Hotels don’t always get their comp sets right. It’s a fact. Think about it: comp sets are the lens through which hotels see competitive performance. Pick the wrong competitors, and the lens gets blurred. That can have serious consequences for hotel performance analysis. This week we publish more research on the comp sets that hotels are currently using for competitive analysis.
Comp set selection need not be as subjective as it is in today’s lodging industry. Since the advent of the internet, we can know vastly more about the way that bookers select hotels from the alternatives available in a given location. Most bookers use some form of web-based booking engine to search hotels. Booking engines facilitate the comparison of hotels, allowing users to filter based on preferences: price, location, brand, hotel class, amenities, etc. In most cases these product attributes can be identified from hotel data. As a result, the substitutability of competing hotels can be measured, providing a statistical understanding of how directly hotels compete with one another.
We recently published an analysis of more than 2,800 current hotel comp sets. That analysis compared the hotels’ nominated comp sets to the results of Hotel Compete’s automated comp set selection process. The selection process uses current hotel data to identify the 100 potential competitors that are geographically closest to the reference property (i.e. the hotel whose comp set it is). It then ranks the 100 potential competitors according to how different they are from the reference hotel, based on major decision attributes like those listed above.
The model works by ascribing “variance” scores to each attribute listed above. So the more different a competitor is in terms of its location, hotel class, future selling rates, etc., the higher the variance score it gets for that attribute. When the variance scores are added together, the competitors can be ranked based how different they are from the reference property. The further down the list a potential competitor is, the less similar the hotel, and hence the less likely it is to be competing with the reference hotel for business.
The ranking list of potential competitors provides us with an excellent yard stick by which to measure the comp sets that hotels nominate for competitive benchmarking. If a hotel’s comp set is truly representative of its competition, it should be composed of competitors that rank highly on the list. To test this hypothesis, the comp set selector process was run for the sample of 2,833 hotels. The resulting comp sets and rankings were then compared to the hotels’ nominated comp sets.
The table below summarizes an initial analysis: the first table summarizes the rankings of the hotels that were included in the Hotels’ nominated comp sets, but were excluded by the automated comp set selector. The second summarizes the opposite: i.e. the rankings of the hotels included in the Hotel Compete set, but excluded from the hotel’s nominated comp sets.
The results are instructive. First, of the nominated hotel competitors that were excluded by the Hotel Compete process, more than 50% ranked 20 or above on the list. It is not surprising to find differences between a hotel-nominated comp set and the results of a statistical analysis. Comp set selection is not a clear-cut decision – in fact for many hotels several marginal properties could legitimately be added to or excluded from a comp set. But this analysis shows that when a hotel chooses to include a competitor that the statistical model would exclude, in more than half of cases there are at least 20 hotels that are more similar to the reference property than the one that they chose.
More interesting, though, is the analysis of hotels that are left out of hotel-nominated comp sets. More than half of the omitted properties occupy positions one through four on the ranking scale. Think about that: when the algorithm spots a gap in a hotel’s comp set, more often than not it is one of the four most direct competitors of that hotel.
There are significant inconsistencies for which there are numerous possible explanations. But they are inconsistencies nevertheless, and they point to an enormous analytical opportunity. And they will be analyzed in greater detail in a series of articles that will be published in the coming weeks. Watch this space!