How BMW uses analytics for supply chains during semiconductor shortage

Mike: Ok, welcome everybody and thanks for joining us for this. Uh I think it's the last session for today. It's an interesting one. So I'm Mike. Uh this is Dennis. We are both from the BMW Group and uh together with, we are from AWS. Um we teamed up and build an analytic solution based on AWS technology uh to help us to steer through the bm uh through the semiconductor shortage.

So quick. Some, some information about BMW. If you don't know the brand, we do cars, we make cars um 2.3 million last year, especially in the premium sector. We basically deliver cars all over the world. We have production sites all over the world and a very complex supply chain which will be a part of the solution for sure. And in terms of our IT structure, we have different IT hubs everywhere in the world. And our course IT centers based in Munich which uh yeah, which basically enables our tech offices with the new technology also provided by uh among others by AWS.

But to lead you into the topic. Um the last 2.5 years, uh we were facing a very difficult environment. Yeah, we had basically as every other automotive manufacturer manufacturer, we had um some supply shortages with semiconductors. And among this difficult environment, we had obviously to use technology to deal with this kind of complexity and with this kind of difficulties and one solution we created, we would like to present you today.

And this is actually what we do if we have a shortage and cannot avoid a shortage of of the semiconductors. So we couldn't solve it in advance. So we basically have to deal with it and to do it, I would like to introduce you the the the complexity of three entities within BMW. So basically we have the sales department, we have procurement and we have production and all three have to interact together. If we have a supply shortage, it doesn't matter whether it's caused by a semiconductor shortage or any other. But the point is normally a shortage happens within our supplier network, which means basically saying they cannot deliver the parts. We we need to build a car and to get the uh to get the extent of it like one car normally on average consists of 100 and 1000 parts. So if we have now assured it, then obviously this can be can can be effective our whole production lines.

On the other hand, we have our customers like you. Yeah, and we give you as BMW the power of choice. So you have actually the possibility to customize your own cars. You can actually decide whether you want to have a panoramic roof in your car, whether you want to have it in green red or whether you want to have, for example, advanced headlights that follow the road with you. It's all your options, all your choices, which makes it, which gives you the opportunity and we don't want to take this away from you, but it obviously makes the world a little bit more complex for us in the supply chain and in the production.

So those three entities, we have to bring together when we have when we have a shortage. So we have the supply chain where we are actually looking on parts that go into cars and those cars are produced for you as customers. So obviously, if the supply doesn't meet the demand, we have to do some allocations. Yeah. And this is where the actually the analytic solution takes into account.

So first of all, then the solution brings for all three entities, transparency and very quickly. So basically when we have a shortage at the parts, we already know, ok, this part goes in this car and in this option, you choose as a customer. Once we have this transparency done, we can now decide what are we doing with that? Is it, is it you as a customer from the states we want to deliver first or do we do we or do we want to deliver a certain car in an asian region. First, there are like decisions that our, our sales department have to do in those situations.

And since we have such a variety of combinations, it's very hard for them to do it only by their own knowledge. So analytics will help us to do this. And this is the second part, the optimization. So if you ask our financial department, they always, they will give you the straight up variable where we have to optimize contribution margin basically to push our our i. But it's not that easy obviously. Yeah, because you, we care, we care for you as our customers.

So we have our sales department takes different variables into account. We have certain markets we want to deliver too. We have certain cars we like to prefer to push into the market than others. For example, we obviously we want to strive the development of our electronic vehicles. And we like obviously to us customers as evangelists who like electronic vehicles to provide you with those cars. Yeah, one is actually at our booth, the i seven. So you probably have to check it out afterwards.

And so basically all those parameters that the sales departments can take into account have, have have an impact on our optimization and to show you this, i would like to also explain you this in a little bit of detail. I would love to hand over to jr who gives you a little bit deeper understanding.

Jr: Thank you, Mike. So as Mike mentioned, we had two challenges as part of this project. The first one being the lack of transparency about which special options were impacted by the chamis short shortage. And what is the optimal allocation related to this special option?

So what we did, we developed a tool based on analytics to cope with these two challenges. So the tool is made about basically three steps. So the first step is to get the data in. Second step is to create this transparency tool. And the third step is to optimize the different special option based on the allocation.

So on the first part, so the data sources. So for those who have been familiar with such a project, getting data in is always the big challenge. So there will be data quality issue, data access issue. So you probably are very familiar with such a kind of challenge.

So fortunately BMW, they already have a data leak in place called cloudhub cdh and many of these data sets were already in cdh. So BMW was able to take, for example, market demand financial data directly out of the data platform and then to combine them into the transparency tool.

So the second step was to create this transparency tool. And for that, we we basically combine the different data source. Uh and then we were able to for each special option, the demand and shortage per model and per market and these conspiracy tools are already the first value out of this project.

And then finally, the third step is our optimization. So like Mike mentioned based on filtering, we were able to optimize uh based on the the transparency cube that we had before. Uh so now we'll let Denise explain a bit more in detail uh about the the technical aspect of our solution.

Denise: Thanks Julian. So let's have a deeper look at the IT architecture. First of all, this is just a simplified version of our real IT architecture. But I think the idea has become very clear.

So what we did, we built an IT architecture that is mainly serverless and relying on AWS Lambda services out of two main reasons to be highly scalable and cost efficient.

Now, let's have a look at the left hand side here. You can see our data source, the CD, the Cloud Data Hub, the BMW internal data lake. We use Glue jobs to transfer the necessary data assets into our S3 bucket, which is the main base for our two modules, the transparency and the optimizer tool.

These tools are following exactly this what Julian just presented to create transparency and to provide the best allocation of special options over markets.

Now imagine you are a user on the right hand side here and you log in in our static web page. Then the static web page is also hosted in S3 bucket and then you can interactively trigger our transparency tool or the optimization tool. And there the magic happens when you see our both modules, they are built very similar.

So both have an orchestration of two Lambda services and a DynamoDB. This is mainly out of one reason to overcome the 30 seconds time of limitation which an API Gateway provides. If you have more questions on this, please feel free to ask us afterwards. Since I won't go too much into detail here.

For now, then either the transparency tool or the optimizer terminates and the results will then be go over the AWS QuickSight back to the end user. The QuickSight dashboards are integrated on our static web page. So the end user can interactively check the results of the tools.

But I think that was now enough from the theoretically part. Now let's deep dive in the demo. I hope you're curious to see how our tool looks like.

So let's focus on the left hand side and imagine I'm a purchaser at the BMW Group and my boss now asked me to check for this part number here. If we have a shortage situation, a part number can be, for example, a specific part which has a lot of semiconductor included. And since I'm the purchaser, I asked the supplier, ok, what is the supply for this part number? And the supplier told me in week 49 and in week 50 we have a supply from 80 units. And then I would trigger the request request here.

And on the right hand side, the AWS QuickSight dashboard will appear I already preloaded this, let me make this bigger for you. And now the first thing is I want to know in which specific special option, this part number is included. So here I can see the adaptive LED headlight which Mike already introduced. This is a headlight which is following your road is demanding the specific part number.

Please keep in mind this is just a demo here. So the data are not real data, they are dummy datas. And in reality, the relationship can be more complex. So one part number can be included in many specific special options. And of course, I can also insert a lot of special as part numbers.

Now I'm interested in the demand of the specific part number. So I have an overview here here, I can also see my supply and then what shortage maybe derived from that. And here we will see, for example, I have a shortage situation in week 49 and week 50. And I can also have a deeper look here on the second screen about the specific special option. Ok. Let give a second. Yes, it loaded. Perfect.

And here I can see ok, what is the demand for the part number with a shortage next to it? And I can also have a more detailed look. For example, here on the heat map. What is the demand for the specific special option over the different markets? And I can also have a weekly zoom in to check which vehicle models have, which demand for the adaptive LED headlight.

Now, I as a purchaser see we have a shortage situation and I will come together with my colleagues from the sales department and from the manufacturing for the production and we discuss the situation and if we don't find a solution to overcome the shortage situation, then I would hand over to my sales colleague and the ss colleague would now focus on the optimizer.

And let's assume I'm now the sales colleague and I will check. Ok. In week 50 we have a shortage from around 9% for the for the adaptive LED headlight. So I would just focus on this specific greek and what hit this button here again? I again preloaded this and now I can set specific optimization parameters.

First of all, the optimization corridor, this is a corridor which says what is the maximum allowed reduction which the optimizer is allowed to take out for each vehicle market combination. So for example, here by 90% i'm only allowed or the optimizer is only allowed to take out 90% for a specific vehicle model in a market. For example, in Germany, we introduced this corridor as a fairness bias.

So you can assume otherwise the optimizer would always take out the special options from the vehicles with the lowest contribution margin or from the markets with the lowest contribution margin.

Then we have another look at the so called 100% markets. So you can assume in some markets, the adaptive LED headlight is not a special option. It's mandatory, it's included in the default package and therefore the optimizer should not take it out. So for example, here in Japan, for this model, it's not allowed to take out the adaptive LED since it's included in this model default.

And then last but not least, I can also add the strategic constraints which Mike mentioned earlier. So for example, I can say now, ok, for example, Great Britain or Japan is a high priority market. And I want to protect this market from any reduction from the optimizer or I could also say, ok, let's protect all the electric vehicles from any reductions.

And now I'm able to choose the target function. So either to optimize regarding the special option contribution margin or the vehicle contribution margin, I choose here. Now the vehicle contribution margin and we hit run optimization.

And what now happens in the background? The Lambda, the optimization Lambda service will be triggered, it will perform a linear optimization by taking all the constraints into consideration and maximizing the contribution margin. And now the Lambda terminates and the QuickSight dashboard will be loaded.

And now I'm able to investigate the results of the optimizer. We can see now the specific vehicle models here and also the different markets. And for example, I see now the demand for Great Britain over the models, the allocation which the optimizer proposes to me. And also the reduction which the optimizer proposes to me.

And then here at the end, I would see, ok, this reduction would now fulfill our supplier. Of course, this is these are just dummy data as I already mentioned before. And also the optimizer is just giving me a proposal. I would then again discuss with my colleagues from the sales from the purchasing and also from the production if this makes sense or if I should reiterate on it.

And we also have here a small heat map included as well. So our optimizer has a lot of other cool different features. So for example, running different scenarios comparing them, but with a look at the time, I would say let's jump back to the slides, let's jump back to the slide so i can give you an update.

Perfect. So I want to say, first of all, thank you to our colleagues from AWS and BC G who helped us to develop this tool super fast in only three months. And I would also say thank you to you as you joined our session today and you were listeners and i hope when you go to the next time to your car dealer or on the BMW web page, you see that we are always aiming to achieve or we are always aiming for our most customer satisfaction. But this is sometimes very complex as you see it. But analytic solutions and AWS analytic solutions help us to overcome this complexity.

And with that being said, thank you so much. It was a pleasure to speak to you all today. If you have questions, feel free. We are around here. If you're interested in BMW, you can see our iC seven series at the AWS automotive booth. Check it out and yeah, have a nice evening at the re:Invent event and enjoy your week here. Thanks so much. Thank you.