AI WATCH EPISODE #13

Manufacturing AI: Insights from Cobham Satcom’s CIO

By Bjørn Olsen, Niels Bomholt

Listen on Spotify

Watch on YouTube

Welcome to our thirteenth episode of AI Watch!

This episode of AI Watch features Bjørn Olesen, Sales Director at 2021.AI, in conversation with Niels Bomholt, CIO of Cobham Satcom, on the implementation of AI in Cobham Satcom’s supply chain. The two engage in a discussion about topics such as:

  • How integrating AI into supply chains can improve transparency and provide superior demand predictions.
  • The key phases of an AI implementation project, from identifying the need to presenting the solution to senior management.
  • The challenges of AI implementation, including change management and bridging the gap between data-science and subject-matter experts.
  • How AI solutions can support sales and demand forecasting, component monitoring, and risk assessment.

Overall, this episode touches on the key lessons learned, including starting in areas with good data quality and building a roadmap for AI implementation.

Introduction and role as CIO

Bjørn: Hello and welcome to this episode of the AI Watch. Today we have Niels Bomholt from Cobham SATCOM with us. My name is Bjørn and I’m from 2021.AI. Today, we’ll be talking about how it actually is to implement AI and how Niels, as a CIO, has been experiencing this.

Welcome, Niels.

Niels: Thank you.

Bjørn: First of all, maybe you could tell us a bit about your role as CIO in Cobham Satcom to start things off.

Niels: Sure. My role at Cobham Satcom, besides the daily operation and normal business driven by IT projects, is to help the company drive efficiency using new technologies wherever appropriate. And that’s basically the reason why we’re here today.

Interest in AI and supply chain transparency

Bjørn: Great. I know that you’ve been doing an AI project, but what sparked the interest of starting out with AI in Cobham Satcom?

Niels: What really started this out was an idea that’s been brewing between myself and my colleague for some time about how to get better transparency and prediction into our supply chain of components. All our products have a huge amount of electronic components in them. And as everybody knows, electronic components are a scarce resource.

So how do we get more transparency into that to make better decisions? How do we ensure that we know when an electronic component is about to either become obsolete or become a resource, a limited resource, and thus either driving a need anytime or a price increase?

The sooner we know that, the faster we connect.

Bjørn: So it’s been more precise insights into what’s going to happen.

Niels: It’s more precise insights to what’s going to happen, because at the end of the day, we serve our customers. Our customers are basically partners, their demand.

How do we serve that best? And making sure that we have the products available when they have a need. It’s key to enabling their business.

Bjørn: And for the ones out there who don’t know what Cobham SATCOM does, you mentioned there were a lot of components in there.

Niels: It’s satellite devices, satellite components, satellite products that manage the communication to and from satellites. So antennas, modems on ships, cars, whatever.

Key phases and milestones of the AI project

Bjørn: You’ve just been through a project, what was the journey, the key phases, milestones in it?

Niels: Well, the key phase was originally this idea: What can we do to improve transparency, make better decisions, reduce lead time, and also reduce cost?

We had a conversation with our head of procurement. They added that you also need to look into the forecast, the demand forecast, of course, for the products. But then you also need to look into the sales forecast. So how do we improve predictability in the sales forecast? Because that leads into our demand forecast. Couple that with, how do we have greater visibility into all the components? And we have several hundred products. Each of them will build up material, on average 700 components. Nobody can look across them and have an overview, no human beings. Then building proof of concept on that.

Bjørn: So did you use the proof of concept to convince the rest of the business?

Niels: Yes, exactly. So we built a proof of concept, and that allows us to showcase that it was actually working and also to build the business case to get some credibility into the business case. Up until then, it was more assumptions.

Based on that, two things: choosing the partner, which would be you in this, and then presenting it to the rest of the senior management, both to give a sneak peek in terms of what the solution was like, but also to address some of those questions. Concerns regarding governance, regarding security, regarding costs.

Working with an AI provider and platform

Bjørn: This is a bit of a biased question, but when we decided, or you decided that, let’s pitch the pilot here and get started and show some results of the business, we came in with an AI platform and you could argue that it was maybe a bit overkill for a pilot, but how has the experience been working with a provider like us, where you get a, let’s call it, precooked platform? What has the experience been after the pilot and the ongoing development?

Niels: We probably underestimated the need to have a platform in the beginning. What we are doing and the type of business and type of customers we’re dealing with, everything, compliance-wise, just gets more and more important. So having a platform established from the get-go, which allows us to have the governance, both governance building from the get-go, but also allows us to improve on that and to expand on that governance in various applications, that is going to be a huge benefit for us.

I expect that will be the same regardless of industry, that you will meet some requirements when it comes to proving what it is your AI models are doing, that you can prove that it’s well documented and all that, and getting that in with a platform, that makes it so much easier. And we’ve seen that benefit from those two use cases that we have actually developed there afterwards.

Challenges and benefits of AI implementation

Bjørn: So the business case gave good results that were presented for the business. I’m glad that you brought us in, in presenting this, but I also sense that, as with many other companies I meet out there, there’s no track record of buying AI systems. So people would like to kind of see what this is about. There’s a lot of interest in the beginning. Cases are all over the place, new ideas and cases.

Niels: New ideas and wild cases are being floated around with huge benefits and all that, and it can probably be a little bit difficult to boil that down to a context that fits your own organization.

In our case, we did the proof of concept and showed that by just using the technology, we could get to the same results when it came to predict the risk on components faster than any of our senior procurement, and validated against that person that the results were right.

Then, we are still struggling a little bit on the sales forecast part…

Bjørn: But maybe that is a great overlap to the next question I actually had, because what have been the challenges? Is it a technical challenge, is the model not performing, or is it a change management, or why are you bringing up this?

Niels: Change management. Absolutely. That’s first. Another thing is data quality. Our data quality is our very high level when it comes to knowing what goes on in these components. Also very high data quality. Then the last thing I want to touch upon is that you still need to bridge that gap between data science, regardless of how well they are, in understanding your business and your subject matter experts, because it is still two different worlds.

Bjørn: Yeah, it’s not the same logic.

Niels: No, exactly.

Bjørn: So communication, change management has actually been the biggest challenges in this.

Niels: Absolutely.

Components and areas included in the AI solution

Bjørn: Niels, maybe for the ones who haven’t been part of the project, like you and I, what components and areas are included in the AI solution they decided to go with?

Niels: So first and foremost, it’s a sales forecast, demand forecast, looking at components availability. So sales-forecast-wise, what we did was train the model on our entire CRM data set for the past eight years. And what we did there was simply to train it to ensure that we knew cases that we have won or lost. What’s the predictability in that? What’s the score? So, I got some accuracy into that. What we realized was that we could fairly easily get to an accuracy of somewhere in the high 70s in predictability of opportunities.

Bjørn: So out of cases in your CRM system, you could tell which one would be one with that kind of background.

Niels: Exactly. Of course, that’s interesting, because that leads into the demand forecast, because then you have a much better input into the demand forecast. You look at all the components, you look at the component library, you look at the component availability, put all that into the model, and that allows us to get a view on a bill of material level. Each of the components, how are they progressing in terms of availability, price, how many years, months left until that component is becoming end of life and stuff like that, do we have on each component, do we have alternatives available? If not, then that component is a higher risk.

Bjørn: As you mentioned, 700 products with 700 components in average to monitor manually or with AI instead.

Niels: Exactly. That’s a no-brainer, right?

Key lessons learned for CIOs starting AI journey

Bjørn: Final question, I think here. But looking back, a key lesson was learned. What would you tell another CIO that maybe hasn’t started their AI journey yet?

Let’s start with the easy part. What was easy?

Niels: Well, the easy part in our case was that we had the CRM data for over eight years in the same system and the same to some extent for our components. Again, our bill of materials, very structured, start out in an area where data is of a certain quality. Otherwise there is the risk of spending a lot of time chasing that data quality. But again, do not let the data dictate. But because at the end of the day, you need to solve a business problem. If you don’t have a problem in an area where you have fine data, congratulations.

Bjørn: I remember looking at some of your CRM data with the data scientists when we were presenting, and lack of data was actually an indicator of an opportunity not being won. So bad data could also be a signal that something bad is about to happen.

Niels: Absolutely.

Bjørn: Something will not happen.

Niels: That’s number one. Number two is that you don’t have to save the entire world in the first goal. So instead of building eighteen runs on that, let’s do something that is much more compressed and where we can get some tangible benefits. Right away. After that, build a roadmap, because the ideas come and try to formalize that roadmap.

Bjørn: It’s not going to be one model that fixes that entire company. For me, it’s kind of like when we were all talking about RPA or the robot process automation, that it’s not one automation, it’s a lot of small improvements.

Thank you, Niels, and thanks for taking the time. And thank you for watching. Remember to subscribe if you want to see the next episode of AI Watch.

Have a great day.

Bjørn Olesen

Bjørn Olesen

Sales Director, 2021.AI

Bjørn has 15+ years of experience in IT implementations, for a large segment of clients. At 2021.AI, Bjørn focuses on transport and logistics industries, enabling our clients to optimize and improve their operations.

Niels Bomholt

Niels Bomholt

CIO, Cobham Satcom

Niels Bomholt is a highly experienced professional with 25.9 years of work experience. He is currently working as the CIO at Cobham Satcom in Denmark. Prior to this, he held various positions at Kopenhagen Fur, including Head of IT Business Partnering & ERP and Project Manager and Owner.

Watch the previous episodes

AI Watch Newsletter: Episode 12
AI Watch Newsletter: Episode 11

AI Watch Video Newsletter

Get the latest know-how from those in the know. Sign up for our AI Watch Newsletter and receive the latest insights from AI experts.