AI in Revenue Management

I'll be honest, it's a pet peeve of mine: hotel leaders still saying they’re not using artificial intelligence (AI), but they are. If your Revenue Management System (RMS) has automated pricing, demand forecasts, or dynamic segmentation, it’s powered by AI. More specifically Machine Learning. Forward-thinking hotel executives, including Chief Commercial Officers, should recognise that AI is not just chat bot's and Large Langue Models; it’s already embedded in the technology you use today.

.

How Machine Learning Powers Revenue Management Systems

Machine learning (ML) at its core means training computers to find patterns in data and make predictions or decisions automatically. In a hotel RMS, this typically involves feeding the system years of historical booking data, rates, and other factors so it can “learn” what influences demand and revenue. Some systems can even take market information, which they use to update past trends to current patterns.

Overall, there are two main types of ML algorithms used:

• Supervised learning: The RMS is given labeled examples of inputs and the desired output. For instance, an ML model might be trained on past dates with features like booking pace, market segment, and competitor prices (inputs) and the actual number of rooms sold or optimal price (output). Over time it learns the relationship, so it can forecast future demand or suggest prices for upcoming dates. In other words, the algorithms use known outcomes from historical data to predict what will happen next.

• Unsupervised learning: Here the algorithm is not given explicit correct outputs, but rather finds hidden structures in the data on its own. In revenue management, this is often used for dynamic segmentation, clustering customers or booking behaviours into groups that weren’t obvious before. For example, an unsupervised model might analyse booking patterns and discover a new customer segment (like “last-minute bleisure travelers” who book high-end rooms mid-week) that the hotel wasn’t explicitly targeting.

  
  

In practice, a hotel RMS combines these approaches. It might use supervised learning to forecast demand for each date and segment, and unsupervised learning to continuously refine the way it segments customers or events. Crucially, the “learning” doesn’t stop after an initial training phase. Modern systems perform continuous re-training and real-time optimisation. As new data comes in, for example today’s pick-up or a competitor’s price change, the ML models update their forecasts and recommendations.

Why does real-time matter? Because hotel demand is fluid, and what you knew last month or even yesterday might not be true now. AI-driven RMS can analyse data as it streams in and adjust pricing or inventory controls on the fly. A great advantage, as your revenue team won't be able to monitor every booking coming in in real time. Especially over the weekends or during nights. Though there is an argument to be made in favour of near real-time data. Systems like Duetto update their pricing decisions three times per day, which works well 99% of the time. The exception being, of course, high demand event announcement.

Another advantage is that the algorithm can recompute optimal rates continuously, where traditional revenue management would rely on rules such as hurdle rates. For example, if a surge of bookings comes in for a date that was initially slow, the system can instantly recognise the uptick in demand and recommend higher prices or restrictions on discounts. A true game-changer. As one analysis noted, relying solely on past trends to set prices fails when an unexpected event (like a sudden travel ban or a big concert announcement) occurs. Machine learning–based RMS address this by emphasising current data inputs and constantly re-calibrating the forecast with the latest information.

Key Capabilities in a Modern Hotel RMS

Machine learning allows us to use a host of powerful capabilities in hotel revenue management systems. Some of which include:

• Dynamic pricing by segment and room type: Instead of one-size-fits-all rate changes, ML allows 'Open Pricing', optimising prices independently for each guest segment, distribution channel, and room category based on demand. Which can mean that your standard room has more value than a suite, or that the price differences grow over high demand dates. Revenue managers would have previously set this up by basing the room type differences on a percentage, rather than a fixed price. Duetto is well-known for it's Open Pricing approach. The result is more nuanced pricing that tests each segment’s willingness to pay, often leading to higher overall RevPAR than traditional methods.

• Pace and pickup-based demand forecasting: ML algorithms excel at analysing booking pace curves to forecast final demand more accurately. A modern RMS looks at historical booking patterns, current on-the-books business, and booking lead times to predict how many rooms will likely be sold at various future points. Like a traditional revenue manager, the RMS would increase prices if the pace is ahead of the curve, and decrease if pace is lagging. The advantage is that machine learning can factor in nuances like different pickup patterns by segment (e.g. group business books earlier, corporate transient might book last-minute) and adjust forecasts in real time as new bookings or cancellations occur. This rolling forecast capability is far more responsive than static “same time last year” methods. A risk is that many of the current algorithm's on the market are heavily reliant on previous booking windows. As we have seen large shifts in this since Covid, it often takes time and a lot of manual overwrites to steer the algorithm back on track.

• Competitive rate intelligence and market data integration: Another large benefit is ingesting large amounts of market data like competitor prices, local events, online search trends, etc. Advanced RMS platforms continuously scrape competitor rates from OTA's and brand sites (via rate shopping integrations like Lighthouse) and feed that into their algorithms. The model can recognise, for example, that your hotel is underpriced relative to competitors on a high-demand date. It will then recommend a rate hike to realign and avoid leaving money on the table. Conversely, if the compset drops prices due to a demand dip, the RMS will catch that and could suggest you adjust as well to remain competitive. These systems also consider things like city-wide demand signals, holidays, or events by pulling in event calendars or even flight search data to adjust forecasts.

• Optimisation based on unconstrained demand: Importantly, ML-driven RMS forecasts unconstrained demand. Also known as the total room nights customers want to book, if you had unlimited inventory. Which it then uses that to optimise how to allocate your constrained inventory. For example, if unconstrained demand for a date is 120 rooms but your hotel has only 100, the system knows demand will outstrip supply. It can then make sure those 100 rooms are sold at the highest possible rates and to the most profitable mix of segments. This is one of the key features of IDeaS G3 RMS, as it explicitly projects unconstrained demand, then “re-constrains” it based on your room capacity to set optimal prices and controls. In practice, this means the RMS might suggest closing low-value channels or raising the price until the demand you receive equals the 100 rooms you can actually sell.

These examples demonstrate why today’s RMS solutions are in fact AI-driven. The machine learning algorithms are crunching vast amounts of data to enable tactics like per-segment dynamic pricing, pace-based forecasting, and demand unconstraining that simply weren’t feasible at such scale and speed a decade ago. And as we all know, the result is that we are heavily reliant on these systems, and rightfully so.

Recent Innovations from Top RMS Platforms

Several major players have been pushing the boundaries of machine learning in their RMS, as they continue to innovate pushing the industry towards more real-time, data-driven revenue management.

Duetto

Duetto has been a pioneer in bringing machine learning to hotel revenue strategy, coining the concept of Open Pricing as an alternative to traditional BAR tiers. Ranked number 1 on the hotel tech report Duetto is a well known player. Their Open Pricing means prices are not fixed in silos or preset increments. Instead, the system continuously adjusts rates for every segment, channel, and room type independently based on demand signals. This methodology, powered by ML, ensures hotels are always open for business at some price point: rather than closing off a room type or discount, you keep selling but at a higher price if demand warrants. As Duetto describes it, this tests guests’ price sensitivity by presenting all options – the guest will choose “no” only if the price exceeds their willingness to pay, whereas legacy approaches might prematurely shut out opportunities.

Under the hood, Duetto’s RMS product, GameChanger, uses predictive analytics and machine learning to recommend pricing for each booking date. It pulls in a variety of data (PMS bookings, web shopping regrets/denials, competitor rates via rate shoppers, etc.) to forecast demand and then optimise prices. Duetto pairs GameChanger with ScoreBoard, its business intelligence and forecasting module, and with other tools like BlockBuster (for group revenue optimisation) to provide a comprehensive revenue strategy platform.

Recent innovations: In early 2023, Duetto announced a new enhancement called Advance – Dynamic Optimisation. Essentially an extension of Open Pricing into true real-time pricing. Duetto Advance uses streaming analytics and third-party data updates as frequent as every 20 minutes to adjust rates continuously. This is an explicit move toward real-time yield management: instead of revenue managers having to accept recommendations daily or manually, the system auto-publishes rate changes in response to live market shifts. As Duetto’s Chief Product Officer explained, Dynamic Optimisation combines real-time market data with automation so the RMS “continually flexes prices in line with changing market demand, up or down, delivering true real-time rate optimisation”. In other words, if bookings suddenly spike at 4pm, by 4:20pm the system may have already inched prices up, without human intervention, to capitalise on the demand.

NH Hotel Group in Europe deployed Duetto across nearly 350 properties, using GameChanger for pricing and ScoreBoard for forecasting, as part of a digital transformation to centralise revenue management. By 2025, Duetto’s platform has expanded into total profit optimisation (including non-rooms revenue) and even loyalty pricing, but at its heart is the ML-driven Open Pricing that continuously recalibrates rates for each guest segment to maximise revenue.

FLYR (Formerly Pace Revenue)

FLYR for Hospitality has been at the forefront of fully automated, real-time pricing optimisation. Pace was a London-based startup (founded 2017) that believed revenue management is a science that can be automated. The company was acquired by FLYR Labs in 2022, joining forces with an AI firm known for airline revenue optimisation. Now under the FLYR brand, the hotel RMS solution continues Pace’s mission of “Always the Right Price.” It uses advanced machine learning to make pricing decisions on an hourly cadence with minimal human intervention.

One of Pace’s hallmark features was Autopilot mode. In Autopilot, the RMS doesn’t just recommend rates, it automatically updates them within the connected systems as frequently as every hour. This effectively outsources the day-to-day rate changes to the algorithm entirely. A Pace press release noted that by moving to full automation and hourly updates, hotels saw impressive gains, and it removed the manual workload for revenue managers. Pace also reported an average increase of ~10% in revenues for hotels using its automated pricing engine. Although it must be said that this would likely be for hotels with no previous revenue management in place.

The granularity of FLYR is also noteworthy. It’s not just a single rate for a night; it can yield by room type, by length of stay, etc., every hour. FLYR/Pace’s algorithm also continuously learns by evaluating price sensitivity: it uses reinforcement learning-like feedback loops (observing how demand responds to each price change) to refine its strategies over time. This approach results in micro-targeted pricing that can adjust to market conditions dynamically throughout the day. In addition to pricing, FLYR’s hospitality platform includes modules for inventory controls (like overbooking and stay restrictions), group business optimisation, and business intelligence. It also offers a Revenue Insights BI tool that aggregates PMS, RMS, and market data into real-time dashboards for analysis of pickup, channel mix, etc. All these pieces are built with a modern, cloud-native tech stack enabling that real-time flow of data.

Recent developments: The core Autopilot concept remains a differentiator for FLYR, even some other RMS require user approval for rate changes, whereas FLYR is comfortable running fully autonomous (with the option for users to set guardrails). For hotels struggling with revenue manager staffing or aiming to centralise with smaller teams, this full automation is compelling.

European adoption of Pace/FLYR has been notable among innovative independents and smaller chains. For instance, London’s Zetter Hotel and Prague’s Hotel Julian were early Pace users, and more recently FLYR has been making inroads with larger brands. The key positioning is that AI is at the core, not an add-on. FLYR’s RMS was built from scratch around machine learning, which means it doesn’t rely on rules or manual configuration as much as older systems.

IDeaS

IDeaS is one of the most established players in hotel revenue management, and they’ve been leveraging machine learning techniques for decades in their products. IDeaS G3 RMS, their flagship system, uses an automated ML-based revenue engine to produce what they call “scientific pricing decisions” at a very granular level. In practice, IDeaS generates forecasts for each room type, segment, and day and recommends optimised pricing and inventory controls (restrictions, overbooking levels) accordingly. The system considers both the hotel’s internal data and forward-looking market data in its AI forecasts. For example, it will ingest competitor rates, market occupancy indices, and even web shopping data to adjust its demand predictions. By combining these inputs, IDeaS’ machine learning models can yield highly accurate forecasts and pricing that often outperform manual methods – reducing the need for users to do repetitive tasks or rely on simplistic rules.

One of IDeaS’ strengths is automation at scale. Large hotel portfolios use G3 RMS to centrally manage pricing for hundreds of properties, trusting the ML algorithms to continuously recalibrate. The system automatically updates rates and availability controls, freeing up revenue managers to focus on strategy. E.g., it might decide to offer a certain room type at €200 for a two-night stay but €220 for a one-night stay on the same date, because the model found the longer stay is slightly less price-sensitive. These optimisations happen through ML algorithms that solve complex mathematical models (IDeaS has its roots in operations research). IDeaS was among the first to introduce the concept of optimal business mix: forecasting unconstrained demand by segment, like Duetto, and letting the system decide which business to accept or reject to maximise total revenue.

Recent innovations: Between 2022 and 2025, IDeaS has expanded its platform to address total revenue management and more flexible use cases. A notable addition is IDeaS SmartSpace, which targets Meetings & Events revenue optimisation. This is significant because for many large hotels and resorts, meeting room rentals and events are a huge revenue stream (sometimes equal to the rooms department). SmartSpace’s analytics can, for example, show that a ballroom booked for a low-rate local event is displacing a more profitable corporate conference. By providing visualisations of pace and demand for event space, and even recommending ideal pricing or whether to accept a piece of group business, IDeaS is extending machine learning beyond guest rooms into the realm of group sales and catering. Radisson Hotel Group was an early adopter of SmartSpace to drive meetings revenue growth.

In 2023, IDeaS launched new capabilities to optimally price "independent products”; think ancillaries or upsell products. For instance, a hotel could use IDeaS to dynamically price a breakfast package or a parking fee depending on demand (perhaps lowering the price of add-ons during low occupancy periods to spur uptake, and raising them when the hotel is near full). This shows how the ML engine is being applied to more than just rooms: it’s looking at all revenue opportunities.

IDeaS continues to be a popular choice across regions. Park Plaza Hotels Europe (with properties in UK, Netherlands, and Italy) selected IDeaS for 17 hotels in 2023, citing its ability to maximise revenue and profit across both rooms and M&E space.

Benefits and Limitations of ML-Driven RMS

It’s clear that machine learning has a great impact on hospitality, what does that mean for hotel organisations in practice?

Key Strengths and Advantages

• Automation & Efficiency: An AI-driven RMS automates countless decisions that used to eat up revenue managers’ time. Price updates, forecast revisions, market analysis; all this can happen in the background continuously. This automation not only saves labor but also allows constant availability. In a world where we continue to face staffing challenges and need to do more with less, this efficiency is a huge win.

• Accuracy & Better Decisions: Machine learning models, when fed with rich data, can detect patterns and correlations that humans might miss. They evaluate millions of data points to come up with the best decision. Hotels using ML-based RMS frequently report improved forecast accuracy and a lift in revenue metrics once the system calibrates. They’re also great at iterative improvement: if the model’s prediction was off, it learns from that error to do better next time. Over the long run, this leads to increasingly precise revenue management.

• Speed of Reaction: Perhaps the biggest practical benefit is speed. An ML-driven RMS can respond instantly to changes in demand or market conditions. No waiting for the next day’s pick-up report, it can reprice within minutes or hours. This real-time agility means revenue opportunities aren’t missed. If a competitor sells out and demand shifts to your hotel, a good RMS will catch that and capitalise the same day, maximising your rates. Likewise, if a sudden cancellation wave hits, the system can quickly adjust forecasts and recommend new promotions before too much perishable inventory goes unsold.

• Data-Driven Insight & Strategy: Even beyond the direct revenue metrics, ML systems provide a wealth of insights. They surface data relationships, like how weather affects your bookings, or which customer segment really drives peak nights, that help inform broader strategy. Many modern RMS platforms come with visual dashboards and reports that leverage the ML results, giving teams near real-time visibility into performance. This improves collaboration across commercial teams, as everyone works off the same analytically grounded picture.

  
  

Limitations and the Human Touch

For all the strengths of AI in revenue management, we can't fire our revenue team just yet. Revenue managers will always remain a key part of your commercial team.

• Data Quality and Availability: Machine learning is only as good as the data that feeds it. We often say “garbage in, garbage out”, and it remains true to this day. If a hotel’s data is incomplete, inaccurate, or biased, the RMS’s recommendations will suffer. For example, if your PMS history has not recorded cancellations properly or segments are all lumped together, the model might learn the wrong patterns. Ensuring clean, reliable data is critical. This means consolidating systems, maintaining data hygiene, and feeding the RMS all relevant information (on books, events, competitor rates, etc.). Hotels often need to invest effort up front in mapping and aligning data sources with the RMS. Another consideration is that if market conditions change in ways you have no data for, the model can also be caught blind. E.g., early in the COVID-19 pandemic, many RMS had to be re-informed with new data because historical patterns broke down. We still see the effects of this in current pricing suggestions.

• Transparency and Trust: By nature, sophisticated ML algorithms (like neural networks or ensemble models) can be complex and behave like a “black box.” They might output a recommendation without an easy, human-readable explanation of why. This can pose challenges for revenue managers who need to trust the system and for leaders who need to justify decisions to owners or asset managers. This is why human oversight remains important: an experienced revenue manager can interpret and validate the AI’s output. If something looks off, say the price recommended seems too high given guest sentiment, the revenue manager can question it and investigate.

• Handling Unusual Scenarios: Machine learning excels at recognising patterns it has seen before. But truly novel situations require human judgment. For instance, AI might not fully grasp one-time events like a city hosting a political summit (if it has no precedent data) or sudden shifts in traveler sentiment due to, say, recent tariff announcements. In such cases, revenue managers must step in to apply context that the data doesn’t capture. A neural network doesn’t have common sense or creativity; it can’t anticipate completely new demand drivers or “think outside the data”. Humans are needed to adjust strategies during black swan events (think the 2020 pandemic). Moreover, AI lacks emotional intelligence and broader business perspective, it might recommend something profitable that a human would veto for guest experience reasons. For example, an algorithm might suggest overbooking to a very high level because mathematically it maximises revenue, but we all know to avoid this to enhance guest experience and retain reputation.

  
  

Embrace AI, It’s Here to Stay

The key message for hospitality leaders is: AI is already here in hotels, often just under the gentler label of “machine learning.” If you have a modern RMS, you’re likely already benefitting from AI algorithms every day, whether it’s forecasting next month’s occupancy or tweaking tomorrow’s rate by room type. Rather than fearing it or thinking of AI as a futuristic experiment, leading hotels are leaning in and asking, “How can we get the most value out of these systems right now?” They recognise that an RMS powered by machine learning is a tool to sharpen their competitive edge, not a replacement for their revenue team. Forward-thinking revenue leaders are upskilling their teams to work alongside AI, feeding the systems better data, and refining strategies based on the richer insights these tools provide. They are also pushing vendors for continuous improvements and staying updated on new features (as we saw with real-time optimisation, autopilot modes, etc., rolling out in recent years).

In the end, AI in revenue management is about augmentation, not replacement. As one hospitality tech leader noted, the optimal path forward is “AI working hand in hand with human knowledge and intuition”, where the two together achieve better results than either could alone. Hotels that embrace this approach will be better equipped to make informed, profit-maximising decisions in an ever more complex market.