How to Build a Data Foundation for the AI Era – with Telmo Silva, ClicData

00:00:00: But at some point in time, we may say well... ...we really no longer need a data flow or connector property screen.

00:00:07: Or even dashboard designer's screen.

00:00:09: We might just have AI manage that piece but the functionality and engine behind-the scenes is still us.

00:00:27: Hello & Welcome to The Data Culture Podcast.

00:00:29: I'm Karsten Wanne founder and CEO of BARC And my guest today is Telmo Silver a veteran in the data and analytics industry.

00:00:39: He started ClickData eight years ago,

00:00:43: And

00:00:43: we get the chance to participate from a lot of knowledge and insights that terminal gathered over time.

00:00:52: For example We talk about these contradicting approaches to data management on one side agility speed On other side.

00:00:59: we need also quality and reliability from Telmo, for example.

00:01:06: Why it's better to embrace than fight data silos

00:01:11: or

00:01:11: when a semantic layer is the good semantic layer?

00:01:14: Lots of interesting stuff in here.

00:01:16: enjoy episode.

00:01:17: Hi Telmo!

00:01:18: Great to have you on The Data Culture podcast.

00:01:21: Thank You Karsten For having me here.

00:01:23: Pleasure

00:01:24: Telmo.

00:01:26: When your started at your company ClickData Data Analytics BI platform of developing the software before starting to sell it, how did that happen?

00:01:38: or why?

00:01:39: Why didn't you do that.

00:01:42: It was back then almost a pipe dream.

00:01:46: I came from a background of corporate BI data warehousing CRM ERP installations.

00:01:53: That's my background and the latest projects i was working on Was really trying to embrace a data culture within the company.

00:02:02: one of the board members was an amazing person, very visionary.

00:02:06: Which kind of launched this idea?

00:02:10: thinking about data differently even monetizing data a lot of things which in that sector... ...which at the time was pharma....was even unheard us right?

00:02:20: And and that propulsed me when I left that company to think you know i had so many resources!

00:02:28: A lot of consulting companies helping me.

00:02:31: I had a substantial budget and for databases, for data warehouses or software etc.

00:02:39: And I have internal amazing internal teams around the world... ...and yet i felt when i left that we did not succeed.

00:02:46: We didn't finish that vision that board member has for this company.

00:02:53: That felt like maybe i could do something here But i was very well aware of who I was competing with.

00:03:01: I was computing with, you know, macro strategies and business objects and Microsoft's BI at the time and Borekals and Business Intelligence so i knew what I was dealing with.

00:03:13: then I go tell them how could you build something?

00:03:17: And more importantly when I came down to it said this is not just about charts and KPIs in dashboards.

00:03:25: that piece is the easy piece, believe it or not.

00:03:28: Everybody can do it.

00:03:30: PowerPoint and Excel can do amazingly.

00:03:33: where's the gain for their company?

00:03:36: And second question to me was how if I'm able to solve this who could benefit from it?

00:03:42: right what type of company could benefit From It!

00:03:46: That started thinking about building.

00:03:50: yes we will have dashboards But because that's the end product for some customers, but we need to build a whole piece.

00:03:59: The connectivity of data and transformation in the data.

00:04:07: And yes, visualization is why it took me five years.

00:04:11: I had to build this thing with full end-to-end products which was useful for people who thought could benefit from small or medium mid-market companies, that did not have the resources I had.

00:04:25: That didn't have an amazing budget and knowledge sometimes to give this power for these companies themselves.

00:04:37: He coincided with the big cloud movement in the year of five or two thousand eight.

00:04:43: Salesforce was already cloud.

00:04:46: SAP was thinking about going cloud so was being very well adopted and accepted by business in corporate culture.

00:04:54: And we said, Well, We also have to do this a hundred percent cloud.

00:04:57: so because of those challenges Yes he did take his time To build A product that Was a minimal viable Product if you will or proper data management and analytics which encompasses all the different stages Of from data acquisition All The way to Data restitution Or data visualization or data analysis.

00:05:17: So Yeah, it took me a long time.

00:05:20: It took the team for a very good team but today our product even supersedes that and I think that's kind of... ...the pleasure i took out of it was really trying to solve problems for customers That had experience in the corporate life.

00:05:41: When when you talk about it that just said hey dashboards and reports.

00:05:45: That's the easy part, but do spend a lot of time on them or data management sort things.

00:05:50: I was just wondering whether we are in this same situation again could be say that its for let us say LLM which obviously now they knew kids under block.

00:05:59: that is probably fairly easy.

00:06:01: for then to build reports and dashboards by the data is still hard part.

00:06:06: I

00:06:07: would agree with that.

00:06:09: And let me tell you a little bit, an anecdote or story when is pitching click data early in the years and we said We're going to decide to build connectors for every single data source out there as many of us can To give really the ability our customers have to be self-serving and so forth.

00:06:28: And i pitched this actually between investor earlier on.

00:06:31: This investor trusted me.

00:06:32: He goes No no...I don't think he has a private care APIs, I hear a lot about APIs.

00:06:38: APIs are going to solve the world's problem with API.

00:06:43: software is gonna talk to other software without issues right?

00:06:47: And you know here we are right?

00:06:49: APIs or just standard?

00:06:50: and yeah they facilitate some things but They don't solve the worlds problems in the definitely not standard.

00:06:59: When people talked about you now which is plug-in AI and deal work It flashes back to that.

00:07:04: every single new technology we come up with, Cloud or Web APIs.

00:07:11: Whatever you want call it comes in sets of huge benefits but still doesn't solve the most problems.

00:07:22: You know, the old metaphor garbage in and garbage out is still true.

00:07:27: Right?

00:07:28: If you'd garbage to the LLM... ...you're gonna get garbage out!

00:07:32: It's undoubtedly true But it does raise a different level of autonomy I find In people that are building these flows Which i think will be very helpful.

00:07:47: I was not so sure of that at the beginning.

00:07:49: And also, here we go another wave of marketing fads and everything else.

00:07:57: but as we looked into how to implement that click data internally for our own use or customers We started seeing some interesting things if again the customer pay attention is sixty percent of the work of anything after that, right?

00:08:16: Whether you're using dashboards reports or a chat bot chats GPT to talk to the data.

00:08:22: Or any other downstream agents too to process the data if that data's not writes everything kind of collapses and we are going back to different type way working where now validating the AI is producing, right?

00:08:45: And that's a lot less satisfying.

00:08:47: Before we look at it again let me first highlight what do you think are most important... Let us maybe call them building blocks.

00:08:57: when I'm thinking about data management for analytics and AI or these types of use cases What do see as one of the most important things companies need to address?

00:09:10: In my view, again it's a dual contradicting kind of approach.

00:09:16: Back in the days... Again back in my corporate days we were all about standardization of data at source MDMs strong master data management very strong data schemas.

00:09:31: The data warehouses had to be cleansed and proper.

00:09:35: There was a lot of documentation so already some of the concepts of semantic descriptive properties and metadata on-the-data itself.

00:09:47: There was a lot of work on that earlier, typically by IT or data analysts working with IT And that would fundamentally make the foundation for The Data Warehouse.

00:09:58: there's problem with that!

00:10:00: The problem is speed number one right?

00:10:04: Speed not only in terms producing this and making changes to these very lengthy, process-oriented processes if you will or at least documentation heavy processors.

00:10:20: But it kind of creates a bottleneck within IT organizations, data organizations... ...to streamline everything through there.

00:10:28: then give back the business community.

00:10:32: The other side is agility where you created people on the other side that basically said, well no I want my data right now.

00:10:40: I want agility and speed.

00:10:42: i wanted to do that today.

00:10:44: have a market impact Today!

00:10:46: I can't wait to draw up functional requirements.

00:10:49: so these two are kind of diverging in opposing things.

00:10:54: we need both.

00:10:56: We Need Agility, Speed.

00:10:58: And yet we need provide proper metadata and structure implement proper analytics on their data.

00:11:10: One cannot live without the other, and earlier when we started to click Data we said... We had this saying which is companies should embrace.

00:11:21: at that time a lot of people were saying stop creating data silos in the company.

00:11:26: don't do data silos centralize everything.

00:11:32: My vision was, no embrace data silos.

00:11:36: Build your process around the fact that here's the thing you are going to have data silos.

00:11:42: there is going to be a manager walking around on their laptop with an Excel file That contains the budget number for next year.

00:11:49: Embrace them Understand how?

00:11:52: You can somehow harness that and say How Can I fix that?

00:11:57: leaving the person, the ability to continue work but embracing that somehow in a structured process oriented way.

00:12:05: So we said embrace silos and interesting enough came about at same time as the cost of data lakes just basically dumping everything into a data lake.

00:12:14: then let somebody sort it out.

00:12:18: well thats kind of the idea is.

00:12:20: say lets stop trying be very strict on all this figure how you can adapt and kind of embrace the chaos, in a controlled manner.

00:12:32: So that's I think where we're headed to again.

00:12:38: AI is going to open a lot more doors.

00:12:40: this time it's not just gonna be one or two people who are Excel experts in your organization.

00:12:45: It will be everybody with an AI chat or agent And they create their own analysis They add data on them.

00:12:57: So now you really have to embrace all these variety of things.

00:13:00: How do you do that?

00:13:01: That is going to be the challenge, I think

00:13:05: Absolutely!

00:13:07: But obviously there are some problems with data silos as well especially at let's say definition KPIs

00:13:13: Yeah security privacy... There're a lot of problem.

00:13:17: but when i say embrace data silos and just adapt it.

00:13:23: It does not always mean just leave them be as they are, but somehow take into account the fact that people who now is running the LLM or run the data on their machine may not in the IT Data Server room.

00:13:40: As we think of it in the past right?

00:13:42: Its'nt a controlled-matter object anymore But you need to get it from them in proper way.

00:13:50: Yeah,

00:13:51: I would agree.

00:13:53: In a way it was never controlled.

00:13:56: these big waves of building data warehouses, building data lakes now doing big data lake houses?

00:14:03: They are all

00:14:04: not representing the organizational reality that... The dynamics even increased!

00:14:11: That will become tougher and tougher to have everything on one place Even though you might want try this.

00:14:18: But we definitely see ideas like a fabric, data-fabric architecture being more flexible on this.

00:14:27: It's definitely appealing to many companies.

00:14:30: but now obviously you made me curious yeah?

00:14:33: You said okay... We need to embrace it!

00:14:34: Now I think you should tell us how can we do

00:14:37: that?!

00:14:37: You know i show these terms with the few other analysts within Berg but potentially even outside of Berg.

00:14:44: But I think the what I call it, the metadata which you know some people also called semantic layer.

00:14:51: The semantic type layer that we need to and not be confused necessarily with a master data management or maybe part of that.

00:15:00: but these terms have been around in the IT industry and they did industry too long.

00:15:05: so pay attention to peoples definitions where their terms are still do.

00:15:10: this day People call BI just dashboards.

00:15:14: to me BI is more than that, but you know the definitions.

00:15:17: But this semantic layer and what it means to me I think when i was coding back in my younger days we used to call these things in C++ called pointers And pointers are just pointing To where the value is finally or where the memory block Is of data?

00:15:35: That says this is the data that represents your KPI.

00:15:45: This is where the data sits, this how it looks like and this my description of It right?

00:15:51: And that includes all different KPIs you can build with them.

00:15:55: if You have that which has no Data in itself just saying Describing Where it Is or you Can get it what it Contains How it Looks Like etc.

00:16:07: Then you can have this data however you want, in silos or not in silos.

00:16:11: It doesn't matter okay?

00:16:13: As long as it's somewhere that you can easily access to any its entirety which is an important item and then You just work with the semantic layer.

00:16:24: You don't care anymore about The data behind bad right because a semantic layer isn't easier like a catalog but it's easier to maintain.

00:16:33: Easier-to-maintain could be a dual-edged word there, but definitely takes some work to maintain.

00:16:40: But once it's proper and maintained consistently and always pointing where the data sits and how it looks

00:16:47: etc.,

00:16:48: I think that is the best way to embrace data silos, to embrace chaos by saying yeah!

00:16:54: There will be chaos behind this wall... ...but this wall is very nice and clean for me.

00:16:58: And This Is What The Company Should Work With Right?

00:17:02: This Where My Business Users Will Work With.

00:17:05: And you may have different semantic layers for different teams, for different departments.

00:17:10: Marketing and sales may have a semantic layer which overlaps somewhat with logistics or production... ...or with finance.

00:17:18: it doesn't matter.

00:17:19: but what's important is that for that world-for that team That semantic layer becomes the truth of not data But The Truth Of What The Data Is!

00:17:31: And the LLMs will appreciate that as well Because now you have an amazing amount of information.

00:17:36: You don't have the data itself yet, but what's inside people's heads which is a description for what that data is etc.

00:17:46: and because we're pointing into where it is, the LLMs didn't know how to go get it or figure this out right?

00:17:51: That's easy piece for LLMs too do!

00:17:54: But that descriptive nature explanation about what that means That's what is more important to get out of people heads and into a proper system that can track them.

00:18:08: How should the good semantic layer look like?

00:18:11: Maybe you could wrap it for me if he says, okay this is a good symmetric layer!

00:18:15: The question about what is a Good Semantic Layer... ...is that the concept on which it should contain And how does it overlap with other processes related to master data management?

00:18:26: So the way we see a quick deal, which may not necessarily be how a semantic layer is.

00:18:31: A lot of people start well you know just basically they're just describing the table of data and your describing each column on what it means.

00:18:41: tagging potentially that's a lot of People use tagging categorical tagging as well as descriptive notion.

00:18:51: If you're doing combined metrics, say this metric here is this divided by that etc.

00:18:57: and it represents this.

00:19:00: That piece is easy!

00:19:02: It's really just metadata applied to the tables.

00:19:06: where I guess a little bit more complicated is companies.

00:19:10: unfortunately in more complex companies do not have one nice table that we can describe right?

00:19:17: So you kind of create a virtual virtualization of that catalog of data, which has existed in the past before and back into Data Warehouse days.

00:19:30: But it potentially has elements of MDM.

00:19:33: meaning if I say well you know If your trying to do this KPI or if you're trying look for an invoice number You need to go to this table but if its not there then you have got these other tables.

00:19:44: so theres a fallback sequence golden record type of thing that you need to somehow incorporate into there.

00:19:52: Because if the LLM doesn't find that data, it's just gonna give you bad data right?

00:19:55: So we need to put fallback items in today and You also need to Put validation because even though you say you described they didn't said this data represents The quantity of items purchased etc.

00:20:09: But you have bad data somewhere underlying.

00:20:12: again It's going to affect.

00:20:13: so you have two also define rules as too For example, quantities here should be greater than zero.

00:20:20: Should not be over a hundred because that would be very bad number.

00:20:24: Date should do it and not be nineteen seventy-one as the default date or something like that.

00:20:30: so you have to put quality rules around data.

00:20:33: So in my view at least It's lot of elements for master data management in some cases Data validation, data quality And even data privacy by the way because the semantic layer must have data privacy and data security rules embedded into it.

00:20:53: So that, um... The people looking at it do not see data they should not be able to see or more importantly that the LLMs don't as well right?

00:21:01: And they don't share important data.

00:21:03: so it has a lot of privacy and security rules around the data.

00:21:11: all in the same semantic layers looks like, and when it should use it.

00:21:19: When you should not use it?

00:21:20: When he should display or not displayed etc.

00:21:23: right um You will then get a little bit more complicated.

00:21:26: is on the combined metrics across different tables And things like that.

00:21:31: But those are just problems That we'll kind of be solved as We go along but nonetheless Those Are The basic elements I see On that semantic layer today.

00:21:41: I just want to refer back to what you said.

00:21:43: You said, sixty percent of the effort is more on their data side and some say it's eighty percent.

00:21:49: but i think we all agree that

00:21:50: seventy-five percent of these statistics are wrong anyway.

00:21:53: so its okay.

00:21:55: It's a big part

00:21:57: yes!

00:21:58: And i think you know nicely laid out yeah with different components uh... We have here and now my question is already alluded before.

00:22:08: AI obviously also has a big influence now, not only how we do things but how to manage data and get supported in these processes.

00:22:19: Where you see the biggest impact?

00:22:21: Or maybe your tool or organization?

00:22:25: where does it have the biggest effect of AI for data management?

00:22:29: In our platform we did strong analysis.

00:22:33: Are we going to start adding AI and where do you stop?

00:22:36: And what are the concerns, so forth.

00:22:38: So this was a very important topic for us.

00:22:41: now We did What I think everybody Did.

00:22:43: uh...we started Adding AI as an assistant For small features Small functionality Um..We thought of it As more of an advanced IntelliSense You know, intelligence that starts typing The word as you or finishes the Word as you type It but basically adds.

00:23:00: Uh....the ability for data analysts To Just ask AI to type a formula for them or do these things.

00:23:07: These are helpers and they were just easy for us to implement, so we did it.

00:23:12: We added AI For the customers to use meaning when you're building a data flow which is equivalent of a data pipeline if you will.

00:23:23: sometimes They want to use AI To summarize text Or sentiment analysis.

00:23:32: So we built little blocks of things that they could start using in their data to enrich their data.

00:23:38: Okay, so we started again adding a few elements here and there generic AI even for them to ask any question about the row of data or their entire table of data?

00:23:49: Again easy stuff.

00:23:51: you asked AI how are my cells for twenty-twenty five?

00:23:54: is finds the tables he tells your sales with this little chart here a little chart there That's fine.

00:24:01: We then start thinking, okay.

00:24:03: Well this is useful but it's conversational.

00:24:07: we can give these to the customers and put on their dashboards.

00:24:10: so in our dashboard they could ask some questions about the dashboards when getting answers.

00:24:16: Then we started asking our customer are you going still build dashboards or should just move totally into a conversation with data?

00:24:30: The answer, you know a hundred percent of the customer group that we have was no.

00:24:35: We will always have control over dashboards.

00:24:37: Dashboards is still our mean of communication.

00:24:40: it's a bit like saying are we stopping Microsoft Word because now AI can type everything right?

00:24:49: No I mean...we're still going to print and save and have these documents and formalize them And ensure there's standardization in repeatable Action here, so we said okay.

00:25:00: Well if that's the case what would you think about for example?

00:25:03: The AI building into our dashboard free?

00:25:07: And they said wow dad That's great even if we don't use everything If he gets us eighty and ninety percent there Okay.

00:25:15: So then we ask well if the AI could build a connector for you Instead of us building the connectors.

00:25:21: What if you just say oh I'd like to connect this system Yeah, I will look at the documentation.

00:25:27: Look at the agent functionality that we have to offer and say yes i can connect this And make a connector for you.

00:25:32: so now We're no longer waiting on my engineering team To produce a new connector For them.

00:25:38: they can connected it?

00:25:39: Then of course ask well between The connectivity individualization all its missing is data processing.

00:25:44: What if he could ask AI To build the data flow for You the data pipeline the transformation and everything else?

00:25:50: and They said wow That's even more amazing.

00:25:52: So thats what were doing right.

00:25:53: today we're building those, if you want to call them agent tools or whatever.

00:26:01: We've launched the Dashboard Builder AI yesterday I believe, Tuesday and now they can ask build me this?

00:26:09: And it will start seeing all these charts and widgets popping up.

00:26:13: and boom!

00:26:14: You have a dashboard.

00:26:15: If you like it don't you'll need to change of course but that's way.

00:26:20: Next release is having the connector and pretty soon has the data flow as well.

00:26:24: So that means right now, potentially by the end of year a data analyst faced with new requirement could talk to Quiggett and say hey I need get these metrics from this system.

00:26:38: And he'll know how to connect build the Dataflow and the Data Pipeline and build the dashboard and built an entire thing for them.

00:26:45: This is our goal in Objective our customers which are most like very data savvy people, data analysts business managers that know data.

00:26:57: they know their business and to accelerate faster.

00:27:02: So it's an accelerator.

00:27:04: It doesn't do magic for you.

00:27:06: You'll do as much as it can And then you need to add additional items your human side-to-it in the little nuances that the LLM did not pick up.

00:27:16: This is why of course the semantic layer so important Because the more they describe what you're trying to do, the more we know about it.

00:27:23: The better the AI is going to do for your job of creating these processes and pipelines because that makes sense too!

00:27:32: It does yeah?

00:27:33: We had already several conversations here with people explaining how an AI can accelerate tasks or speed up or decrease human effort in modeling data flows doing mappings etc... But when you describe this, that AI can do more and more.

00:27:53: I have to ask the question where does that leave view?

00:27:56: If AI takes over more tasks are you becoming superfluous?

00:28:01: Well we have to split our software into two pieces.

00:28:04: there's a user interface piece right And like i said We still need a dashboard agent To work behind-the-scenes.

00:28:13: okay but The question is Do we need a dashboard UI builder ?

00:28:17: We need a designer for that anymore?

00:28:19: That's a great question.

00:28:20: We're posing that question every day, you know for the time being we are going on the assumption because we have to build de-functionality.

00:28:28: building the UI on top of it is not hard so we continue do both but at some point in times when they say well really no longer need data flow or connector property screen.

00:28:40: even dashboard designer screen may just have AI manage that piece but the functionality and engine behind-the-scenes is still less.

00:28:50: AI's not going to do that, I mean we can talk about complexity of some systems connected with security, authentication or everything else built for.

00:29:02: We could talk about embedding dashboards into other applications which are a service provided by customers before integrating analytics.

00:29:14: We can even talk about the functionality of our data flow, which is... Can do amazing things.

00:29:20: Which potentially you could ask an AI to build it in Python or whatever?

00:29:27: Yeah!

00:29:27: It'll basically produce the same code that we most likely produced as in code but essentially with a lot different gaps between.

00:29:37: So I think the engine still... A lot people say software-as-a-service is dead.

00:29:45: I think software as a service, as we saw being a web application with some functionality behind it may change.

00:29:55: The Web interface is no longer the primordial way of potentially talking to the functionality and to data you have behind but the fact that you still need secure data warehouse?

00:30:07: Absolutely!

00:30:09: to have data talk A-to-B, absolutely.

00:30:15: AI does not create databases and doesn't start things for the customer in a vacuum right?

00:30:23: It creates it somewhere as you use tool that has functionality behind scenes to operate.

00:30:29: So yeah I think we may be reduced who knows To a Data Engine software As opposed to a Data Management and Analytics SaaS company.

00:30:41: Okay, since we are already looking a little bit ahead of what's going on and potentially also in the future that would be my closing question.

00:30:49: Since our time is already up this really has flown away.

00:30:54: Looking ahead?

00:30:55: The road ahead?

00:30:56: yeah so... What do you say?

00:30:58: coming back now to the topic data management or building a data foundation for whatever analytics AI What do you think are the most impactful changes or developments that we see in the next, let's say three to five years?

00:31:13: That will affect.

00:31:15: Yeah well if we go back to what I said it was a very important piece which is The Semantic Layer Is is the maintenance of that semantic layer.

00:31:25: So that's one piece.

00:31:27: i may think Of others as i go along but my issue with the semantic layer because It's a concept that is not necessarily new master data management has been around for years.

00:31:36: We've tried to do this, we try to classify and standardize the data.

00:31:40: Where it fails or where it failed at least in many cases is on the maintenance building.

00:31:46: It's hard work for sure but maintaining up-to-date so forth.

00:31:51: So I think one of most interesting things how can you come with something using AI as a tool?

00:31:59: To maintain the semantic layer alive And very agile which was one of the problems we had with previous ways of doing things.

00:32:13: If AI can help us in creating that semantic layer, maintaining it up-to-date dynamically and recurrent way I think that will self feed the ability for analytics to work better.

00:32:32: The second piece is actually a problem that we have not been able to solve, and I'm not sure.

00:32:39: I've heard of anybody being able to sell them maybe some very like Google or an IBM kind of today?

00:32:47: We have A problem with the way that LLM's work with these tokens in so forth where we can ask easily question about A hundred rows of data, a thousand rows.

00:33:01: We send this data to the LLM and they will make all kinds of great analysis for us.

00:33:07: if The world was run on a thousand row tables or ten thousand rows that would be wonderful.

00:33:12: But you know we're talking millions are rose were talking millions arose in multiple tables.

00:33:18: Were talking billions of rows for larger customers?

00:33:21: Um...we

00:33:22: cannot possibly feed from a cost perspective From even an energy perspective.

00:33:26: Feed all the data to llm and have it crunch stuff out for us.

00:33:32: Today, LLM still remain very much like a SQL generator.

00:33:39: they'll say oh that's the table.

00:33:42: yeah I'm not gonna have access to.

00:33:43: your data is too much data.

00:33:44: you're now going send the data to me.

00:33:46: so i am just going run queries here there create little extractions of the data figure what can do or not with And that to me is a shame.

00:33:57: So there we have physical limitation, resource limitation and cost limitations That yellow lamps don't actually touch the data directly.

00:34:08: The barrier continues today for most cases.

00:34:13: Those two I think if you can solve both of them in fact will advance quite rapidly set of applications coming alive using AI.

00:34:30: And, of course the third one which you know is really referring to what we just talked about the UI Is I do not think anymore that people will log on to OneSystems To get their dashboards or reports Or whatever?

00:34:42: i think everybody Will have The choice to log into Mistral to open AI to Anthropic clients Whatever they are Claude Vibe and so forth and talk to it, add little agents to write emails etc.

00:34:59: Well there will be an agent for their data management as well so the interface would change.

00:35:07: We'll no longer go into an application we're going through our favorite GPT client or whatever To access these things.

00:35:16: So those three are kind of my next steps of Data Management BI however you want to call it.

00:35:25: Excellent, so still a lot of interesting times ahead of us!

00:35:31: Things will not become boring here in our space and we could hear your insights from all the experience.

00:35:41: I think also quite insightful statements on how build a data foundation How make that successful And with that, yeah.

00:35:51: Laying the foundation also for AI success in any organization.

00:35:56: Tamo!

00:35:57: Thanks so much for that.

00:35:58: Thank you very much having me.

00:36:00: Hopefully I didn't talk too long and didn't bore ya

00:36:03: Not at all.

00:36:05: Excellent thankyouverymuchforhavingme though Very nice.