Qlik AI Strategy Overview and Agentic AI Deep Dive

Well, thank you for being here. This is the fourth time that I've shared, uh, some of the things we're doing at Qlik with this group. So I'm, um, I'm glad that you all have to see the evolution and, and be a part of it on my side.

Um, my name's Nick Magnuson. I'm the head of AI at Qlik, and we're gonna talk about, uh, what we're doing and what we've announced this, uh, this year at Connect. Um, I thought what I'd do though, as we get into this is actually start with a bit of the why, what informs the strategy that we use at Qlik.

And for me, this is a bit of a, frankly, PTSD, like, these are the same problems that AI has encountered when I was launching machine learning over a decade ago. Right? It goes back to the data.

It goes back to the wrong use cases. It goes back to the complexity of these systems. And I feel like we are, we're seeing this movie play out yet again.

Um, and as Josh spoke at the very outset, part of our underlying business strategy is to try and address each of these pain points in the evolution of building these AI solutions, uh, so that you ultimately can do it, uh, better, faster, cheaper, uh, and with more confidence, uh, and trust. Um, because what I do know is, say, NA, we use chat GBT. That is not an AI strategy.

An LLM is not an AI strategy. Hmm. And I think part of the confusion, guy you touched on earlier in one of your questions is, you know, we've got a lot of pressure from boards and from C-level executives that want AI injected into the business, and that's causing us to start to think about things that, uh, you know, may not actually represent a strategy, but is just throwing, uh, spaghetti at the wall and seeing if it sticks.

So at Qlik, our strategy is very much founded on, there's a lot of different ways you can employ ai. Uh, in some cases that Josh talked about, we've employed AI back to the nineties, um, and it's seen an evolution throughout that course. Um, but, you know, one thing I can say is when we talk to customers, they all wanna do something in ai.

And oftentimes when you unpack what they're trying to do, it's not generative. It's not a gen ticket. It actually might be traditional machine learning, it might be traditional analytics, it could be an automation.

And that's why an entire platform that Qlik represents now is really important as we can solve a lot of these, these issues, um, with, with technology that's been around for a long time. Um, I wanna talk a little bit about predictive. 'cause guy, you touched on this.

Um, we have customers now that are, that are using this for a variety of use cases just within one one particular industry. So within healthcare, we have customers using it for patient readmission. We have, uh, Appalachian regional hospitals, great example.

They use predictive to predict when patients will not show, and then they take corresponding action to fill. That would be, uh, otherwise lost revenue. And it's saving them tons of money, tons of money, very mundane use case.

Um, but yeah, you mentioned the big squid acquisition earlier. This product has now been in, in the product portfolio for three years. I'm proud to announce we're at a billion predictions that our customers have run off of that platform.

That's a billion with a B. So it's a, it's a big, a big milestone. And of course, last year when you were all sitting in a very similar room to this, uh, we announced click answers and that was our generative AI solution.

Now at Connect today or this week, we've announced the, that, uh, answers is going agentic. Um, but I think the, the point here is our strategy helps to, helps our customers get to a point where they can use whatever AI matters, whatever AI is relevant for the use case, but very, be very much focused on the use case first, and then apply the right AI technique to solve that solution. Like 1, 1, 1 second.

Yeah. Gen ai. So, and feel free to contradict me if, you know, anything I say doesn't jive to you, but Agen AI is different from the other.

One of these three is not like the other, right. Others, right. I think of ag agentic AI as an agent.

Mm-hmm. Primarily an agent. It acts as an agent.

It has goals. Those goals can change. And the fact that AI services from disparate, disparate types of many, any different type of AI service can be incorporated into the action of that agent.

That's what makes it an agent. Ai. We've had agents for a very, very, very, very long time for many purposes.

So there's a difference here in terms of whether we're talking about a predictive AI service that's incorporated into a platform, a generative AI service versus agent ai. We're not talking this, you don't run an agentic AI on an inference platform. Mm-hmm.

So, um, aren't you maybe a little bit sort of munging these together? Y yes and no. So being clear about the distinction, perfect Answer by the way.

Yes. Mm-hmm. And no, you will, It's like I was economics major.

It's always yes or no, right? Yeah. Um, so the, the truth of the matter is like, you can solve a variety of different use cases using very different forms of, of ai.

Uh, and as you point out, agentic could use generative and it could use predictive and it could use a whole other slew of, of, you know, techniques and other, uh, solutions to solve, you know, that problem. Um, what we have seen is when chat GD came out, in other words, all this noise about using gen ai, it created a lot of confusion as to what could actually solve the underlying business challenge. And I would say a third of the time when we engage with customers, when, you know, gen AI was the big rage, predictive was actually the more appropriate solution for the given use case that the customer's trying to, and I think we're gonna see the same thing with the gentech.

Yeah. I need an agent to do that. Maybe not.

Maybe so. Well, business problem, that's, that, that's a broad statement. It's not necessarily a problem.

So I, I would see predictive regenerative as providing an answer to a problem or to a question or to an issue. Agentic is more like addressing a goal, uh, or a set of goals, um, uh, you know, acting, um, taking action, um, in order to carry out or support the carrying out of functions. Those can address business problems, business goals, support business initiatives.

But that's a kind of a different meaning of the word problem. That's, It's a fair point. Okay.

It's a fair point. Yep. So I felt like you were going somewhere when you gave that example of the hospital mm-hmm.

With a predictive, right? Mm-hmm. You know, appointments made, appointments kept.

Yep. Generative AI seems like that would be the appointment, how that appointment went, not having had read all of the files. Yeah.

Or it could be, okay, we we're having a no-show, and the generative solution plugs in and starts a invitation to another patient to come to that, that that appointment that we now suspect is not gonna be shown. So following that same, what's Agentic Then? Agen would be an agent who has the autonomy to make that decision.

And then outreach to the customer, the patient that, you know. So, like I said, you know, the initial use case with Appalachia Regional was, we just need to know if the people aren't gonna show, okay, now we can evolve it to all these other things that are now coming along with the, you know, the evolution of the space here. Yep.

Um, so what I think is, is probably the major announcement for us here at Connect is that click answers leveraging this NEWENT framework that we've built out, um, now can incorporate not only unstructured knowledge base like you saw last year, um, and you can still use click answers to solve those things, by the way. But now we can also incorporate two new things. One is click applications.

So all that rich structured data that Qlik has had a heritage for 30 years can now be part of an assistant interaction. Where if I ask a question in the meaning of my question, um, requires that we answer it through a query against a structured set of data, we can now do that. If I ask a question and it is more geared at a set of documents that we might have, uh, in a SharePoint site or, you know, in our email files, it can also answer that.

And then last part I later this year, it will then be able to take corresponding actions that you give to that agent to then, you know, okay, now that you've answered the question, take this corresponding action, send a team's message, notify my team, update some record in our database or our CRM system, whatever that might be. And with that architecture that we've been talking about, the customer has the ability to design to whatever needs they need for that particular challenge that they, uh, yeah. That, that they have.

So, again, our connectors come into play, our API OpenAI, uh, API first development comes into play there where you can access all those via via those things. So, So Nick, when I, when I hear all those things, like I think about a lot of things. I think about apps to start with, right?

Yep. I think about click application automation. Yep.

I about how you can trigger some of those types using click application automation. And I think about how some customers are, I wouldn't say struggling, but they, they're still understanding how that even that flow works, right? And this is a massive step on top of that.

So the, the first part is like, how do you think customers are gonna cope with that and how they're gonna put that all together. And the second thing is, I think people for a long time have been trying to understand how unstructured unstructured data is going to be interacted with in the future. And I think it's great to make a, a statement like, we're gonna solve it with a gentech.

But one of the things that I see is that a lot of people are still used to using dashboards. Mm-hmm. And if you start throwing an agent experience at 'em again, that is a massive leap.

And especially when you take into account that answers has only been around for a year. So what are your thoughts on that kind of chasm? Yeah.

And I'll flip to the next slide just to illustrate that further. That is completely at the discretion of the customer. Now, if they want to just have a couple of apps or a single app as the point of interaction for a particular use case, they can design it as such if they want just a knowledge base to be the, you know, point of interaction for, you know, a, a great use case.

Some of the early answers adopters are using it with field technicians that operate in the field. Typically, they carry a big manual around like how to do things in the field. Now they just have an interaction point on a tablet, right?

But they're only accessing knowledge basis. 'cause those people just need to have the knowledge of how to create, you know, a solution or to clean the surface or to fix a air conditioner at a commercial site. You know, you don't need necessarily structured data for that.

Mm-hmm. But the point of all this is we give customers the flexibility to build it in the way they need in order to solve the problem that they're trying to go after. Um, and what I'll point out on this slide is not only do we bring structure and non-structured together, we also made enhancements to both of those things.

So if you guys are familiar with Insight Advisor, the very first form of our interaction with, uh, structured data through Natural Query, we now use LLMs to decipher the intent of the user question. And we do it at four times better than we did with Insight Advisor based on our benchmark tests. So not only are you interacting with structured data like you have been in the past, but that performance is much, much better because LMS are much, much more intelligent than the old ways of doing things.

Um, with just in a year's time click answers on knowledge basis, okay, rag as a solution. We're now doing hybrid search, we're doing advanced chunking, we're doing all sorts of different things, including graphs to understand the document structure and the entities and the relationship between, if you ask a a naive question that's not very specific, we can do a better job answering it. So, um, you know, these are all big things.

Automations is the next layer in there. Um, our automations framework will need to include human in the loop type of things because we know people aren't gonna put these things together and let them run on their own. They're gonna have to have human in the loop kind of touch.

So that's an important aspect of just knowing where we are in the evolution of, of these technologies. Um, and if you guys remember me, I love to show more than talk, so I wanna get Kyle up here 'cause I'm gonna show what we're doing, um, with some of these technologies, uh, that we announced here at Connect. So The question we're always coming up, um, you, you started by saying, um, that customers sometimes are just jumping on the bandwagon, right?

And, and you have a better solution than when maybe the newest whizzbang LLM thing, right? Um, but having said that, then we've got all this new stuff and kind of kind of bouncing off what you're saying that you're seeing from the field. Um, it, it seems to me just from interacting with the different things that I, it's just very obvious that it's, it's, it's kind of a, not even, it's just LLM with something in front of it.

It's not even an agent yet, but they only did it because they can, and it's so user and customer unfriendly. It's very much built and designed. And this has nothing to do with Click.

This has very much to do with your customers. It's, it's built to solve a need for the business versus to help their customers, which is what they think they're doing. They're actually trying to save money or, or, or get gather information and pass it to somebody.

But it's a, it is just a horrible experience for a customer. Um, do, and I, I just don't see this getting better, you know, after what you said with people, they just went, look what I could do. And they do it and it's out there and they checked off a box that they have and genic experience, and it is just really bad and it's gonna come back and bite them.

So, um, how are you just in, or are you guys' click, are you trying to advise people on, you don't need this, you just need predictive ai? Are you, do y'all help your customers with that? And have you thought about saying, well, if you put that agent up there to just gather that information and your customers are gonna hate you, and this is gonna fail, but like, are y'all advised I'm on that too?

Yeah. And you know, Kyle's standing here next to me, he's one of those field people that is very acute at understanding what the actual intent of the customer is and what solution best solves for it. Um, it is also why we determine when we wanted to, uh, help our customers start down the generative AI path last year, we weren't just gonna give them an LLM out of the box and say, yeah, send your data at this LLM, it'll give you really good, no, we went with Rag.

'cause that's contextually grounded in the data that you give to the LM. Um, and so you can know that the answers are, uh, are gonna be correct. And if it doesn't know, as Steven pointed out in our last podcast, it's gonna say, I don't know.

I don't have access to that information, which I gotta tell you is really reassuring for me if I'm a business executive, knowing that okay, there is a limit to what the LMS gonna do. So, and that approach seems really consistent with Click's strategy overall when it comes to data Yes. As well.

Um, not making stuff up seems intrinsic to the, Uh, overall company. Yeah, I think it, it kind of goes into the bucket of saying it's enterprise grade, right? Because, uh, there's a lot of people who can go out and use their laptop to build something.

Uh, it's not that hard to go build a, a rag pipeline on your laptop if you're doing it for, uh, five documents, right? And then you're okay with answers that aren't gonna look okay. But as soon as you start scaling that to thousands or, or tens of thousands of documents, uh, and then thousands of users, you want to have the reassurance that it's not gonna say things that it shouldn't say, right?

Or, or even if, uh, it's giving the right answer. There are certain times when an organization doesn't want that answer given to certain questions, right? Yeah.

Um, and then to go, you know, to your question about the age agent-based stuff, I think Qlik is uniquely positioned to do some of this stuff, right? And to help advise our customers in that direction in the sense that, you know, Josh was earlier talking about some of the acquisitions we've made along the way, and they're really the three core pillars that go into building a good agentic experience. And Nick showed those, right?

So the predictive components, uh, from the big Squid acquisition, which became, uh, click Predict as we announced this week, uh, we have the answers that we saw last year come out through an acquisition, which is the generative experience. And then we have a product that we call App automations, which is a third party, API integration that allows you to do those actions. So when you are asking questions about the agent being something that can take action, we actually have the technology to make that possible.

And I think that gives us quite an advantage, uh, into being able to do this faster and better than a lot of the competitors in this space. And so we're really bringing those three products that we've had for years into a single experience to build those agents. And then, you know, when someone asks a question that should be predictive ai, our agent will be able to build the model for them and give them a predictive answer, rather than us having to, you know, explicitly tell someone, oh, you should use predictive for this.

It'll just be part of the answer, right? The answer to get used predictive, because that was the right solution to solve that problem. Right?

Question from a, uh, persona that we've had, uh, a group of practitioners give you folks feedback on kind of the perception of Qlik in the market versus in their specific organizations. And one of the things I'm struggling with is who's the champion with this expansion of capability for the click portfolio of solutions? Because the market and the competition is noisy Salesforce, I'm seeing overlap for, uh, agent force.

I'm seeing overlap for AWS native services. I'm seeing overlap for a lot of the cloud providers, ServiceNow, everyone about, of building your own agents for internal agent platforms that do much of the capability that you folks are, uh, uh, enabling. What's the click, not just story, but who's the champion that's, that's gonna get you in front of the CTO to kind of raise above the noise?

Yeah. Uh, a couple different perspectives on that. So the first is, uh, I think one of the value props for click for years and to this day remains is that, uh, we are kind of an independent vendor in, in the space, right?

You go talk to a lot of these other vendors, uh, especially the hyperscalers, their goal is to get you to buy as many of their products and com compute services as possible, right? Um, and so one of the things that we've been able to offer for years is that independent connectivity to wherever you're doing your work today, whether that's, and most companies want to be diversifying their, their cloud portfolio, right? They don't wanna put all their eggs in one basket.

Um, and so they're trying to spread it across. And so they need a, a platform that can really make it possible to bring in all those different data silos into one place to provide them kind of a, a central solution to solve these problems. So rather than you needing to tack on an agent to ServiceNow and to Salesforce and to all these different places, or heaven forbid, think about having to move all your data into Salesforce and the amount of development time that would require to, to do all that, because that's, that's quite a, um, a big ask to move all your data into a Salesforce environment.

And so, uh, how we kind of see our position is to be able to be the source where no matter where you're working today, you can grab and hook into all the places that your business is working already and will be compatible with those different, uh, data silos, if you will. Um, I, I, I get the, I get the, I think that's, that answers my question on kind of the value prop, but it is very difficult for a vendor not named whatever their vendor is to, to handle application development or application enablement to now have a different vendor other than the incumbent. So the question is incumbency, how do you overcome incumbency for this net new space you're entering?

Yeah. And, and this I think is one of the realizations from our strategy discussions is this is a very, it's gonna be and is already a very big ecosystem, right? That's why AWS is here.

Um, and it, for us, it's gotta be about knowing where we have strength in that ecosystem and where we do not. And so where we have strength is our heritage around data analytics and empowering agents to enlist that, to create intelligence. So these agent systems can be operational across, as Josh talked about, multi-agent architectures.

Um, so are we gonna build guardrails? No. AWS does that really well?

Do they do security best in the world? Do they have regional plans and rollout? Yes.

Best in class. So that's why the partnership there is really good. We can bring our strength and analytics and data to an ecosystem where they can help us be.

That's why I mentioned MCP and H two A and, uh, the agent marketplaces, that's where we feel like we can, you know, participate, but in a much larger ecosystem that's require multi bill vendors to do these things at scale. So it's, it becomes a, uh, go to market with partner. So if you, you're common customer with AWS and they're already using answers for something that's, you know what?

There's an easy button for this integration that you're gonna do. Qlik has these extended capabilities relationship, let's do that. As opposed to you having to have to, you know, build the story up and build the, the, the champions within the organization.

This is just connecting the tissue. Yeah. And as we, you know, data sovereignty was brought up, it's a big topic.

We can push Amazon in the direction that our customers need. They wanna expand in different regions. So that's where, again, it's a partnership that has, uh, mutual benefits.

It's just a part of knowing what our strength is in that ecosystem. Appreciate. Yeah.

Uh, I'll, I'll double down on, on that kind of strength in analytics, uh, in the sense that I see that's one. And when I'm out in the market, uh, it's one of the areas that I see as, uh, another one of our advantages in this space is that a lot of the solutions that, uh, people are, are working on to deliver these kind of structured data with LLM solutions, um, it's a lot of text to sql. So it's a lot of, they go ask, uh, these agents a question, and it, it, they're trusting the LLM to write a SQL query for them.

And then they're pushing that down to a data warehouse. And there's a lot of inherent risk in that, right? Uh, is it writing the right query?

Is it doing the right joins? Is it using the right tables and fields? Is it using the way that we calculate that measure correctly?

Um, and so we have, you know, over 40,000, 30,000, 40 30, we have over 30,000 customers, um, that are already inside of Qlik today. And they've built these data models and applications where they've done a lot of that work, right? They've, they've built the data models that define the dimensions, the measures, um, the glossary for the business terms.

And so we get to just take that technology and put it on top of that associative engine that we've had for 30 years, and allow people to kind of use that trusted foundation of structured data to have these interactions with the generative AI models. And I think that gives us a pretty distinct advantage over someone that's just, you know, querying data warehouse, um, somewhat blindly, right? Okay, cool.

So, um, last year we, we introduced answers. Uh, we, so, you know, that experience was all about unstructured data, right? We brought in things like PDFs, HTML files, um, PowerPoints, and we were able to ask questions on that, uh, trusted unstructured content.

Uh, the theme that we've announced this week is all around expanding that footprint into structured data, right? So, uh, it's about taking those applications that people have spent decades investing time into, um, in terms of building data models and analytics and dashboards, and making those accessible to the agents where they can retrieve answers to questions from either the unstructured document or the structured document or the structured data, uh, in their applications that they already have today inside of Qlik. Um, and so what does that look like?

So, uh, if, if I flip over here, you know, this is what an analytics application looks like. Inside of Qlik is something where someone has a dashboard, they have built out their visualizations, um, you know, different charts and graphs. There's lots of really good information in here, but, uh, we're getting more and more into a world where the expectation is to not have to go look at a dashboard and find the answer that you're looking for.

You just, there's a lot more of a, uh, people coming into the market that expect to just ask a question in a, a chat assistant, right? On their phones, um, wherever they, they may interact with it and say, you know, what is this metric? Or, show me this, uh, and to get an answer back that's reliable, um, and answers their question directly rather than having to go find it.

Uh, and so we're able to bring those experiences together now with, uh, the new version of, of click answers. So you can see here we have a combination of not only unstructured knowledge bases. So that's what we had last year when we in introduced click answers, uh, where knowledge bases, where we brought in those trusted unstructured sources, but we've now added, um, the application.

So we're able to go and pull in those, those analytics applications that have all of this trusted, um, structured data models, uh, and make those accessible to, uh, the agents that people are interacting with today. Um, so what does that look like? I might ask a question in here.

Um, that's, that's really broad, right? Uh, what's affecting coffee sales, right? So that has a really broad implication, right, that an LLM could infer.

So, uh, I might, I might come in and, and just ask a really broad question like, uh, what's affecting coffee sales? Um, there's multiple implications there that might be some sort of structured data question that's saying, okay, show me the performance of, of sales of coffee. It might be things about, okay, do you have news articles or, or unstructured information that's related to what might be driving, um, you know, different trends in coffee sales.

So you can see the assistant was able, the agent was able to go and find a combination of things now with the new click answers. So the first is an unstructured type document, right? It went through a series, um, of different source documents that were news stories, reports, um, published by analysts around what might be affecting tariffs, um, other kind of macroeconomic effects, and summarized those for me in a way that answers my question directly.

What's new though, is that it was also able to build a chart for me. So it went to my structured data model around coffee sales, and then kind of understanding the impact of what might be influencing that. It put together a chart for me.

It gave me some insights from that chart. So it actually looked at that chart and that data and said, okay, here are some of the interesting things that are inside of that. And it summarized those for me, um, in a couple of bullets even better, is it showed me the assumptions that it's making.

So we talked about that explainable AI concept. That's something that's really, really important for us at Qlik, right? Is not only giving you answers from these different AI solutions, but being able to help you and confidence that the answer is what you were looking for and using the right information.

So under the assumptions, you can see it says, these are the fields that I used. Here's, uh, the filters that I applied right here are maybe the different ways that I sliced and looked at this data. So it's able to break down how it was able to logically reason through what you asked to produce this visualization here.

So like, and just to be super clear on that, is that chart like one of the charts from the app, or is that a generated chart based on the data model that the the app? Yeah, great question. It can be either or, right?

So it might tap, if you have kind of a saved chart that is, you know, part of your ma like master items in the application, yeah, it can retrieve that. But part of the, the power of what this able to do is generate charts on the fly, right? So What we're talking about here is faceless apps, really.

Yeah. You can build faceless apps and start to use clicks capability to generate the things that you need. Absolutely.

Based on the output that you got, the answer that you got, how do you put it into a Word doc? Or it was like copy button or, Yeah. So that's that, that's, that, that's really that next phase of in the agent world, right?

Which is how do I take action on this? Right? And so, um, here in a little bit, I'll show you how we'll move into that, that phase of, of automation and, and application integration where you can say something like, great, okay, turn this into a, a Word doc or a PowerPoint, or, or do something that's, you know, external actions.

Um, but, but that'll be kind of the, the, the next phase of the click answers, uh, agent improvements. Yeah. Before you move on to that, explain to me what a knowledge base is.

Yeah. You know, two paragraph. Yeah.

So the way I think of a knowledge base and explain, usually explain it to most people, is it's my trusted source of knowledge, right? So rather than the general knowledge of the model, which is trained on the internet, right? Which there's all sorts of problems that can happen there, information out of date, untrustable information, um, it is a knowledge repository that I have curated as an organization with trusted sources.

So it might be, in this case, a combination of a data model that I've produced, um, in my analytics team, um, and or documents from my SharePoint sites, my S3 buckets, uh, my Dropbox that I've said, these are the right documents that people should be using to ask, ask, and answer their questions. Right? So, uh, you mentioned that one of the things that's new is the ability to go after structured.

Mm-hmm. Are we, I don't want to jump ahead, but we're a knowledge base would also be able to access structured Data. Yeah, absolutely.

So what you can see here in, uh, in the content, um, that we've broken down is there's, that have been made available to this agent, right? So this agent can access three different knowledge bases. So it has basically three different knowledge repositories that someone has curated for me around these different topics, right?

Um, some of them are around coffee tariffs, some of them are around training for the employees in the coffee shop, right? Policies and manuals from hr. And then there's that third row is an application that's the structured data model that someone has already built for me.

Um, and it was this app that we just, uh, looked at here. So someone built this app, right? They took all the time to connect the data sources to kind of make the right connections to say, these are the measures and how you should define this metric.

So when someone asks about sales, this is how it should be calculated, right? Someone's invested all that time and effort and it's in, and we can now expose that to the agent to get the right answers. I wanna go a a little bit deeper.

Yes. So I understand is that knowledge base and what they've built, is it, is it based on, say, a SQL Server database, um, Azure, some Azure Source, an Oracle database, um, a series of flat files being accessed to CSVs? Mm-hmm.

Is that what is the data model? Yeah, it can be any of the above. So any of the above.

Yeah. So, uh, we have at click, we have something called the associative engine, our, our proprietary data modeling technology that brings together those disparate data sources. So Is that Talend, as we've heard, uh, It Talend can be used as part of that.

The, a associative engine's been around since when Josh was talking about the nineties, that was our core technology that the company was started on, right? And that's been around for 30 years. Good.

That makes it clearer. Thank you. Yeah.

Super important question. Like all of the questions, all of yours, anyway, you said it, the, uh, we asked a very basic question, but this is the challenge of using LLM as the interface to really rich data on the backend. If I ask that question three times mm-hmm.

I might get three different answers. So, uh, I love the fact that the, uh, we have explainability, so I know what the answers based off of how do we direct customers to better to ask better questions? Yeah, great question.

So one of the other things we've introduced, um, in the new click answers is this concept of something called guidelines, right? Um, so a guideline is the ability for you to add a couple things. One is to kind of introduce additional instructions to make sure that when someone asks certain topics or different things to continuously go down the right path, right?

So maybe you wanna, uh, you know, that someone might ask the question three different ways, and you wanna make sure it goes down the right path every time you can provide those guidelines to the agent to do that. Um, you can also load in here, you can see number three, um, is example questions, right? So you can kind of seed these, uh, example questions that help the agent to be able to provide the right responses consistently, um, across the content.

And then you can apply, uh, these guidelines to different parts of your content. So you might say, these guidelines apply to the structured data components. These guidelines apply to the unstructured knowledge repositories, right?

So it allows you to really kind of get in and, and add an extra layer of protection to get that enterprise grade consistency and quality in the answers that are coming outta the agent. Okay. And Is, is that like an additive thing in terms of what already exists with click answers?

Yeah. Because obviously if you built something with click answers, APIs at this point, it just appeared in click application automation not too long ago, and you don't want to be like ripping all that out and starting again with a new architecture. So how does that set Yeah.

This, this is all just incremental improvements to the click answers offering that's already there today. So, uh, correct. Honestly, it, it came from a lot of feedback that we've, we got in the market from our click answers customers is they said like, look, great, this is solving a lot of our problems, but there are these, these fringe cases in which we want to be able to provide additional instructions, right?

Because we want it to act a very specific way when someone brings up this topic or brings up this conversation point, we wanna route it to structured data or to, to unstructured sources to answer that question. Um, so this provides that kind of extra layer of customization that some of the customers we have today are asking for. So just to be crystal clear on that, if I've got a knowledge base that has 12 documents in it that has correlated a lot of my unstructured stuff, I can drop an app in there and then connect that.

Absolutely. That's the type of process that we're talking about. Yep.

I was super impressed with the way click answers was a drag and drop experience to begin with. So that's, I think that's really important to continue Absolutely. That paradigm moving Forward.

Yeah. Thank you. Um, I, I, from my perspective, we see a lot of, especially the hyperscalers building from their starting point is the developer, right?

And then they, they kind of back into a end user friendly experience. Um, and from my perspective, it, it's not always the best. We kind of work the opposite way, right?

In that we like to build for the, the, the business user first, and then we add the kind of, as those users start adopting it, we start to add the more advanced capabilities over time to, to make it more available to those who have edge use cases to, to kind of get more advanced in the product, right? So I think that's one of our kind of unique offerings is that when the business comes to us and say, Hey, we're trying to do this, we just don't have the resources to develop it internally with, you know, uh, hyperscaler services or with APIs or all these other things, um, they're able to use our product outta the box because it's so user friendly and intuitive for, for those use cases. I think it's big for the question I asked before as well, because like, that's what I'm really seeing is like, you can build it and you've got the pieces, and we can do it technically, but it's not friendly to the business user case.

It's just some of them are not, and it's going to have a really opposite effect on what they were trying, you know, what we're trying to do with AI in general. Yeah. And you know, it's really some of the stories that you've heard this week around from our customers that you'll, um, and this, this click answers product is those kind of stories, right?

Of like, I was unable to do this for years because the services that were available were designed for architects and engineers and developers, and I'm a smaller organization that's been trying to do this without the resources. And then Qlik brought me that, let me do this outta the box matter of hours, right? Rather than weeks or, or, um, months even to get some of these, these products out the door.

Okay, great. Uh, so that's, that's kind of the agent approach, and then I promised you we kind of evolve into what is the next phase, which is the action piece of an agent, right? So, um, again, fortunately, we have already had the components that make it possible to take action in third party systems with, with these agents.

And so, uh, the next evolution of click answers will be, uh, the ability to say, okay, great, you've given me the answer I'm looking for, I found my either from structured or unstructured data, now I need to actually do something with it, right? I don't wanna get, if you will, with this answer you've given me, um, I need to be able to send this as a teams message to someone I need to generate a PDF that I can send in an email somewhere. Or maybe I need to create a, um, a event in Salesforce, right?

Or update a Salesforce opportunity. So with click app automations, we have the ability to integrate with hundreds of different third party tools and APIs. And again, as usual, with the click experience, it's all drag and drop.

So you can see, um, it's a completely codeless and user's able to configure all of these different interactions. In this case, I have an example where I have a teams channel set up. So inside of Microsoft Teams, you're not gonna bring all your users into a, into Qlik for questions.

You need to be able to put these agents in the hands of the people who are using other tools in their day-to-day operations. Uh, so with let's say Microsoft Teams, there's a channel where people want to just ask the question that they have about, you know, whatever you've built your, your agent around. Uh, in this case, we're monitoring the teams channel for a new message to come in.

We're interacting with that agent, um, in a series of kind of automation steps, and then we're sending a reply to their message, right? Uh, so I might come in and just show you this working, right? Uh, if I come in and go to, uh, this teams channel and say, uh, ask a question, I'm just gonna ask the same one I asked, uh, about 10 minutes ago, and just post that in the teams channel.

And then I come over and we'll just manually trigger this automation. But we can schedule that to, to, to set off when someone asks a new question. And so it's gonna go through those series of steps, it's gonna work through the different APIs required to take that action as an agent, that autonomous nature.

And then when I go back to teams, now I have a response to the question. So it went to that trusted structured data source, right? I asked the question, how many people have registered for Kyle's session?

It's a very ambiguous question, right? It can, that if I went and asked that to just an open LLM, it would be like, I don't know who, who's Kyle, what are you talking about? Right?

But because I've connected information around the click Connect breakout sessions for this week's breakouts, and then asked that question, it came back and said, 50 people have registered, uh, for the session featuring Kyle Jordan. And I just wanna go ahead and say the capacity was 50. So that's a sellout.

Okay. I just wanna make sure everyone's aware, aware of that. But, uh, so I can ask that question.

Uh, and then just like, uh, when we have, uh, in our click answers experience, I just, I literally flew yesterday from, from Wales, so I've been in Wales that really Fast, right? Then that's a frightening demo in that you're representing as yourself in the demo. Yeah.

And so the, the implication being is that, um, you know, one of the, the company I used to work for, we did the same thing for a couple people in product management, and they created their own little people. Yeah. And the thought was is like, well, Don doesn't sleep now.

Don's available 24 7, 3 6 5. Exactly. Beat up all you want for questions, you know, post product management's product management.

Is, is this the intent or was it No, this is just easier to do the demo this way with my name. Well, I, I like to make it as real as possible, right? Yeah.

Okay. Uh, I I want you to know that I'm not using generic demo content here. I'm making stuff that's specific to me or to the people in this room, right?

And so, uh, the only reason I've been able to be in Wales for two weeks on vacation leading into Click Connect is because I've now automated myself. Yeah, There's A bill, I'm not sure if you remember the Dilbert one, where it was Wally, and he starts basically sending emails from the past several years, and no one knows that it's, I ha I have a, that actually leads to question I have. So what it seems like there's also another thing to consider is a agentic lifecycle, right?

Because after Click Connect, this isn't going to matter anymore. It's just a bunch of extra things hanging out in teams and hanging out on click. So, um, are there any suggestions on how to, uh, make sure that information that the, the agent is addressing is, is constantly current?

Is there an, you know, are there suggestions and advice you give to your customers on how to, um, how to update or delete the apps when they're no longer necess? Yeah, it's a great question. And it's something that each of these progressions in AI is uncovering, and then also why we happen to have, through like the talent acquisition and some of the other acquisitions that we've made along the way, brought in a lot of data quality, uh, and governance components to the platform.

And, and that's important because all this does is open a can of worms around data quality and governance, right? So when we introduced Click answers, we saw this huge wave of customers saying, oh boy, I just figured out that all my unstructured data is full of out of date documents, documents with the wrong answers. Um, the same name document 10 different times, right?

And they have different information in them. Blackwood, You're triggering me right now. Yeah.

And so and so because, you know, it's exposed all these things. Again, we just happen to be in a really fortunate position that we've made strategic acquisitions over the years that give us the ability to help address those concerns. And so as part of the kind of the data integration platform, we have the ability, and we've introduced these concepts around AI trust scores, where you can continuously monitor your data sources to know how ready they are for ai, how much the information can be trusted.

So, um, all those same concepts apply. We're just helping people open that can of worms. Fortunately, we have, I don't know what a good analogy here, we have something to put on top of that can of worms, or at least to cast the worms out, out and catch some fish with it.

So, So if a Click customer wants to build their own agent, can they do that within the Click framework? Or should they better go to a third party and build their agent? I think most people would be able to say that they can use the, the click answers experience to build those agents between, uh, the structured and unstructured integration now, and then the ability to interact with third party APIs.

And again, if it's not, if we don't have it in our prebuilt connector drop list here, we have open, uh, rest API blocks that you can drop in. So you can call any API that you want, even if we haven't built an, an easy drag and drop connector where you can just drag the action onto a canvas here, you can specify your own API. So maybe a company has their own API they've built internally, and they want to be able to interact with that.

API, they can do that directly from this experience. So I think people can get about 99% of the way there. Um, there might be, there's always, there's always fringe cases where someone says, oh, I wanna customize a little more.

Uh, you know, we can provide a lot of the support services in the background to kind of help get that from A to B faster. Uh, but I think we'll find a lot of people can solve a lot of their problems. What's your strength in the versioning?

Uh, so you have a history tab? Yeah. Um, so is the history indicative of like, um, you're taking a snapshot of the environment with all the, uh, relevant elements or, uh, where, where does a, where does a backup and recovery begin and end?

Yeah, I mean, pretty much any, um, of the, the products that you see here are, can be version out. Okay. Right?

Um, so not only can you see the logs to see what's been running and, and executions. Yeah. Um, but in our Click Predict platform in answers in automations, you're able to kind of revert back to previous versions, kind of see how things have evolved over time.

So you can always kind of manage those experience. Everything's this, all this AI stuff is very iterative, right? Um, you're constantly gonna be evolving it and changing it, and you might want to go back to version three to, to begin version six, right?

Um, so there's always that ability to kind of create that, that timeline as necessary. Thank you. Um, as much as I love to keep talking about agents all day, um, like Nick said, a lot of the time the agentic and generative talk actually devolves, I mean, not Devolves, but reroutes back to predictive ai, right?

Um, in the sense that this is the right solution to the problem. Um, and so one of the things, some of the things we've introduced over the past year that have really evolved the space are almost, I don't wanna call 'em boring, but it's really the groundwork required to do predictive ai, right? And that's around data quality and preparation, um, and transformation.

And so we've introduced, um, these no-code data flows. So the ability to 80, say 80% of data science work is around data, the data prep, right? And getting the data right for the machine learning and, and the, um, the model training.

Uh, so we've introduced this concept of data flows where you can quickly build the data sets you need to build your machine learning model. So you can see that's completely no code, but it uses the power of the click engine that we've had for 30 years. And our proprietary clicks script language in the background.

So what's happening behind the scenes is we are generating Clicks script for you, and you can even pull that clicks script up and customize it if you want to. Um, but just dragging and dropping these blocks on here does all that powerful clicks script generation for you. And then that gives you the ability to pull into a, um, uh, a experiment, a click predict experiment.

Um, and then we've also just, uh, recently announced, uh, the, the table recipes, uh, concept, right? Which, let's say I did my data flow and I realized, oh boy, I got all these columns here with NA values in them, and I need them to be numbers right? For my model to work correctly.

It's really easy with table recipes to come in and say, you know what? I just need to do a replace on those values. If I can click and say, place all those NAS with zeros, and then just like that, um, I've transformed that data set to have the right data structure for me.

As you can see, all those NAS have turned into zeros now. So it allows me to quickly iterate over data sets to get to the right version faster without having to kind of keep going back to my data warehouse, my SQL query, and, and kind of doing that back and forth. It's just in a GUI where you can make quick updates and quick improvements.

Um, and then the last thing that I'll, I'll touch on, uh, here, that is exciting for us and, uh, the, the next big improvement that, uh, we are bringing to the Click Predict platform is the, the introduction of time series capabilities. So we've heard a lot of people asking, uh, for the ability to do native time series modeling, uh, with their data. Because as much as we want to preach the benefits of, of predictive ai, a lot of times it goes back to someone saying, I need to do sales forecasting better.

Right? And the really, the best way to do sales forecasting is to use time series models. And so, um, with this, we'll be introducing a lot of deep learning algorithms that will be part of the Click Predict platform.

Um, and so you'll have the ability when you're configuring an experiment now, um, to quickly select that you want to use a time series, it will ask you what your date column is, and then it will allow you to kind of group your data, um, by the different groups, uh, that you may wanna, so it might be like by store or by department if you're doing a sales forecast. Uh, and then we'll train some deep learning algorithms to build a, a true native time series model. So that will be a capability, um, that we have in the, and That model that you're referring to would be your model or someone else's model.

Um, it, it will be trained on your data, so it'll become your model for your data and your sales or whatever your tar, your target is that you want to, to predict. Yeah.