be considered right up front.

Loris Marini: Yeah. And for once these people might actually have be able to communicate with the business their actual value a lot easier than what it used to be. database administrator, what do you do? You talk to databases. Okay, so what, , but now it's okay, you don't do your job, but data r ai piece is gonna be crap.

It's gonna be bs, not AI so he's your roi. He's the, he's is the compelling business need for having you on payroll, right?

Tobias Zwingmann: Yeah.

Loris Marini: Yeah, definitely. Just a reminder to, again, our listeners, if you are into this type of stuff and you are like me, you can't wait to jump on the book. We are giving away a copy today@thiscoin.com slash ai four bi.

It's just the same as the newsletter ai, then the number four, and then bi. There's the usual book giveaway form. You can subscribe there. We are gonna close it after two weeks from the moment here, this episode airs, so don't try and do it six months down the track because it's not gonna, it's gonna work for you.

But if you do it in for within the first two weeks, you will enter the list and we'll select, randomly select someone that wins it and that someone, it might be, all the links as usual in the show notes. Just expand that one. You'll find re resources, references to most of the things that Tobias mentioned today.

Tobias, I wanna spend the last few minutes with you to, with an I on the gap that we are seeing at the moment. We are not ready for this clearly, like this stuff is moving at the speed of light and we had a problem with data literacy before. Now we just had a, a hundred x size problem.

What are we gonna do to bridges gaps and get pretty much everybody, every data over? It, it kept with enough knowledge to become a data.

Tobias Zwingmann: Yeah I'm not sure if I catch the last one if I called the last one correctly, but we need to start with the leaders. We need to start with leadership in companies because in my opinion, they're, if there's any company out there who still thinks they don't need data, or maybe AI is not for them because they are just selling shoes.

Building houses or whatever, doing any techn, non-technical stuff, that's not how it works. The most or the vast majority of jobs out there, right? Is right now is knowledge work, it's services and these are the areas which are impacted by AI the most. So we have really to start with leadership and make leadership aware of those capabilities and what it means for their business because analytics and AI is nothing that can grow from the bottom.

If you have a bottom movement, it'll always be killed by the top. It needs to come from the top. It needs to be like a top down initiative. And so that's why I think we need to start with leadership really right now and enable them to at least understand what's going on and how that might impact their business.

And how to create a roadmap in order. Bring all people on the right track for that. Whatever that might mean. There is not really a general recipe for every business out there where I say everyone needs to take this or that course. But business leaders need to be aware of those things happening around them and then need to make an educated decision on what to do for their business.

If they decide to do nothing, that's fine, but that should be a conscious decision. It's not ah, we don't care. Just that, do what we have ever done. Maybe that might work out for, I don't know Monopoli or so. I have no idea. But for the vast majority of companies out there, this will not work out.

So yeah, that's my take. We have to start with leadership in that regard.

Loris Marini: Absolutely. And so that's really the tagline for us, and this has been changing obviously in the past few months as we refine a little bit more, what we're trying to do here at Discovery Data. But I think the, that one sentence crystallizes really well, and I think you understood it correctly.

It's, we are all about turning those data. Passionate people did a lovers, and these people are everywhere in the organization. They don't have necessarily data in their title. They could be in sales, they could be in process or procurement or in customer relationship. People that are just have an interest for data and they are the business stakeholder, the number one consumer of those insights.

They crave them. They need them for their job, they need 'em to be effective to reach their targets and get promoted. So it's not an optional, it's a must have. The problem is they. Really know what it takes because it's not their job. There's always been a data team. If you're lucky you have one worrying about this stuff.

Now what if we flip that on its head and we let these people, the data passionate, the data lovers the ones that are on the line that own the business metric and they wanna deliver and turn them into data leaders, not just leaders, but data leaders. So they, and know enough to influence and to find those partnerships.

Cause the data, people are already doing it. That's their job. But it's always a monodirectional effort, I feel, from the data team trying to find your stakeholder, trying to engage them, trying to explain how you can help them. Why does it have to be one way only? Can it be two ways simultaneously meeting somewhere in the middle?

And that's, yeah, that's something I'd love to explore in the future.

Tobias Zwingmann: to hooking on that in the middle there will be a new role, which will be critical, which is a translator role. So we need people who are able really to really focus on that and bring those two worlds together. So I think this old model of having the business and the data side, and then they need to talk to each other.

If you wait for these things to just happen or, figure out this will take too long. You need to implement a layer in between, which is able to speak both languages and to, translate between those. So I think this is such a critical role, especially for companies who are very early in their overall analytical maturity.

And these people. That's what I. Seen can really be a huge catalyst for accelerating the whole journey that these organizations take. Because they understand, not they are not experts in both worlds, but they speak they speak the languages of both worlds and they are able to bring those needs together.

And that's why I think we'll also be critical here.

Loris Marini: Yeah. Think about what happens when you go to a hardware store and you have a problem. It was just yesterday happened to me, right? Go there. Do you find a person who is incredibly expert of the store? You know that they know every tool, every screw, every adapter, everything in the store. You go there with a problem.

That's your need your compelling business need. I have to fix this. I have to, whatever fix. And you describe it to them and they're able to then immediately go oh yeah, come with me. It's i l 49. We're gonna find all the screws and other things, and here's how you put it together, right?

That's the expert, that's the domain expert. And that could be the data person or the translator, knowing enough of this AI tools to be fluent and quickly put something together. So we don't necessarily expect our C M O to be able to do it, but the CMO should, at the very least, be aware of what can be done, who to tap into, and if they can't find the talent.

Maybe do it themselves. What's a shortcut that you can take to go back to the next board meeting and go Hey, I actually did this little exploring thing, and I found this opportunity , it could be you seriously, with enough, with just the bare minimum knowledge you could be that hero in that room.

Yeah, I'm just excited about creating those

Tobias Zwingmann: Yeah, totally agree. I totally agree.

Loris Marini: for all these people. Alrighty, cool. So this is so we talked about many things, but one thing we haven't even mentioned is rapid.ai. What are you guys doing there? What's happening? Uh, 

Tobias Zwingmann: You are basically helping companies to adopt ai. And this involves many things. Obviously we're focused on building prototypes building early use cases, and bringing ideas from PowerPoint into the first production, right? The first real thing. And this spans especially areas of image recognition, but also most recently applying natural language processing and especially large language models.

So fine tuning models building custom models for clients across various industries. Yeah, but also always with the focus really on the first prototyping stages and really trying to do that real quick. So we have different templates. We have our own platform and order to facilitate these things.

So yeah, that's what we like and what we.

Loris Marini: Yeah. Interesting. Yeah, I bet you're gonna be super busy going forward because that type of help is yeah this is big need for the, for those skills. So I'm glad that someone is actually doing it. So discover data com first slash ai for bi again, for your book away. And check out the newsletter, not check out, subscribe.

Now immediately, what are you doing? Like you're wasting your time. Just subscribe AI for by adult rocks, do yourself a favor and get ahead and to buy. I think that's, that wraps it up. And I really every time I talk to you, I'm like, , I come back with that call with a billion ideas.

So we should find a way to, to have more frequent chats and implement some of these ideas sometime

Tobias Zwingmann: Definitely I would love to do that. Yeah, thank you so much for having me. It was really my pleasure. Thanks a lot for the invitation. I was really great chatting with you again.

Loris Marini: No worries, man. I'll see you in Germany sometime soon when when I get there. Are you in

Tobias Zwingmann: Yeah. Yeah. Germany. Yeah. Feel free to drop right.

Loris Marini: Yeah. Yeah. Definitely. I'm to feel like a vice beer and some

Tobias Zwingmann: Oh yeah. It's always good.

Loris Marini: It's always good. It's a great breakfast,

Tobias Zwingmann: it's, it really is

Loris Marini: It really is. I

Tobias Zwingmann: after a L, especially after a long night. Yeah, but that's a different story. Yeah.

Loris Marini: Yeah. Now, for instance, it's what, 7:47 AM here in Sydney now , that could be a great way to start a Friday,

Tobias Zwingmann: You can go directly to sleep after that breakfast.

Loris Marini: Yeah, exactly. Just awesome man. I'll speak soon. And thanks again for for being here with

Tobias Zwingmann: Thank you very much for having me. Thanks.