Redefining Revenue Management for Rail & Bus: Integrated, Real-time and AI-assisted

Redefining Revenue Management for Rail & Bus: Integrated, Real-time and AI-assisted

As rail & bus operators face an increasingly competitive market, revenue management is no longer a nice-to-have – it is a core tool to optimise revenue, effectively match supply and demand, and guarantee that occupancy is as efficient as possible. It is the difference between a full train and one running half-empty.

Yet, most revenue management systems running today were never built for rail & bus. They were borrowed from airlines: operators adopted revenue management in the early 90s, following the airlines that had pioneered it two decades earlier – and most systems still carry that inherited logic. But a train or a bus is not a plane, and that mismatch is what’s costing operators revenue.

The operational gaps in traditional revenue management

The typical RM systems found today in rail & bus operators have several limitations. This is mostly because these are legacy systems, built on revenue management approaches for air travel, since that’s what was available at the time. A lot of these systems are still in place, but they have clear shortcomings:

  • Based on airline logic
    The reality of ground passenger transport is more complex than simply going from A to B. Rail & bus have multiple origin-destination pairs (because of all the stops within a route) which need to be treated independently for revenue management optimisation.
  • Updates in batch
    Revenue management systems need to interact with the IMS (Inventory Management System) – very often, this is a batched update that happens once a day. The RM is always behind, working with yesterday’s data.
  • Hard to build forecasts
    Forecasts need to be built from scratch, using custom logic and requiring high maintenance, eventually leading to them being dismissed or not used at all.
  • Isolated from the rest of the system
    Typical RM systems for rail & bus run as standalone tools – they connect to the IMS through a batch feed instead of being natively integrated. Thus, the data the RM works with is thin (little more than origin-destination, train number and class of service), without direct access to the full commercial picture. The analysts can’t get specialised, real-time dashboards, having limited visibility into what is actually driving performance.

These limitations lead to a gap between what the system sees and what’s happening in reality, resulting in lost revenue. Rail & bus operators need a better system: one that challenges these norms and is built for their operational and commercial reality.

Redefining the norm in revenue management for rail & bus

The Revenue Management module in S3 Passenger, developed by Sqills, addresses these limitations, since it’s built natively into the platform – where reservations, tickets, timetables, inventory, fares and customer data already live. Thus, there’s no integration layer to maintain, and no need for data syncing between systems.

This approach allows revenue management to be always connected to other modules like Navigator, Fare, Pricing Engine, Inventory and Ticket. So, more than a data feed model, what we have is real-time integration, where the module is positioned at the heart of a full-fledged PSS (Passenger Service System).

Besides this, the module also overcomes the legacy logic inherent to many systems currently in place. Since it was built from scratch for rail & bus, it specialises in optimising not only the whole route, but all the possible independent origin-destination pairs.

Forecasting is also improved when compared with typical legacy systems. The module uses machine learning models, which are retrained on the latest booking data, so forecasts can keep up with demand. After each retraining, the system will report the forecast accuracy (measured against data the model hasn’t seen). This way, analysts know how far they can trust the numbers.

In fact, accurate forecasting and price recommendations are very hard to achieve in rail & bus, due to the “booking limit dilemma” and the “curse of dimensionality”, two concepts that are important to understand.

Overcoming the booking limit dilemma & the curse of dimensionality

Rail & bus operators sell tickets at different price points, organised into what we can call “buckets”: a train or intercity bus might have multiple “buckets”, at €10, €30, €70, etc. Analysts need to figure out how many seats to allocate to each of these buckets, thus establishing the “booking limit” for each bucket.

The core of the “dilemma” is how many seats to assign to the different buckets: too many seats in the €10 bucket fill the train fast, but miss potential revenue; too many in the €70 bucket generate more revenue per seat but risk running a semi-empty train. This can become incredibly complex, as seats need to be allocated to different buckets, in different trains, for all the possible origin-destination pairs.

As an example, let’s think of one route from A to B with no intermediate stops, 15 price buckets and 500 seats. Even assuming booking limits only increase from one bucket to the next, that single route already has 18 octillion different possible seat allocations.

And that is the conservative version – a real route with multiple stops will sell many origin-destination pairs at once, and each one multiplies the problem. Across 40 OD pairs (common on intercity networks), the combinations run to more than 1,100 digits.

This is what we call the “curse of dimensionality”: how rapidly the number of possible seat allocation combinations grows, as the number of pricing buckets and origin-destination pairs increases.

Traditional systems deal with this problem using methods such as EMSR (Expected Marginal Seat Revenue) or bid-price approximations, which keep the calculations manageable by relying on simplifying assumptions. But those assumptions create gaps between the model and how passengers actually book – gaps that widen with every extra origin-destination pair and price point.

The Revenue Management module in S3 Passenger takes a different approach, using AI and machine learning to learn the structure of the problem from the available data.

AI and machine learning based revenue management

The module uses AI and Machine Learning to help RM analysts overcome the booking limit dilemma. It is trained on the operator’s own data and it learns the patterns behind it: how demand builds up for each departure, origin-destination pair and class, and how passengers respond to different prices.

From what it has learned, the system recommends the booking limits that maximise the expected revenue for each departure. When a departure sells faster or slower than forecasted, the system adjusts its recommendations in real time, reassigning the remaining seats across price points accordingly. As new data arrives, the models are retrained, so the recommendations keep improving over time.

Thus, the system finds optimal booking limits by learning from the operator’s own booking data. And because the module benefits from the wealth of data of S3 Passenger’s pricing, ticketing, and inventory management modules, it can also use demand signals that appear even before a booking is made, like journey searches, look-to-book data or abandoned carts.

This foundation even enables continuous pricing: having bucket prices as anchor points, the system can set any price in between when demand justifies it.

Improving the learning process: human in the loop

The system learns from the human analyst as well, which is fundamental for its improvement. Instead of simply deploying the calculated booking limits, the AI recommends them. So the human always has the final say. The system tracks how the human analyst responds to its recommendations: which ones (and for which departures) were actually applied in production, and which ones were changed or modified (and by how much).

This feedback loop between the AI and the human analyst is vital and dramatically improves the system’s efficiency, since evaluating the AI’s decisions only in a simulated environment is not ideal.

Conclusion

For rail & bus operators, revenue management is today a core tool and a fundamental asset for their operations, which is why having the right system for the job becomes a competitive advantage. Legacy RM systems, based on airline logic, updating once a day, and living independently from other systems like the IMS and the PSS, can no longer serve operators well.

Fortunately, operators can now benefit from a solution that is perfectly suited to their commercial and operational needs – integrated with other data sources,  working with live data and adjusting booking limits in real-time.

To achieve this, Sqills relied on new technological developments (such as AI and machine learning) but, more importantly, on 20 years of experience working with more than 60 rail & bus operators worldwide. The result is the Revenue Management module in S3 Passenger: built specifically for them, and developed with them.