Stefanie Schuster: Hello and welcome to our podcast Loaded, Talking Software for Connected Aircraft. In today's episode, we want to talk about personalization. It has been a buzzword in this industry for quite some years, but I would say it's only now that it's also going more into technology discussions around how to actually implement it. So let's start with: where do we stand today? How would you see the current state in the context of personalization?

Ralph Wagner: It's different from airline to airline, but the difference is mainly due to the human factor. It's basically those who manage to inform their crew, manage to inform their staff who is actually on board, and so on. So it's basically within the cabin — it's more transported through the people in the cabin. But what is missing is the digital part of it. From a digital perspective, the passenger is just unknown. So systems on board as such don't understand who is actually on board. Certainly the crew does. But the problem here is that you also feel welcomed if an application is welcoming you, if a system knows your settings. In this regard, there is really not much on board. There is a little bit, but I think also due to the infrastructure of how an aircraft is, and the connectivity, or the missing of a really big pipe, this is actually causing quite some headache: how to do personalization in a digital way.

Stefanie Schuster: We can look a little bit into the different layers of personalization. It's, of course, eventually per person. But there is also a lot of discussion right now on how to target per fleet, per tail, per cabin. So it might also be some stages you can approach and target as a first and next step.

Ralph Wagner: Yes. I mean, this is even a subset of it, right? So to have an understanding of where an aircraft is flying to, or what type of aircraft it is, or what length of flight it has — and due to that, this also influences the personalization and the experience of the passenger. That is then another layer to the real personal target, that I exactly know who you are. All of these parts need to be discovered and need to be used. All of these data points.

Stefanie Schuster: And data points also means — there are already some initiatives around personalization, but I would say it's right now still rather different pillars, different boxes, because one team thinks about advertising, another team thinks about the loyalty. It's different teams internally, different vendors potentially. So we should also look into how to centralize that, when we think about it: it's one passenger, and there might be data points from different angles.

Ralph Wagner: Yes. And when we are going into that direction, a very important part is: you're going to make the journey of the passenger better — you make him feel better, you make him feel welcome, and so on. But it's also for operations, it's also for other aspects, for getting stress away from the crew. It's more efficiency in how you plan journeys. So all of this data that you actually need to do personalization is data that you need for many aspects of your operations. And this is why it's a little bit like the question: how do you set it up from a software perspective? What kind of software setup do you need to actually get all the data points right in place, to benefit in different areas from it? I think that is an important question to tackle.

Stefanie Schuster: Let's jump into the technology. We already touched the data points. So one area is about how to collect data, how to get data in and out. The other area is how to analyze it, make interpretations, put some rules in place. Let's first look into: how do we even collect data? What is needed to collect and store and use data?

Ralph Wagner: To connect the dots, you obviously need all the infrastructure knowledge of an airline: all the planning, the flights, disruptions, everything around that. On the other hand, you need the passenger data — you need what the passenger is doing, everything around it. When does he board? What does he watch? What is he eating? What is his preferred seat? All of these parts, and all of the operational parts, are important to connect with each other, which means you need a place somewhere where you can connect these dots, and connect them in real time — because the real benefit you bring is that you can make real-time decisions. To make this possible, there is obviously only one place where you can do this: in the cloud. So you need an infrastructure where you have a type of a data lake, where you collect all of this data, where you let all this data flow in and all of this data also go out, and from there you control it. Which means you need a permanent connection to the aircraft that allows you to collect all the data captured within the aircraft, and you need something on the ground that combines this with other data points. The main thing here is: get the data together to build the real, full landscape of data points you can collect.

Stefanie Schuster: Maybe even more concretely: what does it mean to get the data together? Ideally, for personalization, it would be one passenger record, right? Envision it like a file where everything around that passenger is written in. Of course, there are topics around personal data, about anonymization, about what vendors should be allowed to read and write — if it's just an ID. But in the end, as an airline, it's a passenger record where all this data around that one passenger is stored.

Ralph Wagner: Absolutely. When we see this passenger-centric, then it needs to be a data point where you have a passenger file, and this file is a flexible artifact where you can write in data, also from different vendors, where you need to design an entry point to it. So when you think about it, you have a specific gateway that really controls all the traffic in, traffic out. You need to have the concept of how you collect all of that, and for sure also how to combine it with other data information — for example, is my flight delayed, etc. All of this in combination basically creates your data point, where you say: here I know everything about my passenger, and I allow other systems to interact with that. So I allow my IFE system to call data from that. I think that is a very crucial thing to consider. So when you are setting up new systems on your aircraft — something for interaction, a shopping system — you need to think about how you get the data out of that, or the requirements you need to have: how do they need to store data, and what data do they need to deliver to you. The same way somebody is using an IFE player: if it's on a personal device or on a seatback, it doesn't matter. Where do I stop? Where do I continue? What languages do I choose? What genres do I watch? All of this needs to be collected in a centralized place. And if you have the connectivity, you can do that, right?

Stefanie Schuster: And we talked now about this passenger record with personal data. But of course you also need segmentation there, right? Because you might use that data for services that make decisions just based on patterns — to be able to put the passenger into different segments, where it's not really about personal data shown to him personally, but where you want to put him into categories to show content, to show offers.

Ralph Wagner: All the talks about AI, and how to make it most effective within your company, you need to connect it to the data you have in your company. This is a big headache for many companies: how do I get all of my information in a place where I can, for example, use AI to analyze it? This is a very similar approach. You do recommendations, you do segmentation, you also do analytics for your own planning, for your operations. This means you need this data combined somewhere, so that such systems can investigate it and make decisions that help you make the customer a specific offer that he likes, to show him information that is relevant at the time he's consuming it. This is something where, concept-wise, this architecture thinking — how do you manage all of this passenger data on the ground, to be used on board, before the journey, after the journey — is very important to think about: how to conceptualize that.

Stefanie Schuster: Where do you see the challenges? I mean, when it's about collecting data and making it available. One thing is, yes, you need this data lake, you need some cloud infrastructure. You need some way to assign permissions for external parties to read and write. There might also be challenges around keeping the records up to date, right? Caring for, I don't know, 'continue where you left off', having movie IDs change, keeping information in sync. Any other challenges you see?

Ralph Wagner: Oh, the challenges you mentioned are already challenging enough. The technical approach would be that for all of this information, you need to have procedures in place that are capable of updating the data all the time. So you need a performant infrastructure for that, and the capability, when you connect to systems, to permanently learn — to permanently add data, remove data. And what should not be forgotten is: you need to implement all the regulatory parts around data privacy. So you also need to understand that you do not really need to have the passenger identity — you could also operate on an ID, and think about deriving patterns. When you see that the majority of the passengers who fly from A to B do something specific, you can already derive patterns for such passengers, or for those who sit in a specific area of the cabin, or who order specific food, or consume content on the IFE. You can derive patterns from that, and put new passengers you identify into such patterns. So you do not necessarily need the name of the passenger. For sure, if it's connected to your loyalty program, to a real identity, that's the most benefit you can get out of it, also for the passenger. But if a passenger decides not to agree with that, you can find other ways of using the data he is producing, analyze it, put it into patterns, and then help him make his journey more personal — even if you do not know the name.

Stefanie Schuster: That is also an interesting topic, because it is about content, yes, but if you provide value out of the data, that might be content too.

Ralph Wagner: Absolutely.

Stefanie Schuster: Okay. So now we looked into collecting the data, storing the data, getting it in and out for different systems, having different systems contributing to this passenger record. The second topic we wanted to look into is how to use that data — how to analyze it, where to analyze it, where to make interpretations, how to put business rules into this infrastructure. Because data alone is not the deal, right? You want to be able to use the data in a beneficial and reasonable way.

Ralph Wagner: Absolutely. So what we discussed right now, if we would end here, it would basically lead to 'continue where you left off'. You know the passenger, you know his settings, and you provide him the service to deliver these settings and this information on his next flight, on his next journey.

Stefanie Schuster: Just one data point.

Ralph Wagner: Exactly: what kind of settings I had on my business class seat, or light settings, and so on. You can store that, and if I'm coming back, you can provide me the option to just execute the settings in one step. But these personalizations, if they really should work in a way that somebody feels known, you need to interpret them. And there are basically two options to do so. The one option is to do it on the ground, asynchronously. As an airline, you can analyze the data, segment it, learn from it, and make recommendations — for example, by analyzing what kind of meals somebody likes, or what kind of preference somebody has on a night flight.

Stefanie Schuster: Okay. But there was also the difference between real time or...

Ralph Wagner: Yes, exactly — this is not in real time. So this is something where you are adding, as you described, to your personalized file. I'm adding additional information that I derive from the data I have, so that if you are entering an aircraft and get the menu of the flight, I can already make you a recommendation from the menu. I can say: maybe you want to have this, because I know what you took on the last flight. So this you can prepare; this you don't need in real time. If somebody always goes to the lounge, and you know that he does that, you can provide him directly with the information, when he sits in his seat, on how he gets to the lounge. This is something you can prepare, and then offer as information. An additional way to do it is in real time — to analyze what I'm really currently doing, and make a decision here in real time. Easy to understand: when I'm watching movies, and currently into a specific type of content, then this is analyzed in real time, and in real time it's recommended what I could watch next. Such decisions and implementations look very different from a software architecture point of view.

Stefanie Schuster: But that can also be different steps, right? You can build one on the other.

Ralph Wagner: Absolutely. You can first start with one and continue with the other. But one requires, on the side where you store the data, to basically enhance this. The other requires that the devices used for interacting with your passenger be made smarter, so that they are able to dynamically react to the passenger input — also triggered by the ground, but also triggered maybe by applications that sit on board and make decisions in real time to offer you something derived from your actions.

Stefanie Schuster: Okay, that is now also an interesting area. The decisions in the cloud, that's one topic we covered. But decisions on the device on board also require some thought around how and where to do it. Is it the front-end application? Is it some server software?

Ralph Wagner: Absolutely. It's also: how do you combine that? Do you have your application that actually has some patterns stored, or do you have a server on board that maybe has the capability to be more powerful? It depends on the complexity, how much power you need. So you could think about: you have something on board, or you're using the cloud. If you have a direct connection to the cloud, you use the cloud to process the data fast, and deliver back the recommendation of what you should do next. But the design of the central point — where the data is first used — is on the device. So it needs to be captured there, and then analyzed in a very fast way. It depends on the use case, how complex you want to go, and where you make this decision. For example, if you're looking at advertising, those decisions are often made in the cloud, because they are quite complex — many interaction points are captured and analyzed in real time, and usually matched against a big portfolio of ads. So you need to make a selection of one ad out of several thousands.

Stefanie Schuster: So, even with bidding logic and so on.

Ralph Wagner: Exactly.

Stefanie Schuster: Commercial implications.

Ralph Wagner: There are even commercial ones, totally right. So you have a bidding logic, you have commercial implications. This is something you cannot run on board — you run this cloud-based. But for this you need to capture this data, be able to fire it back, and get your advertising within a second. And this is something that, from the technology point of view, is very exciting: to have the capability to analyze the data you really produce on board, not only what you store in the cloud.

Stefanie Schuster: Okay. So the interpretation and analysis can happen in different areas: in the cloud, real time or any time, and also on board, on the device, potentially on the server, in a fast way, real time. You also mentioned there are different data points. One is the customer or passenger: who are they, what are they doing right now, and what did they do in the past? Then you have operational context, potentially — so what's happening on the flight? Is there some delay? Is it meal time right now? Now let's go into how to connect all those different data points, no matter where, on board or in the cloud. But there still needs to be some logic put in place by the airline to define how they want to use it.

Ralph Wagner: When we are talking about the airline and how they want to use it, and how this whole setup should then look in the end, this is something where you need to look at what they have right now. What are their processes? Today there is a loyalty program that actually is around the passenger data. So, how to extend this, how to learn from this — how to basically have your personalization file and your authentication mechanism combined. How do you get this in place? You need a service that sits in the center, where you have the authentication on one side, your operational flight information on another, your advertising, your IFE catalog, your food on board, etc. This is basically a personalization service that sits in the center, that can be used by different parties to combine the data. I think that is something where an airline should look at how to build that up. And I think, in the first place, it's your authentication, your loyalty, with your passenger file. That's the first step. And with this you can start to tell me where I can continue where I left off my movie. So with this you start, and this is something you make accessible from your aircraft — you integrate it in your player, in your catalog, so that you know where I need to watch further. With this you start, and then you build more and more information around this file. You add more and more system data to it. And this is how I would start.

Stefanie Schuster: So let's summarize what the airline needs to get started, or to get more evolved around personalization. We said the one thing is a data lake, or concretely, a place where data is stored and which allows the airline, but also vendors, to write and to read, based on permissions and rules the airline puts in place.

Ralph Wagner: Right.

Stefanie Schuster: Which also means, potentially, some kind of gateway, authentication, permission logic, anonymization, segmentation, to make sure different systems can contribute, but can also take out information to make the service more personalized.

Ralph Wagner: Exactly. So you need to look at how you want this data to be. Should you know the passenger? Should it be an anonymous ID? How do you want to handle this? How do you derive the first patterns? How do you derive the way other systems can write, can read out of the data? Whom do you allow, which type of access? You need to think about how you do your access management, your permission management, for other systems to read and write. And then, for sure, you need to think about how data gets in — how your own systems write in. You have the systems sitting on the cloud side, but you also have the systems on board: your crew applications, your passenger applications, your IoT applications that maybe remember my light settings in my cabin, or my preferred seat settings. So, how do you get all of this data in?

Stefanie Schuster: Then, on the other side, you need teams and vendors who work on this, who provide the data, who use the data based on patterns and rules you as an airline put in place. I think it will be a requirement, for vendors definitely, but also for other internal teams, to work together based on one central setup and rule definition which is set in place.

Ralph Wagner: Exactly. This is a very good point. You need to set up your rules, your requirements as an airline, to give to your vendor: if you're doing this e-commerce application, this advertising, this entertainment for me, then you need to provide me this and this data point. And here is my API, here is my gateway — this is the way you connect to it, and this is the data I want to have, and here is data I can give you, so that you can use that for your interaction. And this concept should be settled — it should not be the other way around, that somebody comes and says: this is the data I give you. And then you need to combine this with data from another vendor, from a third vendor, and the data doesn't match. You don't manage to make meaning out of it if you combine it. That's the dangerous part.

Stefanie Schuster: Both directions won't work, right? It should not sound like the airline or one team just defines what kind of data they want and have, and the others just provide and use the data. But it's also not that somebody says: here's what I have, deal with it. The airline needs to define what they would like to achieve, and all the other teams and vendors contributing can add value, because they have specific expertise in the area. So with close collaboration, you can probably evolve that very nicely over time.

Ralph Wagner: Absolutely.

Stefanie Schuster: Okay then. Thanks for a nice talk around personalization.

Ralph Wagner: Thank you.