Stop Buying GPUs You Can’t Use: Qlik’s Framework for Trusted, Agentic AI (with Qlik CTO & AI GM)
Transcript
All right, guys. This is going to be a fun conversation because you promised I can go anywhere with it. And I have Sam and Brandon both from Qlik. And you know, Qlik has had an interesting evolution from being the everything product to really being focused. And I got really excited when I told you I talked to clients that have went out and bought H100s in the attempt to augment their SAP environment. They want to get some type of value out of the data.
And what they discovered is that they bought a bunch of hardware that doesn't get them to the solution. Mhm. And Sam, you promised to walk me through kind of a framework of how Qlik looks to solve problems like that. Yeah. Yeah, totally. So, what we're going to do today is we're going to white board this out a bit. Um, and like you said, I think, you know, you talked to a lot of customers, a lot of users who have all different types of data, right?
You got your databases, you got your unstructured data, and none of this none of this really changes in the in the world of AI, right? You've got all of your proprietary data, but the thing is is like you got to get out of these things and then get it into something that's useful, right? The the first part of this is always move, right? So, we have lots of different ways of doing this, right? You've got things like um plain old uh you know, extract and load.
You got things like uh change data capture. , right? And this this goes on and on and on and on. Now, this is coming out in its raw data. Um, the the models today, like you actually need to get it in shape, right? So, the next the next phase of this uh and this can kind of be done in different orders, but for the purposes of today, we will then go into the transform stage, okay? And again, this can be done variety of different ways, right?
You like you've got your different patterns around things like uh ETL, you got things like ELT, you got your uh you know, ELTL, right? Lots of different ways to do this, lots of different vendors that can potentially be involved. And part of the part of the thing once you get this into, you know, your known good state, but these things are constantly getting updated, right? You have things like CDC, which are constantly emitting new events. You could have things like Kafka or a Kinesis stream that are that are pumping in uh different data elements.
You could have uh entirely new schemas that are coming from some of these different sources. And so, the other part of this then is the trust column. And this is really the the I think the the keystone for all of the data pipelines. Because you have to think about things like uh you need to have things like your your quality, right? Do we know do we know that these things are not getting polluted by a broken integration upstream? Uh are we getting things like null values where, you know, where they shouldn't where they shouldn't be, right?
You have different uh formatting things. And all of these things can come in at any time uh during you know, during this pipe. And then downstream, your consumers are basically directly impacted in that. So, with our data product concept, what you can do is you build these data pipelines, you bring the data in. Uh and then we have this uh this really nice system that basically goes through highly configurable, but also opinionated, that basically help you get all of this data into a a trusted uh a trusted foundation.
Then you get into the things like access, right? And this is where I would say the two the two sides of the business always meet. You've got things on this side that are really driven by the data engineer, right? Uh on this side, you all you'll have your your line of business users, right? And this becomes really a contract for providing this data product to the end user for a given use case. So, I want to pause here because I love that you went through the complexity of moving from your data.
So, let's say that, you know, one of these sources is my SAP uh uh ECC transactional data. >> Mhm. Another source is my uh data from my logistics system. That may be SAP, it could be Oracle based system. Really doesn't matter, but when I need to get to this point where I can prepare the data to analyze that ETL process I'm using one or more solutions to copy the data out, maybe another solution to normalize the data. Then yet another solution to get the quality to that gets my data quality uniformed across.
And it sounds simple, but at scale, this is exceptionally fragile. Yeah. And this is where I see customers falling down is they went out and bought a bunch of GPUs to do this point after, but this is where they're falling down. Mhm. Mhm. Yeah, I think um you know, and you go back a couple of years, right? Like things like the modern data stack were very in vogue. You had all of these different point solutions, and in a lot of enterprises went out and bought licenses for those, and they had to cobble all these things together.
And I think what you're seeing in the industry now is there's there's much more consolidation because those those some of those failed implementations or things that didn't work quite well together. Like that's where we see a lot of the consolidation happening, and people more I would just say like kind of go towards our strategy of having this end-to-end platform for, you know, getting all of this data ready for everything to the right side of the diagram. So, I understand the processes. So, you're not saying that these processes aren't needed.
Qlik is helping me on the data side Mhm. operationalize this. And then how you help me operationalize? >> Yeah. Do you want me to take the other half of this? Yeah, maybe I can draw a little bit here. So, as Sam pointed out, we've got data products that actually sit in the heart of all this, right? And that's one of the things that as you think about being an executive or business user who actually needs to make decisions off of this complexity, we want to service it in a way that will allow people understand how it applies to their business.
We all do this, and we think about it in a way that we are running this on top of an open lakehouse, right? The open lakehouse allows our customers to really take advantage of a bunch of different things. One, scalability, right? Everything based on iceberg allows you to bring an unbelievable amount of data in there quickly. In addition, and what's really important to CIOs and CTOs, as you know, they can do cost effectively. So, we can actually reduce their spend.
So, once you get all of this complex data, you get it ready, you surface it up as a data product, the next thing you're going to need to do is you're going actually going to need to provide some type of analysis on it, right? So, what the what the heck is in in this data? What is it tell me? What are the signals that you get in there? And there are things that that sit in here that are like business aligned data models, right?
Is a really important thing. So, say for example, you have a supply chain, you're running supply chain, you bring your supply chain data in there. Or if you're running um an operation or merchandising, you can bring a data model that addresses that. What gets really even more interesting is it's not just the analysis of it, it's actually your ability to make a decision off of it. So, that you can put a predictive model in there, for example, that will allow you to identify patterns within that data that will tell you what you should actually do or make recommendations.
And then finally, the thing that we love is acting. And this is where the future is really coming into this. So, this is going to be things like alerting, right? You get alerts. Uh you may actually get an automation, for example. That may be a big I mean, I'm not kind of short now. A little little longer, right? Um, there are things that you can get reports like you're running your business. And the really interesting thing is you look at all of this.
This is your traditional data stack from raw data to actually making decisions. In today's world, you have this really interesting layer that goes across the entire thing. And what this is is this is an agentic framework. So, fundamentally shifting the way people are thinking and acting on their data and making decisions. So, everything that has traditionally been and you've been in the industry for a while, this used to be super manual. Yes. This is still somewhat manual. Mhm.
But bringing in an agentic framework, what we are going to allow customers to do is use agents to go off and do some of that movement or transformation or even go after trust. So, I I think I think this is the thing that a bunch of the viewers are having a aha moment from from an agentic framework. Repeatedly doing this extract and load or taking data out of my SAS system, whether we're talking about MCP servers or whatever the the way that or API and my agents are interacting with this.
This is where folks from the what we're talking about ML ops or ML data ops are getting really excited about. Yeah. This ability to take the 80% of the time that my data scientists work on normalizing data, now this is I have a a agent that's doing that for me in the background. Help me understand, and this is where I'm going to poke at the the edges a little bit, this trust layer. Because what this trust layer becomes the thing. I cannot make business decisions.
I can't even get to this point if I can't trust the underlying data. How is Click helping me with this agentic AI trust my data? >> So there are a couple there are a couple pieces of the pop the puzzle here that I want to make sure that we're calling out. So first and foremost, we're really focusing on if you go back to the above, it's all about trusted data products. And where where where Click has been traditionally very strong is actually applying the trust scores to the data.
So here we have a trust score for AI so that you can have confidence in that. But what gets really interesting is we have an analytics engine. And this analytics engine actually allows you to combine really disparate and distinct data sources and find pattern within that data that will allow you to make decisions and act on it with a lot of confidence. Ultimately, when you get trusted data products plus the analytics engine, you get context for your decisions and a trusted intelligence layer.
>> So if I'm hearing you correctly, traditional Click proven over the years they can get you trusted data. Add to that analytics on top that decision the ability to to get analytics and decision out of that data is kind of where the two houses are together. I I had this practical problem just the other day. I I had my DJI Spark and I loaded I do not delete email. I am the worst executive in all of history because I have 15 years of email, 103,000 uh messages.
I have this powerful AI uh personal supercomputer that I have all the models in the world. You know where I fell apart when I was when to go analyze my data? Is that I had no idea how to transform the data to any way that I could use it to access and and do analysis of the data. " I could not get to that business decision. So let's finish out with this practical example of where we have this unstructured data. How do you help me with unstructured data getting to this decision and prediction and decision capability.
So we really want to fit into other ecosystems. I think that's an important piece of the puzzle here. So let's just say your your vendor of choice happens to be Anthropic and you use Claude. That's where you tend to use. We want to make sure that you can start in Claude, ask the questions of tell me the best deals I've ever done over the last 10 or 15 years. What that will do is that will work with our agentic framework using natural language and it will invoke a series of agents.
So on a part of the the framework there there will be a knowledge agent, for example. That would be something that would use your unstructured data. You could use a discovery agent. That would actually act on your structured data. And because these are all part of the same portfolio and platform that we have in place, it gives you the ability to combine these things and then combine them with all of the different capabilities that you have across the entire portfolio. So you could start with natural language.
It will go off, grab your information, make sure that it went against your trusted data. Make sure you can get your data products. You know you know you're headed in the right direction. And ultimately, it'll build a model for you. It may actually say, "You know what? " Or it might it might. And then it will finally say, "You know what? Let me tell Keith all about this. Let me run an automation and and generate something that will say, 'You know what?
'" So this is even better than I thought. " You can now connect to my QuickBooks data and actually see the closed deals, see the profit margin of those closed deals, go from the negotiation to the actual closing of the deal in the and me getting payment. Because part of part of what's my best deals, when did I get paid? What was the terms of the deal? If as I put on my engineer hat, I'm as a data scientist, I'm incapable of making that connection.
I don't know how to organize the data in a trustworthy enough point to even get it to the point where I can feed it to Claude because those are just two very different data types. And you're asking two one question that you would think would be common sense, but it's a really tough engineering problem. >> Yeah. Well, and I think you can expand this even like so, you know, you talk about sales. There's other use cases around customer support, right? So, you know, again, this could be like your ticketing some sort of ticketing system, your Jira uh or whatever it is, right?
So a ticket can come in. This this would also be then be able to be paired up with things like customer retention data. You could pull in unstructured data from like your knowledge bases. You can pull in other support tickets. You can have an understanding of what like where might this problem be? Have we seen similar things like this? And so you can actually thinking of like you think about like the KPIs around this, right? " Um And like Brendan said, right?
I mean, I think this this is one of the this this this layer right here is how we're doing all this internally. But I think the mistake that we see a lot of other vendors make is they think this is like this is the thing and they want to go sell this. The key is actually integrating it into that ecosystem. So again, people are going to start their day in Claude, they're going to start their day in Gemini. And to be able to interact with that data, being able to call into our platform via MCP actually ends up being a really really powerful thing.
So we're and we're really excited about that. >> Yeah, I can't wait to see some of the use cases emerge where, you know, I'm a business user. I'm using SAP. I'm in my SAP reporting and my gut is telling me there's something more there that I'm that I'm not able to generate the right view or report to see, then I can go to a Click and get ask that basic business question like, "I think I see I I'm seeing this pattern. " Yeah.
Yeah. Guys, what did we miss in the conversation if anything? I think there are a couple things. You mentioned you mentioned a few three-letter acronyms that which probably flew out there. Right? MCP is a huge thing for people in the way they're thinking about engaging with their analytics. So what Click is really thinking about is playing in other ecosystems, right? Start I'm I'm a huge I know I'll admit I'm a Claude user. I love him, right? I start in Claude every day.
So we have we use MCP to connect to our Click environment so I can get trusted information there. The other really interesting thing that uh we really need to think about is A2A, right? So the agentic to agentic, right? We want know that you brought a great example of SAP. They'll have their agents as well, right? " That's what we want we intend to do here. We want to focus on what we are really good at. This right here is where Click's legacy is and this is where we want to focus.
We don't want to be a customer service agent, right? That's not what I want to do. I don't want to do customer experience. I don't think Sam I don't think any of us want to do that. We want to be that vendor that allows our customers to use trusted data products combined with the analytic analytics engine, make sure that they can get context for every decision and every action they make so that they know that they're going to really use agents with as much trust as possible so they don't get themselves in trouble.
Yeah. And I one other thing that I'll add to is I just I think like this open lakehouse concept cannot be emphasized enough because this you know this is this is presented as a very linear thing, but this could you know if we got you know if we had a bigger whiteboard here, I think we could explode this, right? Because you would have things like maybe this maybe this data maybe this goes up to Snowflake, right? Maybe it goes to Databricks. Maybe it goes to the Amazon S3 tables, right?
And the thing about Iceberg and the lakehouse now is that you can actually optimize this. You can put the data where it needs to be from a performance perspective or if it's more commodity, you can put it into the S3 tables and have more commodity storage. And like the the Apache Iceberg concept of breaking apart the compute versus the storage versus the metadata really allows us to then not and like you're not even you're not locked into our ecosystem either because now like we can connect into the catalogs from these other providers.
They can reference the data that's you know Databricks or Snowflake reference the data uh you know, without moving data, it can reference and compute do compute on top of these S3 tables. So there's so much different like optimization that you can do here that can bring down that cost and basically free up those dollars for all these Yeah, I saw that. Is that a neat? If I already have Click and I have this IP agentic firmware firm uh framework, I'm using Cursor, Cairo, really doesn't matter.
" So I have a from the application developer, data scientist, really doesn't matter. You've simplified this process for me to get it here with Click and data analysis, but if I don't to use Qlik for my data analysis, I can use another service, so I can provide it to So, you're you're literally getting where you're You're getting in where you can fit in, which is this process beginning with, and I think where you win additional trust customers and trust where you can show them that you can go all the way here either with Qlik's in-house online solutions or through something as simple as using a cloud.
>> Mhm. Guys, I really appreciate you taking the time out to sponsoring this content, and we're at re:Invent, and you you you the schedules are busy, and I really appreciate this detailed look at Qlik's data and analytics side of your product suite. com. That's with a Q. com, where we talk about all of this good stuff. Thanks for spelling that out. That's It's a light board. We can have you put stuff up. We can have you put devices Exactly.
Talk to you next CTO talks. Thanks.
