How BMW Group is Driving Business Resilience with Generative AI

Matthias Patzak:
Welcome to the Executive Insight podcast, brought to you by AWS. I'm Matthias Patzak. I'm an enterprise strategist with AWS.

I'm pleased to be joined by Marco Gorgmaier, vice president, enterprise platforms and services data artificial intelligence at BMW Group. Marco, thanks for joining us.

Marco Gorgmaier:
Thank you for having me.

Matthias Patzak:
Welcome to the Executive Insight podcast, brought to you by AWS. My name is Matthias Patzak. I'm an enterprise strategist with AWS and I'm pleased to be joined today by Marco.

Marco Gorgmaier:
Hello, Matthias. Thank you very much for having me.

Matthias Patzak:
Yeah, Marco, welcome to the podcast. Marco Gorgmaier is the vice president, enterprise platforms and services, data artificial intelligence at BMW Group. Marco, would you mind introduce yourself and tell us a bit more about your role at BMW group and what you do?

Marco Gorgmaier:
Yeah, very happy to do that. With our global platform organization, we are a very important organization to roll out and scale AI across the organization, and the platform ecosystem we provide to the teams is basically the backbone for that. So we really try to make sure that we do upskilling, ensure that all the employees know our ecosystem, know the efficiencies they can gain with it and bring it into the organization.

Matthias Patzak:
But it's not just a single platform, so it's not just a single platform for data generative AI. So it's several enterprise platforms.

Marco Gorgmaier:
Yeah, exactly. So it's several platform stacks actually. So one important part of course are ERP platforms, SAP platforms. Then we have our cloud stack where we develop our applications, self-developed applications, which we call standard cloud platform, heavily using managed services there. And then we have our data and AI platform, which are very much growing together.

We started the whole journey with the cloud data hub in 2017. It's when we really made sure to bring all the data together on one platform. So we built ingest for all the system in our systems in our landscape. We formed an organization around that back then called the data transformation office, where we then also implemented new roles into the company, so we had data management and government functions in business, data stewards who have the domain knowledge, the process knowledge from a business perspective to control the semantics for the data. And then of course our engineering organization, so data engineers across our global hubs. We are really spread across the globe, actually, the US, Germany, and our headquarters, of course, then software development hubs in India, in Portugal, in South Africa. So very ,very global organization there. We spinned up our data engineering teams and then helped really speed up the integration of our existing landscape.

Matthias Patzak:
How large is your platform organization approximately?

Marco Gorgmaier:
Globally, it's more than 1,000 people including hubs.

Matthias Patzak:
Your organization, your platform organization?

Marco Gorgmaier:
Yes.

Matthias Patzak:
Wow.

Marco Gorgmaier:
It's a pretty large organization, but it's really the platform for the whole company and our total of engineers in the group.

Matthias Patzak:
Platform is a widely used term in the community. So in the latest Dora research... So platform is a widely used term. In the latest Dora research report on DevOps, 84% of the organization survey said they use a platform from a broader perspective, but the term is not really good defined. From your perspective, what is a platform and what makes a platform successful?

Marco Gorgmaier:
Well, I think for us, and maybe I start with... Because it's really also depending on which platform it is, but starting with our standard cloud platform, there we say it's really a platform where we can ensure the development, deployment, management of course of our applications, and then everything you need around that. You want to have scalability, efficiency. So I think very much the standard definition you would find everywhere.

I think, however, what is really important, and what is also to ask the question, what is not a platform at the BMW group, I think the important thing is that we really include the specifics we have. Every large organization has their specifics, specific policies, specifics in regards to their network set up. So all of that. And that is something that we make sure to implement on our platforms, because that massively speeds up the onboarding process for all of the new teams who use the platforms. That also makes it attractive to use the platforms, because when you have, for example, all your governance requirements, when they're already done and you have a check mark in regard to that, then you're happy to use the platform.

Matthias Patzak:
And how many users do you have for your platforms, so number of engineers or number of teams?

Marco Gorgmaier:
So the number of engineers, it's really more than 10,000 engineers across the platform stack using our different platforms. And when it comes to our data and AI ecosystems, we have around about 40,000 users using that platform, because there obviously you have a lot of business users as well. So we have quite a scale in the company.

Matthias Patzak:
So you became really a large software development and tech organization?

Marco Gorgmaier:
Yeah, you can definitely say that. And I think critical backbone was the approach we had to build up our software development hubs. That was really a massive also in-sourcing effort where we built up engineering teams over the last years, and we keep growing, and we have added two new hubs just recently in Romania and in India last year. So I think we will further grow.

Matthias Patzak:
Cool. And in the context of data and generative AI, what services does your platform provides there?

Marco Gorgmaier:
So I think a very broad set of services. Of course everything around data management, data analytics, the whole governance part for data and AI with the UEI Act for example, or other legislation. I mean, that's of course a very important thing for us. We need to be compliant, and looking at the regulatory requirement for cars even more so, we need to be very, very sure that we fulfill all of governance requirements.

Matthias Patzak:
So you have this government's requirement built in the platform services?

Marco Gorgmaier:
Exactly.

Matthias Patzak:
So that the users of the platform, when they use your service, it's simple, efficient and stress-free, especially from a regulatory and security perspective.

Marco Gorgmaier:
Yeah, exactly. So they're being guided. And for example, for our AI applications, we have an AI framework, governance framework, where they get guided through the risk assessment, and then of course the documentation. And the other parts we have is AI model development, everything around that, the services you need. We have some pretty cool use cases actually in our plants also where we do quality inspections on the cars, so for gap size, scratches, all of that. And then of course GenAI came in and we also have a GenAI self-service platform. That's something we just launched, targeting all our business users also. So we call that Group AI assistant at the BMW Group. And the idea is really that I can build easy self-service applications, GenAI applications for my everyday work.

Matthias Patzak:
Cool. What I see with many organizations that they build platforms, and the platforms, the purpose of the platform is mostly a technical purpose. Mostly it's becoming maybe more efficient or cost effective, but not very often they really support the business. From a data generative AI perspective, could you share a bit what is the actual business strategy of the BMW group in regards of data and generative AI?

Marco Gorgmaier:
Yeah, happy to do that. Yeah, so I think what you mentioned is a very important point there. We always try to make sure... Because, I mean, every platform organization, they love tech, so they love to build platforms and functionality. And that's really something I think it's important to early on align business and IT. That's something we really made sure also from an organizational perspective. As I mentioned earlier, when we started the journey with our data transformation office, we made sure that, for example, for every data asset, that's how we call our data sets that are really then already prepared for data analysis. We made sure that we always have a business owner, so data steward and the engineering side. So that was when we started with the data with the cloud data Hub.

And now we do the same actually for generative AI. So we rather start from the use case and say, "Okay, what is actually the goal I want to achieve from a business perspective?" So I want to ensure quality in production processes and then I see, okay, what is the technology I can use for that? And then what is the data I need for that?

And I think what is also new now with GenAI, and specifically looking at agents, actually we see the next wave coming. So we have the data now ingested in the CDH, but now you need transactional access to all the applications in our landscape. And as you can imagine, we have a massive application landscape spanning from legacy application to state-of-the-art cloud native build applications, off the shelf applications. So you have everything in your stack. And now you need to make sure that you're able to access all of these systems with rights and roles of the specific user so that you can leverage the full potential of agents actually. And therefore I believe it's crucial to have the business and their process and the main knowledge included right from the beginning.

Matthias Patzak:
I was very impressed by the numbers of software developers you have in your platform teams and the number of developers in using teams. Could you share some more facts and figures, especially on data? So I really don't have a clue what type of data, how many data's you create on a single day or per minute. Or what type of data do you have?

Marco Gorgmaier:
Yeah, so as I said, it is really from all the systems, it's from ERP system, SAP systems, it's from self-developed application. And I think in the cloud data hub we have 14,000 S3 buckets. We manage more than 7,000 data sets and we support more than 1,500 use cases. So it's a quite large number we're supporting to date.

Matthias Patzak:
Yeah, sounds very interesting. And how do you know if your platforms are being adopted by the internal users? So is it mandatory to use your platform, or is there any incentive?

Marco Gorgmaier:
Yeah, of course. I mean, that's always the big challenge in a company when you are using platforms. You always have a trade-off, I believe, between standardization, efficiency. So that's what you of course want from a company perspective, and then freedom on the other side that you also need and want in a company, because you want some room for innovation, some room for experimenting.

So I think it's an important challenge to find the right balance here, and it's a constant process. It's never something you have achieved. I think you always have to go the next step. And the other big challenge as a platform team, I think, is you have to make sure that you do not become a bottleneck, specifically when you look at those trade off. And so what we try, of course we have mandatory usage of platforms there that's clearly defined, as I mentioned also from a governance perspective, we do that.

That's one part, but I think the main driver, and it's the same that you experience in a market, you get a winner takes it all momentum. And I think looking at the cloud data hub, that's something we manage very well, because people at some point realize, "Okay, there's already so much, there are so many curated data sets already. I can combine with other data that it absolutely makes sense to connect to this." We provided standard connectors for ingest, good quality for the ingest framework. So all of that added up at some point that it really became very central. And this is now also giving us a head start for our AI platform that we basically did the same thing. We have a good basis now and we can scale in that regard as well.

Matthias Patzak:
An issue with data platforms that I observe a lot is that they store a lot of data. So driven by, you might might've heard the term data is in new oil, and then everyone started to collect all type of data. And how do you make sure that you just store the necessary data?

Marco Gorgmaier:
Yeah, I think it's a big challenge, and we're a automotive company and we are driven by efficiency very much. And that's also why we really try to of course manage cost, and if you, as you said, just store any data, it's very cost-intensive. And even more so when we look now at generative AI and unstructured data. So what we try to implement is a very strict lifecycle management. So data sets that are not being used, you get notifications and that at some point we even delete those data sets. So first we archive them and then they're really being deleted, because otherwise costs would just explode.

And the other part is in our data and AI portal, we always link use cases to the data assets. So you have a very clear lineage downstream to the systems, but also who's using the data sets, in which use cases, and are these use cases as well being managed in an active way. And so inside of our portal, that's really the basis where we have all the statistics where we can manage the data, and if you add a new use case, you can of course then also define that you need new source systems being added with new data for your use case.

Matthias Patzak:
Many organizations find it really challenging to attract talent in the space of data, because it's highly competitive, a lot of specialization is needed. How do you attract talent?

Marco Gorgmaier:
I think of course one thing is the brand. BMW group really has a very strong brand, and that definitely helps. So that's one part. The other part is, and I mentioned it earlier, I think, it was very important that we really started, started to source our talent pool globally to not being dependent on just one market or our headquarter, but really leveraging talent across the globe. And I think that was a very important step for us to get the right talent. And also, I think there's a lot of discussion about data scientists, and of course they're very important, but we've really found out that equally important it was to have the data engineering teams, because if you do not have your source systems connected in a reliant, stable way, then there's nothing for the data scientists to look into.

Matthias Patzak:
There is a survey from New Vantage partners now from Wavestone on big data, and from their perspective and from their service data, it indicates that the biggest issue to the business adoption of data, so business users data is culture, not technology. So what's your opinion on this?

Marco Gorgmaier:
So I believe it's true. Yeah. So it very well matches with my observation. I think it's definitely something that one has come from the leadership. I think it's really crucial that, yeah, I don't like the word data mindset too much or a data-driven company. I think we're still a very much a product-driven company. Nevertheless, it's very important to create a mindset that every decision, everything you do, needs to be backed by data. And I think that's something we've really developed over the last years and that takes time. So you have to start early with your journey.

And the other part is we do a lot of enabling for all of our employees. So I think that's critical. You have to do trainings, you have to take the fear away. I think the same applies now for AI. You just need to give people environments where they can experiment, where they can try out, where they feel safe. And that's something we try to do with the business employees on the one side and the other side of course with our engineering organization. So for example, in all our hubs we have what we call platform academies. And there you get onboarding with all the specifics, and we're working very closely together also with AWS for our cloud stack in that regard.

Matthias Patzak:
Interesting. And this is what I observe a lot, is that this is missing, that the platforms are just building services, but they are not really investing in enabling and training the users of the platform services and especially the business people who should drive actions based on the data. And this is why many organizations fail to really leverage all the investments on data.

Matthias Patzak:
Would you mind share some innovations and use cases BMW group is currently working on?

Marco Gorgmaier:
Yeah, so happy to share some of the use cases, and I think some are very typical JNI use cases. So we have just launched a tender assistant there for... So when we work together with external partners, we typically do tenders. And writing these documents in a very standardized way, we built a little JNI service that guides you through all of that, make sure you have all the right legal paragraphs in there. So that may sound really simple, but it brings a lot of process efficiencies in this process. And the same applies for marketing text generation. So I think typical use cases where we see the power of GenAI at the moment. So a lot of those use cases are in place. And another one we're currently rolling out is, so we have our CIC agents, so the customer interaction center agents working with generative AI to give the right answers.

And the same now is implemented in our website and our MyBMW app on your phone. And in next step we will also bring that to your intelligent personal assistant in the car. So I think that's a great example of how platforms work. You build a service once and then you can reuse it in different contexts, use reuse the technical building blocks to services. And I think that's a pretty cool use case, really increasing the quality when it comes to our-

Matthias Patzak:
Yeah, indeed.

Marco Gorgmaier:
... our services for the customers. And another thing actually we are working together with AWS is we are piloting a use case for continuous pre-training of foundation models where we incorporate BMW model specifics. And this is important because if you want to have very short answer times, rack will not work in this case. And I think that's going to be exciting, and that's then important to do implementations, for example, in the car and in other contexts.

Matthias Patzak:
So it looks like it's a very large organization, it's highly distributed with use cases found totally different fields. So you just mentioned legal tech, market tech, customer service tech. So I really wonder about the resilience of this highly distributed setup. So how do you set up your organization architecture for being resilient?

Marco Gorgmaier:
Yeah, so I think it's actually... So it spans across a lot of processes. Yeah, that's absolutely a right observation. So from production logistics to customer brand sales, really we cover all the internal processes. I think the great thing about that is that you actually can break down silos very much in the processes. So that's a big benefit of not having single organizations just covering their specific process. And the other part is the BMW group, although it's a distributed organization from a global perspective, it's still managed very centrally. So it's easier for us to ensure governance and implementation of standards. And that is definitely helpful, because overall, the organization is not very decentral.

Matthias Patzak:
Interesting. And what trends do you see coming in the automotive industry space, in the data space? Anything you want to share with us?

Marco Gorgmaier:
So, I mean, what I see and what I believe is very important is to enable agents now in an organization, AI agents. I mean, it's not a very automotive-specific trend, but it's really something where we see a huge potential to gain further efficiencies in the organization. And I think a big challenge not many talk about is what I mentioned earlier, you need to prepare your existing application landscape for that. And so of course if you have modern applications that are in the cloud already, that gives you a head start.

But I think the reality in every large organization is that you always have a mixture of legacy and modern applications. And what I really see, and that's where we put a lot of effort in, is enabling those for APIs, APIs that are described in a way that you can access them with a large language model. Do you have roles and rights across the applications so that you can really access them with the individual user rights? And that's really a field where we invest a lot at the moment. And that's then the link to our AI self-service platform, our group AI assistant, to then being able really on an employee business level to develop your own little agents and use cases.

Matthias Patzak:
So for becoming more resilient, you need to decouple the organization and the architectures via API. Would this be their advice?

Marco Gorgmaier:
Definitely. I mean, resilience has, I think, a lot of dimensions obviously, but that's definitely something if you do not manage to get the decoupling in place, I think you will not be able to scale it.

Matthias Patzak:
As a final wrap-up, do you have any advice for your peers in the industry on how to build a resilient data strategy?

Marco Gorgmaier:
Yeah, I believe one thing is invest in data quality and metadata. It's a very easy one or something every one of us has heard, but it's really key. I think data quality, not only in your frameworks, but from a technical perspective, but really from a business process side, make sure that already in the business processes, we get in the right data, because some of the data cannot be corrected from the data engineering perspective.

That's one side. The other side is for an enabling gene AI, you need metadata. And that's something where we are also investing at the moment to really scale that up. Then I would say when it comes to the interplay between data now and AI agents, I mentioned earlier, invest in your landscape for transactional possibilities and opportunities. So that is really something that shouldn't be underestimated and I believe is very important. Then leverage the power of generative AI to break down silos and gain efficiencies. We're doing that in our engineering organization ourselves, so we use it heavily in software development. We use it to automate ingest scripts to leverage the potential you get.

And the last piece of advice probably strike the right balance between built and buy. Similar to our cars where we give really the freedom of choice to our customers that they can say, "I want a combustion engine. I want a full battery electric vehicle. I want to a plug-in hybrid vehicle." Or even now, Hydrogen will also bring into production for our customers. I think the same applies to a software organization. You have to choose when is the best decision to buy, when is the best decision to build your own, and keep the flexibility as an organization-

Matthias Patzak:
And when would you buy and when would you build?

Marco Gorgmaier:
Yeah, it's depending on the cost. That's one side. I think it's depending on how strategic it is differentiating for you on a strategic level. And one observation I see is really licenses in the buy stack. They're increasing a lot, and I believe with AI we will see consolidations, and this sort of leads to price wars. So I believe it's good as an organization to have the capability to build when you need to.

Matthias Patzak:
Thank you very much, Marco. It was really a pleasure having you on the podcast and I learned a lot. Thank you very much.

Marco Gorgmaier:
Thank you, Matthias, for having me. It was a pleasure.

Matthias Patzak:
Thank you.

Listen now

Listen now

Listen now