Building an observable enterprise for AI-driven operations
Company: Bayer
Industry: Pharmaceuticals, Healthcare, Agriculture
Bayer, the global life sciences company behind crop science, pharmaceutical, and consumer health products across more than 80 countries, entered 2025 managing thousands of applications with observability distributed across multiple tools and providers. A major IT reorganization had made its observability team responsible for a critical legacy estate, and the team set out to build a unified platform that could scale across the entire enterprise. Within months of deploying Grafana Cloud in a bring-your-own-cloud configuration, the team had onboarded nearly 4,500 users, gained a single correlated view across the estate, and found other Bayer teams asking to join without being invited.
Challenge
Bayer's IT department split into two lanes: a stable legacy stack that rarely changed, but had to run without fail, and a fast-moving cloud-native layer with high deployment rates. Dirk Baumert's team was handed the legacy estate — thousands of applications — to free cloud-native teams to focus on new development.
The environment added complexity. Bayer was operating under a fully outsourced model, with separate service providers managing infrastructure, applications, and pharma systems, each with its own tooling. Coordinating root cause analysis across this multi-provider ecosystem added time and complexity to incident response. "We recognized that we don't have that much information where we can say that is actually going on in our landscape," Baumert said.
The team knew the solution had to scale. The goal was end-to-end observability — from application front end through backend systems, network, and firewalls — with enough correlation to find root cause fast and free up engineering capacity to focus on innovation.
Solution
After evaluating several vendors in early 2025, Bayer selected Grafana Cloud in a bring-your-own-cloud configuration, connecting cloud consumption data from Azure and Amazon at launch and activating Grafana Assistant on day one. The team held to a single-stack principle throughout: all data, all signals, all correlation in one system — no new silos.
The approach delivered results quickly. Around Christmas 2024, still within the first implementation window, the team was able to surface the root cause of a complex incident that had been open for three days. "We were able to show the providers where their errors have been — major incidents ongoing for three days were not solvable by them, but Grafana showed up root cause problems in that whole stack," said Christian Billig, Agentic Foundation Platform Architect.
From there, Grafana Alloy — built on OpenTelemetry — was rolled out broadly across the estate, and Grafana IRM was integrated so every incident automatically triggered a deep investigation. The team built a contribution model that let application teams onboard themselves, configure their own dashboards, and set their own alerts. Adoption moved faster than anyone expected.
Impact
Bayer moved from fragmented signals across a complex, distributed estate to a self-service platform the rest of the company wanted to join.
Nearly 4,500 users onboarded to the self-service Grafana Cloud environment
Application teams now configure their own dashboards and alerts independently
A complex three-day major incident resolved within the first implementation window
Teams outside the initial scope joined the platform voluntarily, without a mandate
Agentic observability activated, with automated remediation underway across the legacy estate
"It's one of the easier projects in IT which we ever did. It's a self-selling business, to be very honest," says Christian Billig, Agentic Foundation Platform Architect, Bayer.
Looking ahead
Bayer's next focus is integrating Grafana Assistant into its broader agentic ecosystem to enable automated incident resolution across the full legacy estate, while maturing end-to-end observability so every user-reported issue maps to something the platform can detect first.
Speakers
%3Aquality(90)%2F&w=3840&q=75)
Christian Billig
Agentic Foundation Platform Architect, Bayer

Dirk Baumert
Senior Observability Platform Leader, Bayer
Dirk Baumert: Hello everyone, and thanks for having us here today and the opportunity to present our little journey with Grafana and the bring your own cloud environment. So a few words we are supposed to say about us besides all the family topics that we all have, maybe two things important about me personally. So I'm still playing handball in a league in Berlin. I'm based in Berlin where we play regarding the Berlin championship. So I'm still active a little bit in sports. And secondly, I also have a passion for woodworking. The ambition is that all the furniture I have is built by me. But the one thing I learned is also if you're passionate about something, sometimes it hurts a little bit. But no worries that happened by playing handball. Okay. So I'm 25 years with the company for Bayer, had many different roles. And since 2025, beginning of 2025, I'm now in that role for observability.
Christian.
Christian Billig: Yeah, Christian. So first a little bit fine [inaudible] picture. Now this is where I spend my free time with. That's Cologne Carnival and making music. So this is a picture from one of our concerts, which we give every year on Carnival Sunday. So for those in Cologne joining Carnival, you're happy invited. There are rest tickets available also for coming here. No, just joking. Music is something which is a great thing on getting off the IT stuff and getting your hat cleared with another dimension. So for Bayer, I'm now with 15 years. I joined Dirk on the observability architecture initiative 2025. I now moved a bit on towards AI. And yeah, let's give you some insights on what we did.
Dirk Baumert: Okay. So this is the headline of our presentation, but before we talk about the content, I would like to, I have to show you that slide of course for a second. Moving on, talking a little bit about the company, what the company is all about. So we have a theme on our mission and vision is health for all, hunger for none. And I personally can connect to that very easily because it says just that health for all. So we want to treat and cure and prevent diseases of course as a company for as many people as possible around the world. And at the same time, we also want to provide quality nutritions which help people with their healthy and keeping hunger away from more people than possible. So therefore that is something we can connect to about the company itself. There are three divisions for Bayer.
It's crop science, the pharmaceuticals and the consumer health division. And we have representations in over 80 countries. And that is also something which I like a lot because working in such a global environment and in some projects then working in the morning with the colleagues in Asia, then over time with the colleagues in Europe and maybe at a later point at the day, then in the afternoon also with the colleagues in South and North America is actually really a pleasure. And being based in Berlin and going to the headquarter in Berlin, being there and having all the physicians and medical experts being there, it's like a big crowd is always very impressive and nice to see. Okay. You also see the revenue. The company had a quite challenging time over the past years as some of you might know, but it seems like we had just a result of the last quarter which were quite promising.
So hopefully things start to change in a positive way. Okay. But how was our move to bring your own cloud and to the Grafana platform, which is actually the reason we want to talk here today. And when our journey started in the beginning of 2025, we had, as we have in such big companies, a reorganization of IT. And we got task in our department to manage and operate a few thousand applications, which was actually new. And why was that made a dedicated effort for us was so you have more, in a company like us, a lot of traditional applications running on a few more on a legacy IT stack and those need to be maintained. You don't have that many changes in regards to functionality. They just have to run stable more or less. And then you have all the new stuff, the cloud native applications, which have a very high change rate and deployment rate.
And to allow the team to focus on the new stuff with the development, they decided, okay, take away or create something specific for more rather the legacy stuff so that they can provide another environment where that runs stable. So that was the task we got for a bigger team, for quite a big team. And then, okay, how do we do that? How do we run that in a reliable and stable way and how to come up with that? And we though about that and said, first of all, so we have to remediate all the issues that we have quite quickly. That is one of the things which were important because even if that is a legacy IT environment, it is critical and the foundation to the company. It doesn't mean at all that those systems are unimportant in any way. It is very critical. So therefore resolving any kind of issues fast and finding the root cause fast was essential.
But if you do that then for one particular area, it is also important that you can scale that. And this is also in the previous presentations, we saw a lot of the things that work sometimes for an isolated case, very good. But if you want to scale that for a whole enterprise, then it becomes challenging.
And if you can scale that quite quickly and can be also very fast in doing that, that would actually free up resources and engineer capacity who can actually then move on and focus on the new stuff that we actually want to drive rather than operating the more traditional applications. And that is the reason why we came up with the approach, we have to change something significantly. And what we then thought as a starting point is, or what we recognized that we don't have that much information where we can say, okay, that is actually going on in our landscape. So we do have to signal the alerts and everything you're required to make an assessment or to identify an issue quickly or to solve that issue even quickly. So we said observing the environment end-to-end is the first thing we have to work on. That is our first step.
And this is when we actually started the journey with our platform where we looked at providers and partners, how can we actually get there? The next thing is we want to do that for the entire enterprise. So not only for a single application. So we are also talking about backend systems or backup systems. We are talking about the network, we are talking about firewalls. All of that is connected to the actually workloads and can therefore have an impact. So correlating all of that is very important to find the root cause. And then as a last step, then you also want to act on that. But not only as a human being, you want to automate that, you want to have self-feeling scripts, whatever it is and whatever it takes at the end, but you want to really automate that. And this is actually how we wanted to drive that and how it was actually presented to our management.
And then we got actually the endorsement to start with that journey also in a more practical way. And what that means in more detail that Christian will tell you.
Christian Billig: Thank you, Dirk. One anecdote to the negotiations on one of that sessions we had that Gen Z moment on different vendors. So the one was more on the older population and showcasing their solution while then Grafana showed up a bit more in a Gen Z style. That was quite funny as an anecdote back in the days. So we started 2025 with selection processes and implementations during roughly a year ago we got our bring your own cloud instance. And we also at that point came from a fully outsourced IT ecosystem with service partners for infrastructure, service partners for applications, service partners for the pharma applications. So a multitude of ecosystems suddenly into the world and joining Grafana. We immediately starting integrating those aspects while we also had transformation programs running. We now are zero data center driven. We are fully in the cloud besides having still in manufacturing facilities on-prem IT systems.
While we were doing that movement, we switched on Grafana. We integrated the serverless cloud consumption of Azure, of Amazon, and we collecting since then those data. We on the first day already were a customer of the Grafana assistant and we immediately used back in the day the investigation functionality already in major incidents. And at that time we were still in the face of having the service providers doing stuff. We starting to do our stuff. And it was interesting that we at that time already at that early stage of implementation around Christmas last year, were able to show off the providers where their errors have been, where major incidents ongoing for three days were not solvable by them, but Grafana showed up root cause problems in that whole stack. And this is, well, build out over the time now we rolled out Alloy to multitude of systems.
We increase hold that thing, but we are now on that next step to the future on the right hand side, thinking about how we can bring that ecosystem further to life. We are still in the present way challenged with having people onboarded to the platform and solutions like this system, but also Pathfinder were of tremendous value on onboarding nearly four and a half thousand users to the platform and getting them engaged on creating their own dashboards, setting up their own alerts because we are changing in that journey also from observability as a server, some team in some room is doing the job for the whole company to a self-service functionality to all application teams. So this is more now a driven contribution model than anything else. Now the dimension on top which Dirk already mentioned is we want to get that thing driven forward. So we are now lying the ecosystem on that and we activated as soon as agent observability came to the game.
The storyline on if we want to do system reboots in a simple way, if we want to change sizes of machines and we want to get that agentic done, having an incident triggering the right resolutions, we need to monitor not only the systems, we also have to monitor the agents. We need to get the reasoning behind that understood. Well, there are also nowadays jurisdictions which are all of us, at least in you, driving us to be mandatory, that thought process thinking to have it in a documented way. And this is now for us in the IT ecosystem now done with agentic observability. We are more and more now getting that system extended to our business divisions. Crop science and pharmaceuticals are also starting to use agentic solutions. This is to be the fair, the quite early stage for our industry because we are highly regulated.
And as we all know, agents and that hallucination things and the non-deterministic aspects are one of the core challenges in our business. And getting AI doing the deterministic stuff is what we are setting Grafana as one of the key aspects in that whole thing. And what we are now, as the next steps, want to do is to integrate Grafana Assistant into our own agentic ecosystem so that it can talk with every agent we have got and get systems triggered, actions triggered and the aspects of data integration also into Grafana. So was that journey smooth to the place today?
Dirk Baumert: Before we move on, maybe I make one comment. What we don't have on the slides, but what is also very important is we started with our initial scope that I just outlined, a few thousand applications. But as we then rolled out and showed people what actually we can do and how we can monitor and observe all the systems, then also all the others came to us and asked if they can participate in the journey and asked us actually to join. So we had another task on our plate actually to come up with a contribution model so that we can also onboard others which can then already benefit from the data we have collected, which was another huge benefit of that.
Christian Billig:
So from a change management perspective, we have to say it's one of the easier projects in IT, which we ever did. So it's a self-selling business to be very honest. So with all the advantages of the system doing dashboarding and all that stuff, we set the hurdle down to a level that everyone can digest that. We are currently working on the integration with teams so that everyone can, who has got the access words at Grafana into their chat and have got the direct dialogue. We now are utilizing the deep investigation functionality of the IRM triggering automatically. So all incidents get deep investigations triggered. And this is just the first step of the journey. We want to get starting with simple things like disc spacing trees or clearing our block files automatically, all that kind of stuff now triggered with Webpokes or with agentic integrations to find the right solution for the right system because every system for us is a bit different.
And this is, well, it's a tough thing. Security is getting into the game. We have got a multifacet and quite interesting identity and access management concept in our company. So this dimensions where that on behalf of the user is getting really interesting. If you use with your primary identity, Grafana and the system that's using your identity, you may not have got access rights on a server.
Interesting background noise. And this is where that whole and that whole integration also in regards to identities of agents is getting really interesting and where we are recognizing that we are at an early stage of technology and thinking that through in a legacy. I think it's mine, Laura. I think it's in your pocket. Hello? Good. Thinking that through is a quite interesting journey currently, which we hope to get more actionable over the next time. But on that journey, we still have got, as I said, some challenges. And one of the biggest things, and for us on the whole contribution model, we wanted to go with a single stack approach. We know many of our followers or different competitors or even partners are using different approaches with multi-stack. We definitely said we never want to come up again in that siloed situation. We want to get correlation over everything and we go that extra mile on doing RBOC and LBOC and getting the data in one single system and getting the value out of the system and the correlation functionality.
And I finger into the wand of Grafana, not every feature is RBUG enabled and LBUG enabled as you would like to have it. I have to say it.
But we have good dialogues, I would say, with those guys. And I hope those is being catched up in the next sprints on their development roadmap. And we, well, like everybody have got all our challenges on labeling and getting their structure in. But yeah, that's basically the biggest topics I would say. Dirk, you wanted to do...
Dirk Baumert: Yeah.
Christian Billig: Last slide, I think.
Dirk Baumert: Yeah. So what comes next after talking a little bit about the good things, what we have achieved and also the challenges, what comes really next for us? And one thing which is very high on the agenda is actually that we mature in the end-to-end observability for our applications. What does it mean to us is really that we can actually from start to end, so from front end to back end and infrastructure and also related systems that we can actually observe an application and have all the signals and alerts and metrics locks, whatever it takes at hand, because that allows us then also to act on it and correlate on it. And this is very important and a good KPI that we came up with and think about it, that every technical issue that is actually reported by a user should be taken and transformed in something that is observable.
And at the end, you can then measure how many technical issues are still get reported because actually if there is something technical happened with the system that in user experiences, then you can actually also observe it and then you can actually maybe also recognize it before actually even the user is impacted by that. So that is the one important thing that we actually have actually the insights about the data. Then we also want to work on the standards in regards to the telemetry data that we have, that we can do the correlation for everything. I guess we have pointed that out quite a bit and quite often that this is key, but it is because we saw that in our investigations and also in resolving MIs that the impact or how it's recognized by the users is very different than what actually caused the issue, right?
Root cause analysis. And this is something where we do need this correlation going forward. Yeah. And with that, with the correlation, then we can have the AI assistant also really helping us to work on that and to operate the system. And then as a last point, we can then really build the automations, the auto healings, everything required to remediate and resolve the issue. We have heard a similar story earlier today, I think, but I want to say that is more for, again, for the traditional IT DBMs that we have ESXOs and so on and so forth. This is the scope that we have. Everything that we have seen before is more on the cloud native infrastructure and that is a little bit the difference here. Yeah, that has been our journey and also this is what we have in front of us. So thanks for that.
And if there are any questions, then Carl, we would be open to it.
Christian Billig: Sweet. Thank you for the presentation.