The Problem with Forecasting in the Hospitality Industry

The hospitality industry has shown promising signs of recovery. Overall industry trends, however, aren’t meaningful for hotels managing operations on a daily basis. Robust forecasting at a very granular level is how hotels can boost their efficiency and help mitigate against so much uncertainty. This hotel forecasting framework shows how to develop those granular forecasts and why they are useful.

Great news for folks in the hotel industry, the market has seen strong recovery and occupancy rates today are effectively the same levels they were at pre pandemic. And so this is really nice for the industry as a hotel and for you to make strategic decisions.

But there’s a real problem with looking at the data like this. See, this is the occupancy rate at the US average level, but for your hotel or your chain of hotels, you probably don’t see a nice smooth line like this. Maybe on a Tuesday, you’ve got 30% occupancy and then the next day it bumps up to 50%. But then on the following day, we were down to 10% and then at the weekend, we’re up over 90% and you might see these day to day and week to week fluctuations, end up looking a little more like this with volatility in your hotel occupancy.

So when you’re trying to manage your hotel or your chain of hotels, and you’re really interested in operations and operating efficiently, this green line is not super important. Even if you can forecast it with really high accuracy, like a lot of economists and industry experts are able to do what would be really great for you is to be able to forecast this blue line, even with some error, to know what your occupancy rate is going to look like on a day to day basis.

Local Models

At Elder Research, we specialize in building forecast at very large scale in situations just like this. And when we think of this type of forecasting problem, sort of two different ways we think about building these models first is local models. What I mean by a local model is say, you’ve got a hundred hotels. Well, we’ve got a model for each one of those hotels.

That’s going to forecast the occupancy at that hotel. And we can even get much more granular than that. Maybe the occupancy is important, but perhaps you want to know what types of guests are gonna be staying. Are they going to be eating or drinking or using other hotel facilities or booking other activities through the hotel? Well, these local models, like an ARIMA model or a  Prophet model or traditional time series forecast, they can take in an input and they can produce a forecast for each of these hotels that you’re concerned with. And that can be really useful, but I want to compare that to a global model.

Global Models

What a global model does is build one single model that takes input from all of your guests at all of your hotels and, and builds one single model to forecast out at each one of those hotels. So you still get your individual forecast for each hotel, but the problem with the local model is there might be some signal that’s really small at just one series at one hotel. But when we combine information from lots of different hotels, we’re able to tease out that signal and the impacts that some hotels have on other hotels, these models might be something like an LSTM, which is a type of neural network or a hierarchical model that will build models at a low level, and then aggregate them up for your entire demand.

Exogenous Variables

These models don’t just have to rely on historic occupancy at the hotels. You can certainly include other things that really drive your occupancy rates. These we like to call exogenous variables that can feed into the model. So these might be like events and activities. Maybe you’ve got a world series baseball game coming up, that’s gonna drive your occupancy a little bit higher. We can also include things like discounts and promos in general pricing information on the hotel. We certainly know that those impact the occupancy rate at your hotel on a daily basis.

Forecasting in Action

So I’d like to tell a quick story. We had a client who staffed their own call center. So when any customer would call in with some type of issue, somebody would be on the other end to answer the phone and help them through that issue. Well, too often, the company found that they had a call center full of people and no phones ranging, but they still had to pay the wages for all the people there. And the converse was also often true. They had only a few people in the call center when call volume was very high and folks had to wait on the phone for very long periods of time. So with a very granular level of forecast at the hourly level, to be able to predict what that call volume was like and be able to predict some metric for how confident that forecast is. They’re able to staff their call center much more efficiently.

You probably don’t need your hotel occupancy at the hourly level. But if you had it, if you had a forecast that was at the daily level, it would allow you to run much more efficient operations with the way you staff and stock your hotel. So if this seems promising or interesting to you, please click this link and you can learn about this call center case study and how Elder Research could help you build a forecast that can be useful for hotels.