A follow-up to "What Airlines Get Wrong (and Right) About Dynamic Pricing"
In 2025 we published a piece breaking dynamic pricing into three frontiers: continuous, dynamic, and atomic. That ladder is useful, and most airlines we work with still find it a helpful way to locate themselves on the journey.
But recording a recent Branchspace Future Picks podcast episode with Radu Iliescu (Managing Director of Branchspace Consulting) and Niels Colemont (Wiremind, ex-FLYR), they give more insight into the topic:
Dynamic pricing isn't one capability. It's at least four, and they move at different speeds.
Dimension 1: Granularity. What are you actually pricing?
Today, when an airline files a fare, that fare isn't tied to a single service. It's tied to a bundle: the right to fly, a cabin bag, a hold bag, a seat type, certain change and cancel rights. The price is set for the bundle as a whole.
Atomic pricing flips this. You price each element individually, then build the bundle price from the bottom up. The bundle can (and should) still be optimised as a total on top of the sum of the prices of its atomic elements, but you start from a much more informed position if each atomic element is individually and scientifically priced.
The problem: A lot of airlines today can't tell you how their £500 bundle price breaks down across these components. They've never had to calculate it that way. Revenue management teams who genuinely value flexibility as a segmentation lever often can't tell you how much of that £500 is attributable to the 'Right to Fly', versus flexibility conditions (the 'Right to Change' or 'Right to Cancel'), versus the included bags or seat reservation. And without that data, atomic pricing is impossible. Price optimisation relies on historical data, and the data has to exist at the level you want to optimise.
This is the first thing to fix, and it's a process and data problem before it's a system problem.
Dimension 2: Pricing model sophistication and optimisation
There are roughly four levels here:
- Table pricing. Static. "This route on this date costs X."
- Rule-based pricing. A reference price with rule-driven adjustments. Common today for ancillaries: a £20 base seat price, plus £10 for a window, minus £5 for a middle. Until a few years ago, this was actually called by some vendors "dynamic pricing." It isn't.
- Demand forecasting. The pre-computed demand forecast bid price curve approach airlines have used in revenue management for decades. Output: a bid price for the next seat sold, which is optimised at an aggregated level (not contextualised for each specific request).
- Machine learning and Willingness-to-Pay models. The current frontier. Vendors are using models ranging from logistic regression and clustering analysis (for segmentation) to architectures inspired by large language models for price optimisation.
The line between the last two is somewhat blurry, and that's fine. Most airlines have a mix across the portfolio. A flight might be priced with a forecasting model which has been optimised and refined over decades, while a checked bag or lounge access are priced based on a simple table. Knowing where each product sits is the first step to deciding where to invest.
Dimension 3: Real-time vs offline optimisation
An often-overlooked dimension, but one that matters: a bid price curve is usually re-calculated offline (as a scheduled job and/or triggered by booking events). Best practice is now to re-optimise it every couple of minutes. Most airlines only re-optimise once a day, or every few hours.
A Willingness-to-Pay model, by contrast, can be executed in real time, synchronously, triggered by each actual shopping request. Different signals, different cadences.
Two airlines could both call themselves "dynamic" and operate on completely different clocks. The one re-optimising every two minutes will respond to demand shifts the other hasn't even seen yet. The same is true once you start thinking about the sell decision itself: in a world where availability is no longer a binary yes/no but a question of "at what price are we still willing to sell," the speed at which you can recompute that answer becomes a competitive variable, not a technical detail.
Dimension 4: Contextualisation
Not personalisation. We're cautious about that word because it implies individual-level price discrimination, which is broadly illegal and the wrong battle to fight anyway.
Contextualisation is different. It means feeding the pricing model signals that improve relevance without targeting individual persons, but individual searches: search parameters, loyalty tier, competitor pricing, time of day, even weather. (Yes, weather. For some airlines, the worse it looks outside, especially in summer, the higher the Willingness to Pay tends to be.)
The more signals the model can absorb, the more responsive the price. This is where AI-driven pricing earns its place, not as a black box, but as a way to handle more inputs than a human or a forecasting model can.
A note on what this is not
There's a temptation to read "dynamic on every dimension" as the goal. It isn't.
As Niels pointed out in the podcast, humans don't want infinite choice. A restaurant menu with 50 dishes makes diners anxious and signals lower quality, not higher. The same applies to fare brands and bundles. The realistic future state isn't a fully bespoke bundle for every shopper. It's likely a couple of standard fixed bundles, with another one or two contextualised options layered in. The point of these four dimensions isn't to max them all out. It's to know where you are on each, and to choose deliberately.
Implications for your roadmap
The reason this reframing is useful: an airline can be advanced on one dimension and primitive on another. We've seen airlines with sophisticated WTP models still running their availability and bid price curve re-optimisation only once or twice a day. This can be attributed to computational limitations, legacy PSS constraints, integration complexity, organisational process maturity, and dependency on offline optimisation engines.
There are airlines doing real-time pricing on the Right to Fly, but pricing seats from a static table. And there are airlines which have invested heavily in algorithmic sophistication, but still can't tell you the atomic price components of their own bundles.
When you sit down to plan a pricing transformation, the three-frontier ladder asks "where are you?" The four-dimension model asks "where are you on each axis, and which axis is holding the others back?"
That's a more useful question, and in our experience it's the one that changes investment priorities.
If you're not sure where to start: pick the dimension with the biggest gap between your current capability and your competitors', not the one with the loudest vendor pitch.
You might also like:
- What Airlines Get Wrong (and Right) About Dynamic Pricing
- The Pricing Frontier podcast episode (Spotify)
