This interview is part of our exclusive ‘Airline Leaders Series,’ conducted in collaboration with Branchspace – digital reinvented, aiming to transform airlines into better digital retailers. Throughout the series, we’ll highlight the key concepts that are shaping modern airline digital retailing, ensuring you gain valuable insights from each discussion.
Leader: Ivan Karlovic
Airline: Norwegian
Role: Director – Data, Analytics & AI
Location: Norway, Oslo
Talks about: Data, Analytics, Artificial Intelligence, Machine Learning, Cloud
Why do you need to read this interview:
Ivan is a seasoned expert in data, analytics, and AI. Ivan brings a clear vision to the table on how to seamlessly integrate data science and AI into airline organizations, empowering businesses to fully leverage AI technologies for enhanced decision-making and operational efficiency. Through this interview, Ivan shares invaluable insights on constructing an agile, cloud-based AI and data architecture, offering a blueprint for airlines aiming to innovate and stay ahead in the competitive landscape. His perspectives not only illuminate the pathway for embedding AI into the core of airline operations but also highlight the critical strategies for building a future-ready, data-driven organization.
Ivan: If you looked at our summer program announcements, etc., we continued a modest growth path. Not resembling our previous decade of explosive growth, but still continuing to extract the benefits of scaling. Increasing fleet, increasing number of destinations, etc. Again, we continue to do that, but continue to focus on the short-haul market, European and North African destinations. So in that sense we continue or keep the direction which has served us really well. During this year, we are reporting almost record financial results.
I would say most companies are potentially reporting some negative effects of inflation, general economic circumstances, but luckily, at least when it comes to our booking patterns, of course, we’re reporting that we’re doing quite good. But I would say evolution of the current strategy has been executed on. So we definitely look forward to next year.
Ivan: We just continue to deliver our long-term plans, which are around decommissioning our legacy data warehouse. Just recently, a month ago, we committed to a date next year to phase out our current legacy data warehouse and to move everything, all the use cases, all the reporting, everything finally to AWS. It’s been a while since we started building the foundational pieces to be able to do that, and now it’s clear enough that we can put a more hard date on phasing it out.
This of course requires retraining, rebuilding stuff, etc., so this is something that at least our team will be heavily focusing on during the next year.
Ivan: Of course, [migrating to the cloud] drives a lot of benefit. Basically, we don’t want to own two systems which have similar functionality that double the cost. So we of course want to also make sure we phase out things with that lower cost and simplified because we don’t want to have the complexity of managing a couple things in parallel. But of course, AWS isn’t just another thing to report on. It opens up other types of use cases. These days we’re launching some of our first machine learning use cases, something that we previously couldn’t do. Of course we’re looking at all different gen AI opportunities, either things we can build ourselves or with some vendor involvement. So again, having a modern platform enables us to not just rebuild all things but really move into new areas.
You continue to grow as you mature. We built this analytical platform primarily for those purposes, but of course there’s a whole IT cloud migration effort in parallel. So yes, again, these technologies allow you to scale when you need it, etc., and give you a very wide set of opportunities to go for. Of course, they require specific knowledge. Even though the services offered across the cloud vendors are similar, the scale that’s needed or specialties are different, so a lot of training and finding the right profiles of people that we recruit, either through employees or consultants, to make best use of the platform. Again, it’s a very flexible platform to grow.
Ivan: I think generative AI s a highly interesting and explosively developing thing, and it’s not slowing down.We do pay close attention to it. We do experiment with it. We have our own use cases with it. But we don’t expect it to be the silver bullet for everything. We still need to do our day-to-day core operations correctly around those less dazzling jobs regularly with good quality to make sure we have the quality of data we need for any type of reporting and machine learning or running some type of gen AI on top. Garbage in, garbage out.
In that sense, we do of course want to understand the capabilities and limitations of these things. How much we should build or buy, what kind of models we should be using. We also want to get to a stage where we have a certain level of internal competence versus not rely on – because all of our vendors are pitching in. Either existing vendors say, “Hey, now we have an AI thing” or new vendors say, “Hey, my AI beats your AI.” In that sense, we’re working with making sure everybody in the organization knows they can talk to us within my team to help guide them through this.
Some of the vendor implementations of gen AI, typically around large language model involvement, have been I would say on different levels of maturity. Some have immediately from Day 1 understanding that they should be enterprise fit, they should be GDPR compliant, etc. Some vendors were running out of the gate so fast that “Okay, we’re going to run stuff on the U.S. servers. What’s GDPR again?”
We positioned ourselves to make sure the business has somebody to talk to when it comes to these types of questions, because it’s easy to get dazzled by shiny new things, but when you scratch below the surface of these things, sometimes you discover some of these weaknesses – weaknesses that can lead to non-compliance on important things, which introduces risk.
But there are other use cases where we support our own developers. We’re concluding a pilot for CodeWhisperer. It’s AWS’s AI-generated code suggestion service, and it’s geared towards our developers. It’s less risky because it’s not customer-facing, so you have more room for error. Even if it’s not perfect, it still might be useful, whilst whatever of these deployments you put in front of a customer, you can easily hit some headlines in the newspaper afterwards. In that sense, again, we do distinguish between these two angles of customer-facing and employee-facing. We start with some employee-facing ones in terms of deployment, but the customer-facing ones are being actively explored.
Ivan: We spent probably the last two years with the recruitment training, etc. We’re still continuing to recruit. Our biggest focus on recruitment last year has been on data engineering, again, to complete this foundation, that we have the data we need, that we make sure it’s good, that it goes into the foundation as it should. But now, because we’ve collected much of the data that we need to start executing on concrete use cases, I think going forward we’ll be more working towards actual deployment of more use cases. And with that, actual delivery of more things into business.
So far it’s been the guys are working on something there, it’s fairly big, fairly complex, takes a bit of time to get there. But now when somebody comes to us with a specific use case, now we have usually a very clear path to go. We know exactly what we need to do and how to deploy it, and again, we can really deliver it fast with good quality. I think next year will be more of the execution part. I would say less foundational and more use case execution.
Ivan: Very obvious use case when it comes to conversion and next best action type of approaches, so we’re looking into stuff like recommenders, etc. These are your very typical cross-industry more classical machine learning type of improvements that just makes sense to do. These are not any type of secret sauce. These are things you can pull from this long list of very obvious, I would say even fairly low-hanging fruit type of things. Yes, they do come with complexity. When we start looking to recommenders, I think they’re both an art and a science as well on how you frame them. But again, this is nothing that – I would say any other airline has either already deployed or has somewhere on the roadmap. In that sense, it’s more like covering the basics and to bring us more up to par with the best in the industry.
I think there’s more personalized – the better things we recommend, the more personalized the customer journey will be. Let’s say for using real-time data, we want to take out non-relevant communication. You booked a car; great, we shouldn’t bother you about the car rental anymore, right? Sometimes it’s a little bit tricky to make sure you’re actually able to deliver that type of experience. But again, those are sort of natural things, but sometimes technically fairly challenging when you talk to the partner ecosystem, etc., orchestration to make sure they work as expected.
I would say at least from my team’s side of contributions, and then there’s the whole CRO team delivering their different customer experience experiments, I would say we really want to grab some of these low-hanging fruits to bring us up to par.
Ivan: I think all of us say, if I just put my customer hat on and interact with different businesses, yes, I want our chatbots to be smarter. But it does come with certain challenges. You don’t want to take out a generic thing like asking “How many kilos can I bring on board?” That might be different from airline to airline, so you need to use specific techniques like RAG or otherwise to make sure the quantum fits your actual specific airline and not a generic type of average of the internet airlines type of thing. So in that sense, chatbot upgrades are very obvious use cases.
We see of course dramatic movement in the marketing creative space – how do you add creation, etc., automatically? The visuals. Maybe even videos in the future. Of course, a lot with expectation that this can both increase productivity and improve quality because you can generate the specific picture you want rather than maybe find it on Getty, etc. So in that sense, yes. I think these are maybe the two most aware things.
The chatbot – it could be somewhere around the concept of conversational booking flow if somebody sees a good fit for them, like Expedia rolled out, etc. And I’m sure a lot of the others will also roll out all of the phone assistance, etc. So it’s coming, and it will come in different flavors.
I think companies will need to adapt it a little bit to their brand and tone of voice and to make sure – what I internally always stress is we want to make sure when we have these multiple AI components running, possibly from different vendors, that we have the same tone of voice, that we have the same look and feel of what AI in Norwegian looks like, that we have similar wording. If we say it’s luggage or baggage, that should be standardized so you don’t feel you get two totally different experiences with two different AI components. Even though they might be coming from two different vendors.
If you want to learn from leaders like Ivan about how to advance your airline data science or want to be the first to know when our next Airline Digital Talk will be published, please:
No Comments