Learning from Deploying AI in B2B SaaS for PMs | Dominik Rose | ProductTank Cologne

exactly cool big thanks to moneta and as jona also presented some nice ai highlights of river and actually when we thought about like what do we want to do we also approached bjorn who a big spoiler will scare us a bit maybe maybe also not maybe gives us some like good ideas how how product manager role evolves in the future what change might come up on us and we thought like we want to set a bit of theme to say ai is here ai will change what we are doing obviously but but let's put it from a positive perspective let's show some of the struggles let's be candid about what are we doing but really also share some experiences to say yes it's working it's working for companies so linux in the in the b2b space but then obviously it's also working for individuals all of us can embrace this technology and use it to make it as an improvement for his career, and going a bit from this direction. And really what I wanted to open with is, and you see it here, and yes, I talked about model context protocol, would not be to say, ah, have you used it right now or not or something, but really like spark some enthusiasm, spark some ideas. yes approvals is always there so i cannot just do everything but what would you do if you could and i'm obviously really really excited to now look into this and let's see what we find out what's mcp good let's let's see mcp who mcp used too much context i was searching for ideas get information out of confluence writing product requirements connecting figma to 11 labs good somewhere two participants are typing. Three participants are typing. Use skills instead. Create much better release notes. Let them test our product, cool. Let's wait for the two others as well. Create automated tests. Put it on the slide and see if more leads want to buy our software. I love it, very, very good. No, seriously, I mean, in the end, As we know, AI is evolving fast, so it's the job of a product manager. And one core message, and it's not to shy someone away, is I believe in 2026, product managers need to learn about those topics. So model context protocol, we'll take a look into the timeline in a second. One of the protocols that are very, very old right now, it was released in the end of 2024, so like decades away. can argue, and some of you did, whether it's now the right thing to do, whether it's efficient or not, whether it's in the end some nice move from some vendors, et cetera, et cetera. But the argument is let's use time with it and let's understand what this is about. And this is a bit what our journey in the last, let's say, two years with SAP, but also before was as well. we have gone through waves as well in ai adoption i show it a bit we now have some excitement both in our customer base but also in our employee base but it was a rocky road as well on the top saying yes we need to play with ai use cases going to some frustration some hallucination to say it's not ready yeah what we are doing first of all is creating b2b software yeah so it's a different space when you go b2c i'm happy to discuss but but this is coming from a from a b2b site what we are doing rather successful is helping large companies to clean up their their stuff in it guess what it's more important than ever right now if large companies i don't know whether the river guys can can relate or others as well want to think about something like agents in production. They need to think about like, okay, what systems do I have? Where do I connect them with? Who's responsible for this? What kind of technologies do I have? What kind of technology do I want to have, et cetera, et cetera. So all of those are things that we are trying to resolve at Lina X and again, B2B, we did it. And if you reflect about it, we did it in the last decade, 10 years ago. Not a lot of people thought about cloud native and cloud, et cetera. This was still the time when you needed to go to companies and said, ah, we are on premise. No, we are using something like Docker and Kubernetes, et cetera. And we are hosting it in the cloud. Now we are almost feeling a bit like the dinosaurs ourselves. We are facing right now startups, which we might know or which might not know, say who start with AI, who still say like, yes, we can build entire stacks on AI. yeah and it's our challenge our business challenge at hand right now to say what are we doing about it because and this is what me observing the the b2b space and again b2c might be a bit different but especially b2b what we believe clearly is we will see different players in there and coming out of some some intense sprints from the SAP side, also the big players like SAP, they will see some disruption by AI native players. So it's our job as a leading vendor to say how do we stay in front of the curve? How do we maybe transform what we are doing? And so I talked about the Stone Age, about MCP, I have it here as well. This technology is easy. evolving very, very fast. We all know this. There was a world before chat GPT. Yes, there was a lot of AI as well. But then obviously since like 2022, a lot of things happened like chat and everything, function callings, and there were agents, and there were reasoning, model context protocol, A to A, and there will be more, et cetera. So what we try to do and do this. sometimes better sometimes worse is to say let's keep up with it so yes similar to the word before gpt we had some functionality ai functionality before as well then experimented put it on a slide whoever wrote it in the slido clearly some tactics as well to say yes let's put ai on the slide and try to sell it yeah and step by step come to more more sophisticated use cases And I brought one, there are others we can talk about, but maybe not today. I brought one which I feel is very helpful to look into more depth. Because first of all, implementing this has not that much to do with AI itself as well. So we are not hosting any LLM or something for this. But if we think about what do customers want to do with a with an enterprise product like linux we always had the user interface where people can consume it but then also we always had like apis where people want to exchange and taking a decision actually last year to say we host an mcp server a model context protocol server ourselves unlocked all of a sudden a lot of use cases from the customer side So it unlocked and MCP as a protocol, you can think about it like a USB plug for agents. You plug USB into the agent and all of the sudden the agent has access to data and guess what, the agents get smarter with it. And all of a sudden it unlocks cases which were normally possible in the UI to say, Give me a life cycle or give me something of my application and I want to change something and you're not doing it via the UI, you're doing it via an agent. So this sets us in a situation right now that we can unlock more and more use cases and to show this for a second, this sets customers all of a sudden. Oops, the video got lost here in the Google location and I do it without a video. What you... There's a nice flickering in here. So what I wanted to show and what you can imagine for a second is now a screen with Cloud Code where you not interact with the UI typically but where you interact with your agent to say I want to create a certain decision and then the agent is going and say I scanned for the tools right now. So I'm not scanning for any UI function, I'm scanning for the tool. I'm doing the job, all of Cloud Code. and it still lands in the enterprise UI. It's a nice little story of we are not doing any hardcore AI features ourselves but we are unlocking our customers to do this and then getting rapid feedback and having our customers share this. It's interesting and I'm really observing this. I linked you an article from the ImmoScout folks as well. who did exactly the same pattern recently so they said we unlock use cases so that other like developers other customers can build on top of our data can build on top of what we are doing it comes with a lot of questions obviously it comes with security and data privacy questions it also comes with questions and we can talk about it also along on pricing model so if you have a pricing which is dependent now on on seat licenses it might hurt you we are doing the same you know we have a vendor in our our company where we are paying money for seats so we are paying them 150k we have now exposed a lot of content from them y and mcp servers so in theory we can reduce the bill right now to one user instead of 150 yeah so it has implication also on on pricing model but the one thought I wanna leave you with from our side really is those little things like publishing some APIs, making data available, having customers to invent those AI use cases with you, they can get you started. And then obviously you wanna go on and say, build agents yourself and do more and more, but sometimes it's really about these small steps. As I said, it was mainly like setting a bit stage for Bjorn as well, but I wanna give six learnings from our side, which I learned from working with a lot of product managers in our teams, but also in others in the last months. So vibe coding, and we haven't talked about Lovable, and we haven't talked about prototyping and Figma, et cetera, that's one thing. If you really wanna think about transforming your role as someone in product, it's not only on like, ah, I do wipe coding and I do faster prototypes right now. What's interesting, and no blaming, no nothing, but it's really also chance to mingle and to exchange here. I fully believe that getting an understanding of things like model context protocol or AA, A2A, or some of the technology is crucial. for every product manager who wants to keep up with AI in 2026, that's one. Number two is also still some encouragement here. What we learned with this example, with other examples as well, is that building great product stays fundamentally a team spot. So you need to have the language with your engineering team, you need to have your language with your UX and also looking into a larger product organization. people learn, people fail, people exchange, it stays like it. Third lesson clearly also to share is, and MCP server here might be a good example, customer focus, so learning with customers stays more critical than ever. Again, with a B2B lens, it's clearly you wanna ship something, you wanna have customers learn, you wanna ship something, customers learn, et cetera. Ideally, you climb up the stack it's interesting what i can share working with a lot of large corporates in the last weeks customers are looking for our guidance if we are vendor so those examples like like like imo scout as well we are explaining the mcp technology on conferences and we are guiding customers to say what are the use cases that you can do so at customers people like you are sitting as well and they want to have guidance like what what can they what can they do what we also learn and again also from a from a product perspective is yes to invest into architecture to to build those things right what what happens if this thing scale you want to have those exchanges with customers okay how do you use such an mcp server on scale can you invest into tool selection can you invest into authentication etc etc it stays there so it's it's nothing that comes only for free but again it's also something that is worth it and again you start it with a small step and you iterate what's my take on where we stand in beginning of 2026 and how we drive it forward it's it's crazy how how easy it is to build great products, not only with tools like Lovable or Cloud Code, but also with this open standards, et cetera. It's absolutely crazy and we need to appreciate it because somebody else might build a great product faster than we do. And on the other side, it stays very, very hard work, but it's rewarding work where I feel this knowledge that we have as product people, understanding business model, understanding customers, financials, technology. it it stays super relevant so we are not sure whether you want to shy everyone away right now but just for some of the learnings i wanted to share and take it

AI is no longer just a productivity boost—it’s actively reshaping how B2B SaaS products evolve. In this talk, Dominik Rose shares practical lessons from deploying AI within a leading enterprise product organization, including what changes when discovery, validation, and delivery collapse into continuous, AI-driven feedback loops.
You’ll hear how product teams can adapt as hypotheses emerge from user data, prototypes are generated faster than ever, and decisions are increasingly prepared (and sometimes executed) automatically. Expect actionable takeaways for what you can start doing today—so you’re ready before AI-driven product loops become the default operating model.
About the Speaker:
Dominik Rose is SVP Product at SAP LeanIX, a leading B2B product for Enterprise Architecture Management. He combines scale-up product leadership with hands-on experience driving AI transformation inside one of Europe’s major tech companies.
→ See upcoming ProductTank Cologne events here: https://www.meetup.com/producttank-cologne/