Alvarez Arguelles Hoteles

Maximo Pisa

Yogi Berra, the legendary baseball player, once remarked, “It’s tough to make predictions, especially about the future”. I still remember my days at university, when we were trained to build solid, theory-driven models grounded in a strict set of classical assumptions—linearity in parameters, homoskedasticity, normality of errors, and so on—to chase the “best” estimators and the most robust predictions, often with a quiet sense of achievement when our sample size (n) reached the magical threshold of thirty observations, just enough to make those assumptions defensible!

Nearly two decades later, the rise of big data has rendered much of that practice obsolete or at least profoundly altered. Large-sample asymptotics and flexible machine-learning methods now dominate the analytical landscape. Yet this shift has not necessarily made analysis easier, nor outcomes more accurate; it has simply replaced one set of challenges with another.

Getting the most from our data

Accurate demand forecasting in hospitality is key for revenue management and related decision-making, such as supply chain management. The industry operates under constant uncertainty, shaped by seasonality, post-pandemic behavioral shifts, weather volatility and macroeconomic forces —such as FX movements. On the other hand, inventory is relentlessly perishable: an unsold room is revenue lost forever. What is more, demand also extends beyond room occupancy into restaurants, cafes, events and ancillary services, while data remains fragmented across different systems (PMS, CRS, OTAs, ERP) rarely aligned and designed with forecasting in mind. Together, these factors make hospitality a fundamentally messy forecasting environment, where complexity is not an exception but the norm.

“Real-world demand does not unfold along a straight line, and forecasts are more useful when they describe a set of plausible futures rather than a point estimate.”

In this context, extracting maximum value from data becomes critical. After testing a wide range of models from autoregressive and pooled panel-data approaches to transformer architectures and long short-term memory networks—we found that one factor consistently dominated forecast accuracy: access to booking data. Historical booking curves, clustered through a machine-learning (ML) approach, proved especially powerful when modeled in an autoregressive framework and enriched with additive components and forward-looking calendar effects, such as holidays and festive periods (for those interested, see Viverit et al. 2023).

Looking back, this realisation echoed an earlier chapter of my career. In the UK optical retail sector, where I previously worked, customer appointments played an equally foundational role in demand forecasting. The lesson is obvious, yet easily overlooked: in complex, noisy environments, attention is often drawn to ever more sophisticated algorithms, while the most meaningful signals sit quietly at the heart of the business. Real progress comes from understanding the business well enough to see what has been in front of you all along.

Back to best practices

Always start with the fundamentals. Before investing in AI, put your data house in order by fully digitalising end-to-end processes. Planning often relies on unspoken expert knowledge, making it essential to surface, codify and structure this intelligence so ML initiatives can deliver meaningful, scalable outcomes.

In complex demand environments, such as hospitality, forecasting may also fail when it pretends to be precise. The instinct to produce a single “best” number is understandable, but it´s misleading as well. Real-world demand does not unfold along a straight line, and forecasts are more useful when they describe a set of plausible futures rather than a point estimate. Thinking in scenarios—supported by confidence ranges instead of fixed values—forces organisations to confront uncertainty honestly and to plan for it, rather than be surprised by it.

Consider a city hotel facing a major international conference. A point forecast may suggest healthy occupancy, but a scenario-based forecast tells a more actionable story: one scenario reflects strong early bookings and aggressive pricing; another accounts for delayed confirmations and higher cancellation risk if exchange rates shift. Revenue management, operations and staffing can then align their decisions around risk—not hope.

This shift in perspective points to a broader truth about forecasting itself. At its best, a forecast earns strategic value not by predicting a single outcome, but by inviting exploration. What happens if prices change? If weather turns unexpectedly? If exchange rates shift or demand surges? By embedding systematic what-if analysis into the forecasting process, organisations move from passive prediction to active learning. The forecast stops being an answer and becomes a tool for thinking—one that helps leaders navigate uncertainty with clarity rather than confidence alone.

This is where modern forecasting must not be understood as a contest between humans and machines, but an ongoing conversation between the two. Machines excel at processing booking curves, historical patterns and calendar effects at scale; humans contribute context—market knowledge, operational constraints and strategic intent. The objective is not to replace decision-makers with algorithms, but to empower them. Because AI does not replace people; people who use AI replace those who do not.

Ultimately, perfecting demand forecasting and building AI capabilities starts with people, not technology. Invest in continuous learning for yourself and your team and approach online narratives with healthy skepticism (including this article!). Understanding what’s achievable today—and what isn’t—helps organisations prepare for the skills and role transformations that AI-driven ways of working will demand. My knowledge and tools bear little resemblance to those I learned as a young college student twenty years ago. Yet the lesson remains timeless: staying up to date, remaining curious and never forgetting the fundamentals, are what truly equip us to face (and forecast) the challenges ahead.