Skip to main content

Driving Revenue with AI-Powered Dynamic Ancillary Pricing

airBaltic has been partnering with PROS for over 10 years to deliver airline-led offer creation. Using PROS Offer Creation and Retailing solutions - PROS Shopping and PROS Merchandising, the leading airline in the Baltics is managing their entire fare and ancillary catalog in real-time and customizing offers based on granular customer segmentation. Join this session to learn how the airline is taking their retail capabilities to new heights and gaining a competitive edge by adopting PROS Dynamic Ancillary Pricing to automate and optimize seat assignment pricing, and drive +6% revenue increase per passenger.

Full Transcript

Paul Hohler: Good afternoon, everyone. Welcome to this next session. I think we are the next to last one, so thank you very much for staying with us. I know it's been a long three days, and hopefully you learned a lot this week. And so, but again, we're glad you're here. My name is Paul Hohler. I'm a product manager at PROS, and we're gonna be talking today about one of our most recent products from PROS, which is PROS Dynamic Ancillary Pricing (DAP). With me coming up is gonna be two representatives from airBaltic. So we did a recent study with them on Dynamic Ancillary Pricing and they're gonna talk all about the wonderful project. And so it's Jolanta Rema who's gonna be joining us, and Iuliia Granja Velasco, also from airBaltic.

Paul Hohler: So I'm gonna turn on over to Jolanta, and she's gonna talk about the project from airBaltic....

Jolanta Rema: Thank you, Paul. Hello everyone. I think I have met already during those three days with many of you. Some of you most probably were in the masterclass we had on Monday on Offer Creation, where I shared some insights on what we are doing with PROS. But today before we go into use case on how airBaltic is using AI to automate and optimize seat pricing on Let me still give some background of this, because here could be some new faces, [laughter] which I didn't meet before. So I'm with airBaltic e-Commerce team for since time when we had first this crazy idea to start to sell airBaltic flight tickets online on Then again, we moved to another crazy idea to decouple our baggage from flight fares and started to sell meal pre-order for our economy class passengers.

Jolanta Rema: Yeah. And then at some moment we came to the conclusion that we should charge customer to reserve a seat on plane, obviously. Yeah, many things have changed since then. But I can repeat again and again, what I said on Monday and what I heard already today someone was quoting that we, airlines, are here still to make money through delivering excellent customer experience. And this is what we do at airBaltic. First of all, by having new modern single type fleet and constant investments in new technologies. By end of this year, we are going to have 50 Airbus A220-300 aircraft, which are operating more than 130 routes from five bases, 64% of our flight tickets we sell on And this sell share, of course, we have achieved thanks to proper technologies behind.

Jolanta Rema: And as well, thanks to our partner support PROS and as well, I wanna praise here our IBE provider, 2e Systems, definitely because it's three party cooperation that led us to that success. I can say, yeax, our story with PROS starts back in 2011 when we teamed up with Vayant Travel Technologies to fix flight shopping and pricing issues we had on our website at that time. And I think we were not the only airline having those issues when prices we showed on were not actually available for purchase. But yeah, Vayant at that time helped us significantly to improve the situation. And later, as we know, Vayant was acquired by PROS. Today, PROS already plays significant role in our technology portfolio. We use it for flight shopping and pricing, and as well for merchandising.

Jolanta Rema: Recently we introduced this new thing, PROS AI-Powered Dynamic Ancillary Pricing (DAP) exactly for seat reservations on Why did we go for this? First of all, I must say that PROS Merchandising has a lot of capabilities to explore, to use, but of course it requires some human interaction, which is time consuming, efforts consuming, and that is part we saw that probably we could find some improvements with. And as well, if there is human interaction, then of course errors, mistakes can happen and it's not always easy to spot them. As well must say that PROS Merchandising project success, which help us to increase seat reservation revenue per pack by 30% was really encouraging factor to try something new, something modern. But main driver, I think from our airline perspective, it was that airBaltic in nature we like innovation.

Jolanta Rema: This is our nature to do something new, something that no one has done before. And if you speak about AI, then at airBaltic, we always have embraced AI since it appeared as tool that can help us, that can help us to drive revenues and not a threat. So, and today, this modern business world, modern airline world is not imaginable without AI. So all this made perfect combo to go for this project together with PROS. And now I will invite my colleague Iuliia, who is actual project lead on airBaltic side to implement DAP and as well for future activities we'll do in this area with DAP. So Iuliia will share more on goals and objectives we had.

Iuliia Granja: Good afternoon. Thank you for introducing me, Jolanta. So, as a project manager back then, yeah, last year, last September before we were embarking on this journey of DAP implementation. Of course, I worked together closely with ancillary revenue management team in order to define a goal. So at the end of the day, we could be able to measure the results of this pilot project for us. And without doubt, it's not a surprise that one major goal was to maximize revenue. And at the same time not only gain maximum revenue but also keep our passengers find the price of the seats reservations still attractive and keep their desire of purchasing the seat willingness to pay is still high. And back then we were thinking about modest to 3% of a revenue increase per passenger before DAP actual implementation.

Iuliia Granja: At the second goal we saw some potential in DAP to help us cut the maintenance effort on our side in terms of the resources we use of a pricing team to manage and adjust pricing. Before DAP implementation, we had in place rule-based pricing system which was pretty sophisticated with a lot of rules. And whenever we had massive change, we had to spend some time to not only just to set up the prices, but to do the cross-checking. So everything runs really smoothly. Yeah from that perspective, we had high hopes for AI and finally, of course as airline industry is very competitive. And that is why we were looking for the ways to stay ahead of the competition and to have some competitive advantage. So these are the three major things we were pursuing. And before I share with you a bit on more details, how we actually implemented everything step by step. I want to give back the word to Paul who will open a little bit this black box for you of how AI works, and maybe you won't be that afraid to try it out as well. Paul, you're welcome.

Paul Hohler: Great, thank you guys both. As mentioned, I'll talk a little bit about how we've implemented this and how the system is actually generating the prices for airBaltic for it. And so again, the key word of the week has been AI. We're using AI technologies to price those ancillaries. And we've set, as we'll show in the next in a couple slides, that there is still a product catalog for the ancillaries that they sort of set up the rules, the conditions, the base prices of these things. And what we're doing with the dynamic ancillary pricing is providing recommendations on modifying those prices. We're doing this in real time and that we are also each day learning through this cycle, so that the model gets updated each and every day to improve on this.

Paul Hohler: There are three key components. There's a customer segmentation, reinforcement learning, and efficient exploration. So talking a little bit more about each of these. The customer segmentation is really the key driver of the price differentiation. This is sort of determining which different segments and how granular I'm going to price the situation. This segmentation we use 15 plus dimensions to determine it and it may map and be very similar to some of the rules that airBaltic had in place beforehand. But we can also go much further than that. For each of those 15 dimensions, there might be 2, 3, 4, 10 different values for a particular feature. And so that allows us to have millions of different customer segments by which we could sort of price for. And so, if you were doing this manually, you would have to put in those millions of different strategies and rules into your merchandising system to accomplish this.

Paul Hohler: And obviously through some simple setups, we can do that here and power that differentiation. The second is the reinforcement learning aspects. And this is where the model is updated, each and every single day. We make offers into the marketplace. We see what those prices were, we see which were successful offers. And then based upon that and what those were, the conversion rates there we're updating the models and so that the models are continuously updated and learning from this process. Now, this is sort of one aspect of that key AI sort of process of continuously learning on these things. And then last is the efficient exploration. What we're doing is that we provide in terms of the configurations different price points that you can be offering for each of the different ancillaries.

Paul Hohler: And so this provides sort of guardrails, so a min and a max, but also which prices in between those guardrails they provided. Now, if you were just doing a exploration process on your own, maybe you'd be offering, the first price in this case, maybe $22 on Monday and then $24 on Tuesday, $26 on Wednesday, et cetera, et cetera. That process is not very efficient. It spans all of the parameter space. But it has the possibility of offering unoptimal prices for large periods of time. And so it is gonna generate revenue losses in that. And so what we're doing through the algorithms that we're implementing is that allows us to sample that parameter space in an efficient manner so that we can converge fast. And but also to minimize your the loss basically and to maximize your revenues. So we still do that exploration but we don't do it in a clunky manner. So again to minimize that loss of things. So again, these are the three sort of key components of the AI model that is powering the results for it.

Paul Hohler: In terms of technology integrations, there's three major components. We have the booking engine, which is for, in airBaltic's case was 2e systems. The merchandising catalog was powered by PROS Merchandising, and then finally the PROS Dynamic Ancillary Pricing for that. And so in real time that the request would come into their website, and again, this was doing for seat requests. So we would send us the booking engine would send a seat request, request to PROS Merchandising, PROS Merchandising would identify which ancillaries are available, which are the applicable ones, what's the base fare of that. And they would send also requests into PROS Dynamic Ancillary Pricing. And at that point, we would use our algorithms and say, what is the model state of today? What is our recommendations for the prices?

Paul Hohler: And send those back for it. And then that would be sent all, all the way back to the booking engine for consumption and presentations to the customers. And again, this is happening in real time for every single request that's happening. We are also capturing all of those offers that were sent off in the Dynamic Ancillary Pricing system. So for every single request that comes in to PROS Merchandising that is sent on over to DAP, we capture that information ourselves. What we don't know is that what was actually successfully booked. And that's where we do need some integration with the booking engine. And so, this is sort of a batch process or sort of a nightly process that we would want to understand what were the successful bookings on this. We identified these through this offer ID, so we're using in this offer ID concept throughout this process.

Paul Hohler: And so we would want to know what was the successful bookings, which were the successful ancillaries that were purchased. And then we can again, use that information in terms of dynamic ancillary pricing to, again, calculate that win-loss and the conversion rates to update the models for that. And so that process here of this update, thing is happening in a daily rate, so that we are again, learning sufficiently fast, but not too fast.

Paul Hohler: I will comment too, that while there is this learning and an updating process, what we don't do is change the price within a single day so that we do have guardrails in place so that someone doesn't go to your website and hit refresh like 20 times looking for the best price. So if you have the same set of conditions, the same set of features, on a particular day, we would offer the exact same price, in all of those situations. Now, the next day maybe the model updates and we offer a different price, but within the same day it would be static of things. Okay, I said for the airBaltic case, we've integrated with PROS Merchandising. However, the API that we have here, we believe is agnostic enough to integrate with other vendors. And so if you're interested, we're happy to come talk to you about that integration as well.

Paul Hohler: So, that's sort of the setup. I'm gonna turn it back over to Iuliia to talk about some of the results that we had.

Iuliia Granja: So, speaking about airBaltic case, before I comment on the result, I want also to quickly guide you through our implementation journey. And, as a project manager, I'm all about timelines structure of the project. So, the project, roughly took us six months to implement. And here, I mean that, we had the six months from the very start of the kickoff meeting until the moment we could really measure the results to have some driven conclusions by the data actual one as a result of DAP implementation. And, the first month and a half that was all about preparation. And here we were working on segmentation. As Paul mentioned, we were defining some key parameters, which were important for us. And, among, most important were flight buckets, passenger buckets, and, a range of price points.

Iuliia Granja: Price points, which we defined for different seat characteristics on board of our aircraft in order to price them as accurately as possible and in the best way. And, on top of that, in parallel, we were doing technical integration, to enable our booking engine pass the information about our passengers booking preferences, prices with which they book seats on So this information was passed to be processed by PROS system, and from where AI was picking it up and processing in order to do the forecast of pricing, for the seats. Well, right after that, actually the second phase, embraced a go-live of DAP. And it was all about learning and adjusting. Even though DAP was working pretty okay, of course, it didn't have much data, but, with every passing day, this information was coming, updated once as Paul mentioned.

Iuliia Granja: And, finally like, it began to produce really like, logically correct and good prices, of course, how good they were we found out only during third stage. But in this second one, we were just observing how AI was learning and adjusting. And, during third stage, we ran a few A/B tests, meaning that we split the traffic at our website and had to measure how the DAP driven prices, were generating revenue per passenger and comparing this information with the rule-based pricing in place, so during this split, and also comparing to the results which we had in previous years. And so, after this exercise, we were able to draw the conclusion that revenue per passenger with the help of DAP, went up to 6% revenue increase per passenger.

Iuliia Granja: Which is, if we compare it to what we said at the very beginning, is, actually more and that exceeded our expectations, which is, could be considered as a success for us for a our particular case. Yeah. And, well, among other goals that we set and which materialized, there's actually this maintenance part that I mentioned that indeed, with DAP in place, we didn't have to have anyone set and update prices. So DAP does this work for us, which is very convenient, and we want more such things, more automatization. That is why we are really looking forward to try DAP with our other ancillary products with PROS. So Paul, if you can wrap it up.

Paul Hohler: Yeah, I think, yeah, so again, thank you guys again, the key things, we were able to launch an AI powered product with the help of airBaltic that provides a proven revenue uplift for the ancillary prices. This is a trusted and open solution. Again, we can go through the different models and talking through the different things and throughout the process that the airBaltic was able to monitor the behavior of the system as well throughout the things.

Paul Hohler: And the key aspect really, again, we were able to do this at scale. We were able to do this fast. The implementation time was fast and obviously it was a large increase in revenue. And I guess, I should say too, I mean these guys didn't mention it, but I mean we rolled this out for the entire network for these seats. So, it wasn't just sort of one or two markets, but they really jumped into all this and then we really appreciate that too. But it was for their entire network and all these things. So, if you have more questions, I don't think we have time. Okay. But please reach out. We're happy to answer any questions now or later on. So, thank you very much.

ABM Resources Test Stream for Your {demandbase.industry} Industry