Upsales Technology AB (publ) ($UPSALE)

Thank you, everyone, online for joining us today. My name is Daniel. I'm the CEO and Founder of Upsales, and we are extremely excited to present our new AI agents for you today. I will be joined by David Dernuff, our CTO; and Kristina, our CFO, and then we will end with a Q&A. And I wanted to start just by talking about what started this whole project. So we, by accident, listen to a podcast with this guy, Jason Lemkin, who is somewhat of a legend in Silicon Valley, a tech entrepreneur who now runs something called SaaStr, which is the biggest SaaS or software event, one of the biggest ones in the world. And he was on a podcast talking about how he had a couple of really well-paid people on his sales team who just left with 1-day notice. And he started thinking about, okay, my business is not growing. my sales team is not as predictable and stable as I wanted to. So he started experimented with AI agents. And around 1 year later, this is what they managed to do at SaaStr. So they replaced almost 10 people with AI agents, and they went from negative growth to almost 50% growth. And it's a very interesting story. It might not be applicable to every company, but I highly recommend everyone to listen to it. So what happened at Upsales when we listened to this podcast was that our first thought was that we can do this. We can help our customers do the exact same thing to reach incredible levels of efficiency and really use AI to be more than a chatbot or an assistant. And actually, at the same time, my colleague, David, was in the second iteration of building the new AI assistant in Upsales. And the idea was to build an AI assistant, a chatbot. And after we started talking about this, it turned out to be something way, way bigger, which we are very excited to present today. So I just want to talk a little bit about this also. If you have ever tried launching an agent or use AI to do anything useful in your sales team, you might have some bad experiences that the AI doesn't work as good as you want it to. And the numbers back it up. So most of these projects show no financial return. So it's -- the data out there of most companies who have tried deploying agents are quite depressing. So -- and our view is that this is not a technology problem. There are cases like this and Lemkin, there are many more out there. The models are good enough. The problem is it's a workflow problem and an iteration problem, and I will talk more about what I mean by that. So I think what we see is that the #1 mistake when deploying AI agents is to treat it as something you set and forget. So you treat it as a traditional piece of software or a traditional system integration. And the problem, as many of you know, is that AI models are nondeterministic. So you know that they can do something. It will hopefully be useful, but you don't know exactly what will happen every time. So what we see is that the companies who are able to get the kind of results that Lemkin was talking about, they do 30, 40, 50 iterations. So it's hard work, tuning these agents, giving them feedback and making sure that they are doing exactly the right thing. And the most companies give up after iteration 2 or 3. So you try it, you get a little bit encouraged at start, and then you get a little bit discouraged and then you come to the conclusion that AI doesn't work. So it does work. The models are good enough, but you have to have some stamina and you have to go through the number of iterations to make the AI do something useful for your specific use case. And [ David ] will talk more about what this means in more detail. So why do we think we can do this? Why can Upsales be the company delivering the same kind of results as Jason Lemkin is talking about? So I think number one, our very long track record and experiences of nerding around sales processes. We've learned from the best. We interview the best people in the world, and we have worked with thousands of companies implementing our sales and marketing software. We've seen what works. We've seen what doesn't work. And what we're doing is we're taking all of this knowledge and putting it inside these new agents. So generic AI is generally useful for some things. What we are building is something very specific for the core sales and marketing use cases our customers have. And the third reason why we think that we can do this is we have had an almost 20-year track record of working with not only selling the software, but also selling the data because the biggest problem in our field in sales and marketing tools is that the data is broken and your strategy to get the right kind of data is to have users enter it manually into the system. And if you tried it, you know it doesn't really work. So we've always worked with third-party providers to make sure that our tools and system contains the real quality approved data, data that you cannot scrape online. You have to buy it from a third party. And when you have all of these finely tuned agents together with actual data, it's a lot easier to make them do something useful. And this is just one example of many, just to make it a little bit more concrete, what we see our customers will be able to do with these agents. So we're moving from a chatbot waiting for a prompt and an assistant waiting for a question to autonomous agents working 24/7, taking initiatives, doing stuff, preparing stuff for the team, contacting customers, booking meetings. So we move from supporting the work to actually doing the work. And this is one example of one of the agents we will launch. It monitors every single customer you have 24/7. It keeps track of everything going on. And it finds the early warnings that usually exist before you lose a client 6 or 12 months later, and it gives you a strategy and a proposed plan and all of this, they never take a break, they never go on vacation. They are running 24/7. And the elephant in the room, the question that is too too polite to ask out loud. Do we think this will replace the sales team? I don't think so. And I agree with what Lemkin says in his podcast. I think many of the agents that are out there today that are doing useful sales work, they are better than the bottom half of salespeople. They are better than the mid-pack, but they are not better than your top performer. So we see that the work of the agents is enabling really good salespeople to produce even more by removing the non-meaningful repetitive admin work. But the ones who lose are kind of the ones whose value is the administration because that kind of work is quickly being replaced by different types of agents. And before I hand over to David to show you what it looks like in practice, I think when I talk to -- I'm out seeing our customers every week. And when I talk to them, I think that a couple of things happened in the last 18 months that was not really true 18 months ago. The models are good enough now. They were not good enough 18 months ago. The cost of running them have dropped significantly. If you look at the kind of reasoning power and intelligence you get for $1 today compared to 18 months ago, it's a huge difference. So you can experiment and you can be a little bit creative with AI without kind of breaking the bank. And I think the platforms that will win are platforms like Upsales, where you connect to real data and make sure that these agents are not trying to go out on the Internet and just find something and hope it's useful. And I think the CEOs and the sales leaders I talk to who are winning, they have like a 24-month plan, and they realize that this is a sprint. And if you're not doing it now, your competitors will. So what we recommend when we talk to our clients is to not try to have a huge, super ambitious AI project, find something, find one process, start somewhere and just iterate until it's as autonomous as it can get and then move on from there because I see a lot of companies kind of getting stuck in the planning phase. I think I will stop there and hand over to David for some of the real stuff.

Thank you, Daniel. All right. Hello, everyone. My name is David. I've been working at Upsales for, I think, over 15 years now. I've been CDO of Upsales since April of this year, and I still work hands on in the code base. And I want to continue a little bit and talk about what Daniel said. So I want to talk about what I've seen happened for my line of work in the last 12 months. Okay. So about 12 months ago, we used AI, of course, during coding, but AI was nothing more than like a better auto complete. And today, I can write a spec. It doesn't have to be very detailed and I delegate it to the AI, and it completes the feature while I just go grab a coffee basically. So I come back to work and most of the time, it has completed the task very well. So the thing I get paid for has changed. That's not like a prediction. It's the reality. That's happening now. So nobody at Upsales is writing code by themselves anymore. That's just not how it works anymore. AI does that. And I think -- I mean, we see it first as developers because we're closest to the tech, I guess. And I think the harnesses around the models and the tooling for coders are probably like at the cutting edge, which is not the case for most other lines of work yet. So I think the same shift is coming for all white-collar work basically, and it's coming really, really soon. It's not going to be 5 or 10 years from now. It's going to be starting this year, next year, yes. So that's the real AI revolution. And like Daniel said, the models doing this kind of work, they're already as capable as most humans by any reasonable metric, I would argue. They can reason, they can articulate their motivation and they can make decisions. So there's an obvious question here. If the machine can already reason at the level of an educated professional, then why does everyone's work still look basically the same anyway. I'm sure most of you or probably all of you has used ChatGPT. Some probably use it every day or some other provider. But it hasn't really changed how your work gets done. It's just a tool that you reach for and then you use it sometimes, but it's not the foundation of the business. And why is that? I think that gap is probably the most important question in tech right now. And I'll tell you what I think the answer is. So the model is not the product. The model is basically just the brain and the brain alone can't do much unless you provide it good context, memory and a way to take action. So that's what we've built. It's not a smarter brain. It's basically the the body for the brain in sales. So we built a platform, a foundation where AI isn't a feature. It's basically just how the work gets done. And it has to solve 2 specific problems. So a model is only as smart as the data that you give it. If you ask a generic AI, should I call my customer today, it has no idea. It doesn't know who you are. It doesn't know what company you're talking about or anything like that. But obviously, the information exists and you could just taste it into ChatGPT or your personal favorite provider and you would get a pretty good answer. But the bottleneck isn't whether the context exists or not. It's who assembles it. and when. So no salesperson is going to pull like financial data, news data, activity history into a prompt to an AI for every decision they make because they have a job to do. So what we've done is we've built a layer for them that does this for them. It knows the context needed, it fetches it, delivers it at the right moment. So the user will never have to think about that. And like Daniel said, at Upsales, we've been collecting this kind of data for over 20 years now, proprietary company data, financial data, credit data, monitoring of news, et cetera. And that's our edge. And I think that's really hard to copy. The next problem is the intent. So I'd argue that most people don't know how to write a good prompt for the AI nor should they have to. I don't think anyone should have to do that. I mean I write prompts every day. And it's a bit annoying to be honest. And especially annoying when it doesn't understand what I mean. So asking a salesperson to learn how to prompt, I think that's just bad product design, honestly. So with our 20 years of building sales software, it means we know what good sales work looks like and bad sales work as well. So we know what decisions move the deal forward. And we've baked that information into the agents themselves. So our customers do not have to teach our agents how to sell, even though we might encourage them to problem in one direction or another, but they already bring 2 decades of best practice with them from day 1. All right. So even if you have a good platform, someone still has to ask the AI to do something. This is basically what Daniel talked about before. You sit down, you type, you get an answer. The human is still the trigger. But the obvious next step is an AI working without being asked to basically. So you set up an agent, it has a job, make sure every salesperson on the team has at least 10 leads to act on at any given moment. It will silently watch and once the lead count drops, it will act, find candidates that fit the ICP, draft outreach and surfaces it for approval. So nobody told it to do that today. It's just doing its job basically, like a good colleague. And like Daniel said, I mean, this doesn't mean salespeople stop prospecting. It just moves the work, basically the grant work. So they stop spending their morning searching LinkedIn. They start spending it deciding on which of these 10 leads should they focus on right now. So the job doesn't disappear. It just gets better. Okay. Let's see. Let's try to make this a little bit concrete, though. I'll walk through a salesperson's workflow here. So currently today, they need to prospect, do outreach, have the meeting, recap the meeting, maybe update some data, propose the next -- maybe they send a proposal and try to close the deal, right? So some of these steps will happen multiple times, obviously. But yes, so with our platform, the agent will do the prospecting. It will propose outreach based on our sales experience and also based on the voice and tone of this particular salesperson. And once the meeting is set, it will help you prep for the meeting by giving you a brief with -- yes, maybe an agenda and whatever details about the deal you need to know at that time, basically. And once you've had the meeting, the agent will send an automatic e-mail summary. It might draft a proposal. It might update the deal data or it might add a contact based on what happened during the meeting. And at the same time, we have agents working on the platform, making sure your data stays clean, everything is updated automatically, duplicates are flagged and forecasts are adjusted. And all of these things need to be -- needed to be handled by a human person up till very recently, most often the salesperson. And now agents handle them. And the salesperson does the parts that needs them, the conversation, the judgment and the relationship. All right. enough talking. Let's see if we can get this going. Is this yours? Daniels. Let's see if I can get a log in here. Okay. We'll start over here. So this is basically the start screen. It looks like any regular AI assistant. And obviously, it works like that as well if you want it to. So in this case, I'm going to ask it to show in my pipeline. So it puts up some data from the CRM, cross-referencing activity history, checking external signals and computing a risk score here. So I get a basic brief pipeline summary as well as some deals that are in risk at risk. And then I ask it here, okay, actually, could you send this to me every Monday? And it's like, yes, sure. I'll send it to you every Monday. And then I'm like, okay, but could you also flag anything that looks weird in the -- when you look at the pipeline and make sure to notify me about that. And then it says, yes, sure. I'll watch for anomalies, sudden stage changes, unusual deal slippage, et cetera. And you can obviously just ask it again to be like, yes, I don't care about deal slippage. I don't care about that, just focus on this. And then you will get this report every Monday basically. So I think this can probably replace a huge part of our current analytics platform, but we'll see. So Daniel and I talked a lot about the prospecting and outreach. So we have an agent working during the night here and it found 4 candidates for us to review, click it, you get a list here of the prospects. So Tink, Klarna Group, Trustly, Lendify. And we can see that it found some news about it and added the credit score from our proprietary data and the contact person we should contact and then it writes a draft here. And I can just approve and send, read this one as well, looks good, approve and send. I'm going to skip these ones for now. Yes. And then we have the data hygiene. So maybe the boring agent, but we have this cleaning crew agent. So it runs a sweep during the night, looking for duplicates, maybe like opportunities that lack contact persons or stakeholders or find deals that have been stuck in a stage for long, and you can click it and see what it proposes you do about it basically. So it gives you reasoning what to do, you can approve or you can reject or request changes. Same for all of these. And then obviously, like you can ask the chat to do basically anything. And if it thinks it's going to take a long while, it might decide to run it in the background. And if you know that you don't want to discuss anything with the agent, you can just tell it up here by giving it the task. So in this case, I have, okay, it's very ambiguous and not very good task, but that's the thing. People just write whatever is on their mind. So check and see if we know anything new about Stega then you run the task and it starts working immediately. So we're looking at CRM records here, looking at our data and then it goes on the web and fetches, see if there's any news on their site basically. And then once it's complete, it synthesizes all this information into a short summary here where you can read, okay, they had a Series E closing last week, new VP revenue -- VP Revenue Operations hired. and headcount is up, like, okay, I don't care about that, but you get the point. Yes. So these are just a few examples of what we think this can bring to our customers basically. Let's see if I can switch back to slides. All right. So what I showed you right now is just the foundation and the same architecture obviously extends a lot further. So today, the agents do work alongside the conversation. It does drafts, recaps, preps, follow-ups. And tomorrow, the agents will handle the work around the work. We will generate like proposals with the correct pricing, correct branding, legal terms. sales material on demand prepped for this specific client, agents that know when to ask a manager for approval, when to escalate the legal or when to involve another colleague. So I think also in the near future, we'll have agents that don't just work around the conversation, but they will help during the conversation. So basically live coaching in the moment, in the meeting, suggestions in real time. And going further than that, I mean, I think the limit is just your imagination. And I think we'll find many really cool use cases going forward. All right. So I'm about to wrap up. I think a lot of people might be nervous about AI, and I understand that point. I mean the press talks a lot about what AI might take away from us. And I want to leave you guys with the opposite framing because I think it's more accurate and also more interesting. And just like Daniel said, I think the work is just moving up the stack basically. So the repetitive parts, the boring parts, the parts you do the same thing every week, those are going away. And what's left for humans is what's always been the most valuable anyway, the judgment and the relationships. So I think the days of a salesperson spending their day updating records and chasing dead leads, those are gone. The they'll be spending it deciding which of the leads to pursue instead and how. And that's a better job and a more valuable one, and it's the job our platform is going to make possible. I think the companies that figure this out first, the harness, the context, the initiative will be the ones defining business software in the years ahead, and we intend to be one of them. Thank you. And here comes Kristina.

Thank you. Hello. I'm Kristina, and I'm the new CFO of Upsales, and I've been here for only a couple of months. So I'm not going to share my 23-year-old story from the inside. I'm going to share what I've seen since I came to Upsales. So the first thing I noticed was how this company is actually built. So I'm going to talk about 3 things: how we work, why we're independent and how we see the market. So this is the foundation. We have been profitable every year. We have no debt. We have well above 90% recurring revenue, and we have had organic growth since 2003, so since start. And all the investors are familiar with these numbers, but I want to spend a minute talking about what this does to our customers. So by building the company this way, we can stay close to the people who actually use the product. And we haven't changed trends. We haven't made big bets. We've listened to our customers and solve real problems year after year. And this is also shown in how we deliver the product without a big consulting team in the middle. So I think that's what makes us unique. And the discipline isn't something we just put on the wall. It's our operating model. So 23 years of running a profitable business teaches you something. It teaches you what works and what does not work and what to leave behind. And second thing I noticed is the independence. It's not that common for these kind of companies. So we're founder-led. We're listed, and we have never needed to raise capital, and we have no exit clock. And we also have a founder, CEO and largest shareholder all in one person, Daniel, and he's been all in for 2 decades in Upsales. So we have no fund waiting to flip. We have no bank telling us how to run the business, and we have no outside agenda, and that makes us pretty special. So for investors, this is a really solid financially stable business. And for customers, we really do our decisions based on what's right for you over long term. So this is where it really matters, I think. So our old business funds are new one. So the core business funds. We do not have to raise money to chase the AR market and we do not betting the company on agents. So we have a profitable debt-free recurring revenue business that pays for what comes next and what David has built. So it means we don't have to choose between protecting what's already in the company and building what's coming next. A lot of companies do need to choose, and we do not have to. So the market situation. There's a narrative right now that SaaS is dead. A lot of people are talking about SaaSmageddon and that AI takes over. Truth is way more interesting than that, I think. We do not think that the market is shrinking. We think it's shifting. So the big players in this category were built for big deals with many partners involved, but that makes their projects slow and expensive. So the architecture that works and that builds our companies -- our competitors also is the things that trust them today. So AI is changing what customers want to buy. They do not longer want big projects or long rollout or small like partner army of partners anymore. So they want something that just works for their business fast today. And that's where we come in. Yes. So companies have bought business softwares for many decades for one simple reason, you can't build everything yourself and you can't do that with AI either or you can, but do you want to spend time on it? I don't think so. Tools that run your business are worth paying for the ROI is still there. And no company will stop investing in tools that run their business or their revenue. They are too busy running their own business. So to wrap this thing up, I want to just summarize with 3 things. We have built the company with discipline quite hard, and we have also kept it independent. And the market is now shifting to something that suits us very, very well in the future. Yes, let's go to the Q&A, Daniel.

All right. So we will open -- Inge, do you have the questions from online, right?

Yes, I do.

Yes, it's possible to use the chat online if you want to ask a question as well. Do we have any questions in the room?

Where did you start? Like what was your starting point?

Yes. So my starting point, I think like many entrepreneurs, it's kind of a combination of coincidence and maybe some courage. So I was actually working at a company who didn't have a tool like this. And I'm an old developer myself since I was 13. So I built a tool. And by accident, I was on a client visit actually at the company, someone here used to work at who became my first client who needed something like this. So it was humble beginnings, but I think that culture is still an upsales to like grow with each client and do the hard work every day.

Yes, some questions regarding the AI and the road map going forward. We have had a few -- used a few AI agents from you. And my -- first of all, it's nice here. I don't know, nice, but I mean, you need 20, 30 iterations for it to work. That is also good for you to bring to the product with the customers. We weren't told that. So I mean now we're scrapping on. I mean, the users have lost faith in it and so on since it's providing sh** answers and so on. That's one thing, just something to bring with you. Others, what do you think about for us, I mean, our view on AI agents and transcription and so on when you have a quite small niche business, it's really -- I mean, the transcription choosing words that doesn't exist and so on. It's really -- I mean, how to build good agents. I mean, I want to put our lingo and so on into it, so it can learn from that because otherwise, yes, the business -- the users are losing faith in the agents because they are just providing shify recaps and all that is just months. Any reflections on that?

I think I can ask -- I can do half of the answer, and I think David should do the other half. So I agree on the lingo. My favorite one is Upsales. It's always Uppsala in our meetings sometimes. So I know the problem. I think Dana will talk about how we solve that. But one interesting thing is like how fast the market is moving because the platform you're talking about, I think, is the agent platform that we launched like a year ago. it's a white label partnership with another company, a very fast-growing automation company. And we were thinking when we signed that, should we sign like a 1-year, 2-year, 3-year, -- it's like it's so strategic. So let's sign a 3-year contract with this company. It's almost already obsolete, I would say, that platform. because what we saw is that technical products require too much from the typical customer of ours. So I think that's one of the big reasons why we're building what we're building now where you get your agents by talking to the platform without having to become a technical person or use an engineer. And I think with regards to the other part, I think you're better to.

Yes, maybe. I mean, I'm very sorry that, that has been your experience. And I want to be clear that the agent platform we're talking about today, like said, it's -- they have nothing in common. And one of the issues is, like Daniel said, like you need to iterate. We need to iterate on the agents, and they need to learn. They need to have a memory, and we need to be able to provide like customer-specific context and also for specific agents, we want to have to provide like agent-specific context as well. But I think one of the most -- the best way to improve this is to have -- to let the agents have a memory. And that way, you can say never translate that to this again. We talk like this, and you should never do this. And that way it learns over time, and you can make sure to never have it repeat the same problem again. And it's easier to do this than you might think. So we have done it. It's going to -- you're going to see how it works very soon, I would say. So hopefully, this will be a much better experience than what you had last time. I hope that answers. We will see.

No. But I mean, of course, it's important in the projects to really stress the importance of iterations because I mean, typically, the users out there, they give it a try or 2 if it doesn't work...

Yes. I mean I think -- I mean, we will provide a lot of agents, and they will be trained already, like we will iterate on them internally and make sure that they are good enough. But then in order for you to get the most out of them, once you like add your specific context and you have your specific needs, then yes, you might need to iterate even further. And you can also build your agents from scratch. And in those instances, it's very important that you try them out and see what works and what doesn't and just tell it. You just write to it and say like, yes, don't do that, focus on this, never focus on that.

But I would say this is the first time in my 23 years that these guys are ahead of the sales pitch. in all software companies, usually the other way around, you go out and you talk about the future and all the cool stuff you're releasing and then engineering guys are trying to catch up. But as I said in the -- if you saw the video on LinkedIn when we invited to this event, it's -- I think it was also Lenin who wrote it sometimes that founders typically have like a very annoying sense of detail. And I'm always the guy who finds the 10 first bugs in new releases. And I was literally blown away about the first demo, which is many weeks ago now. So you will be impressed, I promise.

Yes. I'd like to add on that as well. I mean, I work as an engineer and I have for my whole career, and I don't like to say things that are not true or promise things that can't. But I'm very confident in this future and what we will deliver and build in the coming months. So yes, I'm very excited. Take that to the bank.

Just so I understand. So how is this going to hang together with the normal up sales that we use every day? Is this something separate? Or is it integrated into the normal one we use every day?

Yes, it will be integrated, yes. So currently, it looks just like I showed you where you can't really see anything of the regular Upsales. But I think what we'll do is we'll show the top bar or some things you can navigate around to other parts of the system as well. But yes, it's part of the regular system.

And I think we had an interesting discussion when we were talking about this when we started looking at the design for this. I mean a lot of SaaS companies, software companies when building AI, they added somewhere like it's an assistant somewhere in the corner, but the product stays the same pretty much. And we had this kind of metaphor of like if we were a group of 25-year-olds in a dorm room in Silicon Valley somewhere with a goal of like taking all of Upsales customers and making Upsales go bankrupt, what product would we build then? So that has been kind of the idea to challenge everything because we have a long experience, and we had a lot of knowledge, but we have to reinvent a lot of stuff. So the -- what will software look like in general 5 years from now? It will not look like software look like today or a few years ago. So it's very integrated, and it will replace a big part of the product as it exists now.

Yes. I just want to add quickly. I think there might still be a case where we do add the assistant in different parts of the system as well because I think there might be -- I think that might be useful as well. But we will still have the like workspace view that you saw here.

Yes. We have a question from online as well. How do you see these AI agents accelerating ARR growth?

Yes, maybe that's a question for me. So I think when we are talking, especially to our existing clients, I think in almost every meeting, there's like 5, 10, 15 use cases like problems to be solved and ideas to be explored. So I think for existing clients, there is like a lot of work to be done and a lot of more value to be created. And for new clients, I think it's about staying competitive, making sure that we have the best product and the best product in 2026 is to have the best AI. So that's the bet. And of course, we are not done. We will never be done. We will continue doing this as long as the company exists, it's always a matter of innovation.

I have more of a trend watch question. What will great salespeople look like in 5 years or in 3 years?

I had an interesting conversation at Lumestre actually about this. And I think great salesmanship like 20 years ago when we started Upsales, I think it's the exact same thing today. I think it's about knowing your product, understanding your customer, being able to build relationships. And I mean, I think the kind of negative view that people typically have of a sales guy, I think that kind of sales guy was never a great salesperson. So I think it's kind of universal and timeless. But they hopefully will not spend their time adding phone numbers and e-mail addresses into a CRM system.

Yes. I want to add to that as well. I think like the agents will -- they will make you look good, but then you have to show up in person and talk to people. And if you're not a good salesperson, it's not going to work either way. So...

I think AI amplifies it amplifies your strength, but also it kind of very painfully displays your weaknesses.

First of all, I just want to say congrats for this big launch and this new feature. It looks -- I mean, it seems amazing, really. And we're happy to be an upses customer. One thing, I don't want to be the most boring guy in the room, but about the GDPR and the data. Could you tell us about a little bit more about it? How do you share the data? And how does it work?

Don't reply. No. But yes, that's a real concern. And we've thought about that, obviously. What we're doing currently with these kinds of products that we have launched today already if we have to sign like separate contracts where you agree that this data will be subprocessed by third-party providers. I'm looking into finding providers in Sweden, which keeps the data here and that are GDPR compliant and ISO 27 that have all certifications. So we don't have to worry about compliance. But to be honest, there's one big problem. And the problem is that America has the best agent -- the best models, but the open source ones are catching up. And I think like we both talked about like the models, they're not really -- the models aren't a problem anymore. So I think we can get by with having not the top-tier model, but an open source one that is compliant, GDPR-wise and otherwise. So I think what might happen is like you might have to wait until you get the compliant version, and then you might get to choose if you want to be completely compliant and -- or if you want to go crazy. But an important point is that we never train on any customer data. So all customers, they -- your data is your data, and we don't do anything with that except store it for you and do what you tell us to do with it. I think it's also interesting. I mean, the kind of -- a lot of companies, we have the trend with a lot of companies moving away from U.S.-based software. I think it's not because of GDPR. I think it's more because of of Trump and his adventures. And we see some very promising European companies like Mistral in France, which are building models, which is catching up to both Anthropic and OpenAI. So yes, time will tell.

We have another question from online. What kind of behavioral changes do you think the sales teams need to do if you want to successfully work with AI agents?

I think they need to do a little bit more, unfortunately. I mean, what I see, again, talking about the best sales reps I've worked with and the best sales reps I've seen, I think they have a kind of systematic way of thinking about how they do their work, how they do sales. And I think that's the kind of behavior you need to be able to get more leverage from using like any kind of tool, but AI in particular. So I think the boring answer is to like you need to be a little bit more interested in processes and tech in general. We have an example at Upsales now, a guy in our sales team who's 22 years old, has like 0 technical background, and he's like building tools using Claude connecting to Upsales because he he had the opinions about something lacking in the product and he has 0 technical background, but he has a lot of drive and the curiosity. So I think like expanding your views and be more curious is an important trait.

What is your predicted cost for using AI models like cost like for tokens, et cetera, for the next 12 months?

So there's no specific number that we communicated. But I mean, I think this is a challenge that every software company has now because the AI agents, they can provide a lot more value than traditional software. And I mean, the cost is variable. That's just how it is. And as a customer, you want predictability in what will I actually pay for. So we don't -- we are not -- we don't have a perfect answer yet, but we are looking into like how to find a fair pricing structure that allows the customer to keep some kind of predictability.

Yes. I mean, we will add like limits per organization and agent and user, so you can be sure that, okay, the maximum cost is never going to exceed this because obviously, it could be possible that you build an agent that does a lot of work, but it's not really valuable, and you want to avoid that, obviously. So I think some limits will do the trick.

We have one more in the back.

I'm Lars from Pro Hearings. Very interesting. fantastic. You did not address the outreach. I come into my office, I'm a great sales guy. I get the 10 prospects. But what about the next step, the outreach? And how can the models predict this guy you should talk to on LinkedIn, this guy you should e-mail, you have to meet this guy could suggest a Teams meeting. expand a bit on that.

Yes. So it was part of the demo, but it was really small and I probably just skipped over it. But basically, how to decide what to write and when is basically just like how do you know that? You know it because you should always reach out -- I don't know sales. You know sales. But like you have to teach it, and we have to teach it. And you correct it basically. It will provide a draft for an e-mail to say, and you look at it and you think, okay, this looks pretty good, but I would never write it like that. I would write it like this and then you correct it. And next time, it knows that, it provides a better suggestion. And then you might correct it just a little bit like, okay, never do that, just do this, and then it gets better again. So it learns how you like to communicate. And that's something that we can't ship that for you. I mean we can't make it talk like you. You have to make it talk like you. But I would guess that's basically how it works. You can always like provide a lot of context if you have sales material already, you just dump it in there to get a good starting point.

I think the view you should have is like see it as an employee, like how would I onboard a new rep on my team. I think it's pretty much the exact same thing. And the good thing is you will never have to tell it the same thing twice.

That's a good analogy.

Yes. And we have one more question from the online audience. Have this been tested together with customers? And if yes, what feedback did you get good and bad?

So it's like a select few customers together with our internal sales team, and we're kind of rolling it out gradually because we don't want to -- we want to make sure that we don't release 1,000 agents that go crazy and reaywalk. But I think the customers that have tried it are kind of our most enthusiastic, most technical customers, and the feedback has been very good. And again, I think we're really on to something because typically, we need to iterate many times before getting to where we are now. And yes, I'm quite impressed by how fast we got to where we are now. Yes. And then the release plan is to just continue throughout Q2 and gradually release it to more customers.

We also have a question here. You mentioned moving faster than your big American competitors. Have you compared their AI agents to yours? And how is yours different?

You want me?

Yes, I don't remember saying that.

Yes. So I mean, we've tried all of them, I would say. We're not alone in trying to do this. I think the 2 things that -- the 2 advantages we have is, number one, we have the data. We have the data and we have the 20-year advantage of working with real kind of data. And that allows the agent to be more relevant every time. So it gets you from iteration 1 to 10 without having to do the iterations, I would say. And the other thing is that -- I mean, there are many kind of in our category of software pie streams, I mean, the sales guy selling a CRM will say that it's going to work perfectly and you're just going to grow your business if you just buy your CRM. All of us know that's not really true. So we -- knowing what works and what doesn't work and all the experience and putting that into the agents, I think that's the main difference. I think it's time to wrap up. So thank you, everyone, for coming, and thank you, everyone, for joining us online, and see you on the rooftop for a few drinks.

Nice. Thank you.