The Great Hospitality AI Disrupter
Hotels use an exponentially increasing volume and an assortment of data to help predict the future and answer questions such as “Who’s coming and when?” “Are those the right people or groups? If not, how do they find the right people or groups?” “Which marketing levers are the right ones to pull to land specific customers?” “What technology needs to be in place to make these predictions?” and “How can they continuously improve the reliability of these predictions?”
The industry has been steadily adopting artificial intelligence (AI) solutions to help answer these questions, and many others, across all areas of operations. Building, configuring, deploying, and training AI systems are a huge challenge for any organization, but the cost-value proposition has been continually making it a worthwhile investment.
Model Degradation
Models are built and trained with the data at one’s disposal, whether it is a single set of data or multiple, related, sets. As with any AI system, once the model is put into production it immediately begins to degrade in accuracy. The data that trained the system isn’t the same data that keeps feeding it.
A common mistake made when implementing an AI system is assuming once it is in production all is well. What’s really required is continual monitoring of results, planning for changes, then making model adjustments accordingly. Additionally, there is the human equation, and ensuring the human parts of the process if they exist, are monitored to make sure the operational behavior matches the expected behavior.
Enter the Great Disruptor – The Pandemic
If normal data changes cause models to degrade in accuracy and need to be monitored, what happens when the system is thrown a curveball like COVID-19? Processes built with relatively stable trending are thrown into disarray. The effects of the pandemic on the industry changed everything. Data that made training AI models simpler now says something completely different, or worse there is no data at all.
This is not the first time hospitality patterns have been thrown curveballs and it won’t be the last. As an example, the global economic downturn in 2007-2009 completely disrupted industry patterns. This caused a major problem for the data science systems that were forecasting during that time and beyond.
One might think that, with the advances in AI since then, this would be a simpler problem to overcome. To some degree, that’s true. The AI systems of today are more powerful and more flexible than systems 10-15+ years ago. Even though they’re handling more data and often from many diverse sources, AI systems can be trained to handle different situations much better than the data science models of old. They are excellent at pattern recognition or logic, when provided with sufficient training, and can do it against volumes of data impossible for a human to handle.
What Should the Industry Do With COVID-19 Data and Its Patterns?
One school of thought would be to throw out the period and just declare it an anomaly. That worked reasonably well when dealing with data from 2007-2009. Compare post-pandemic periods to pre-pandemic periods and plod forward, right? That makes some big assumptions though: COVID-19 didn’t change things and data from the period isn’t relevant.
Keep an eye on emerging new patterns
Despite the obvious lack of occupancy and usage during the time, there are likely new patterns in data collected during the COVID-19 period. Remote work and technology boomed. Group events went virtual and then hybrid as things progressed. But will that behavior continue in the future and what will that do to the transient/group mix?
New patterns will be critical for your success, particularly from a group sales perspective. As time progresses, we will need to consider whether hotels will be able to count on group business to fill rooms, while keeping an eye on how business travel has changed. We also will need to understand if these changes are temporary or permanent.
Your sales teams will need data to understand if your property’s competitive set has changed. For instance, having insight into whether groups will still go to the same types of hotels used in the past. Even more important, your sales teams will need to know if groups are going to the same cities and using the same amount of event space. Having this type of data available as they begin outreach to build out your group pipeline will be tantamount to having the most recent revenue management data.
Model the COVID-19 data
Modeling the COVID-19 period data in isolation could present an opportunity to pick up the new patterns. The focus could even be on the later period to zoom in on potential new trends. If this data is too sparse to be of much use, then the overall system might be better off ignoring all of it. Otherwise, model it like any other data.
As travelers cautiously begin to return to normal travel patterns, the data collected over the past two years may prove interesting indeed. We have seen the shift in work patterns, with many workers choosing to continue to work from home. Some businesses have shuttered their offices in lieu of the new work-from-home/anywhere model.
There is even a question about whether the 9-5 workweek is dead. According to the U.S. Bureau of Labor Statistics, in 2019, the average full-time American worker spent at least 8.5 hours working every weekday. Inc. Magazine reported that while this was once the norm, the pandemic has shed light on the fallacies of this structure for efficient and meaningful work, and leaders are now looking to reinvent it permanently. The pandemic certainly brought to light that today’s employees want to be empowered to own their schedules.
Train models to mix the results
If there are new trends, and certainly there are, we will need to adjust models and processes to recognize them. It will be necessary to carefully monitor the inputs, outputs, and actual outcomes just like one would for normal model degradation. Mixing these two sets of data, along with the new data coming in, can help the essentially new model come up with more accurate forecasts.
It comes down to what data analysts have always done. Closely monitor the progress, compare actual results to predicted results, and adjustment accordingly. Then be patient as the data models progress.
The Future
With a disrupter as large as the COVID-19 pandemic, it’s going to take time to regain the industry’s ability to predict future business as accurately as before. Regardless of whether a hotel bases its forecasts on gut feelings, spreadsheets, data science, or modern AI, revisiting training of those models is going to be required. Learning to trust them for budgeting and forecasting is also going to take time until they’ve been proven out.
ABOUT THE AUTHOR
Rob Landon leads the software development and IT efforts at Knowland with a focus on enhancing the role of AI and machine learning to drive predictive analytics for group data. He brings over 25 years of software industry experience across a range of operations from innovative startups to large enterprises. For the past decade, he has focused on software solutions to help hoteliers maximize their revenue in both the transient and group spaces. Rob has a BS in computer science from Louisiana State University and holds an MBA from the Terry College of Business at the University of Georgia.