Metasearch sites, such as Google, Kayak, TripAdvisor, and Trivago, have redefined the travel landscape. They offer travelers a simple way to specify preferred dates and desired amenities, as well as easily compare rates among competitors.
For hotel brands in particular, metasearch sites serve as a marketplace to compete for consumers at a property level, and advertising on them can drive significant web traffic and qualified leads. But managing the sheer amount of data involved in metasearch ad campaigns can be very challenging because of the vast number of online travel agencies (OTAs) operating in the space and the variety of real-time customer inputs metasearch enables. The optimal bid could be influenced by the user’s location, type of device, day of the week, time of day, proximity to the booking date, and many other factors.
Artificial intelligence offers a way to address that. Platforms such as DerbySoft — a company that provides distribution services and digital marketing services across the hospitality sector — make it possible to set up a campaign within minutes and let it run by relying on automation to optimize and make real-time bidding decisions.
SkiftX recently explored the challenges of metasearch and how machine learning can address them in an interview with Norberto Degara, head of data science, and Christopher Callow, head of product for marketing services at DerbySoft.
SkiftX: What are some of the obstacles hotels face in making the most of metasearch?
Christopher Callow: Metasearch primarily sells as an auction. How much you bid is determined by your return on investment target, or the balance between the increasing cost of the clicks you buy versus the number of bookings you hope to generate. The challenge is in understanding when to invest, weighing risk versus reward, and how high you can bid without falling on the wrong side of your return on investment target. This requires relatively complex analysis and predictive capabilities to do well.
Metasearch sites have sought to support advertisers by providing more flexibility and control in tailoring the right bid, but it’s a bit of a double-edged sword. While the greater flexibility does offer improvements in efficiency and performance, it also requires exponentially more time to analyze and apply successfully. This results in advertisers focusing their efforts on the top 10 percent of their hotel portfolio — on properties with large volumes of data — and gradually whittling away on the long tail, eventually turning off placements with low data.
SkiftX: How does the dominance of OTAs complicate matters for hotel brands?
Callow: OTAs will always try and win as much traffic share as possible. They typically have very large, dedicated, and well-funded teams that are completely focused on optimizing metasearch. They have relatively more flexible budgets which they can freely reallocate based on shifts in performance, across a broader range of hotels and markets.
Comparatively, hotel brands will often source funding from hotels or regional teams, locking budgets to a narrower range of opportunities, making them more susceptible to macro factors. These factors place hoteliers at a disadvantage when it comes to competing efficiently in metasearch, but most of these challenges can be overcome.
SkiftX: What would a hotelier be up against if they tried to tackle metasearch without machine learning?
Callow: The bids you submit can never be considered optimized or final since metasearch is a live-auction environment. They will always need to be tweaked in response to changing demand or shifting rate effectiveness, rate parity, pressing commercial needs, and the bids of your competitors in the auction. These adjustments need to be made constantly, but the amount of work in calculating the appropriate bid can be incredibly time consuming.
SkiftX: How can machine learning help hoteliers be more precise and efficient?
Norberto Degara: One of the biggest challenges is being able to predict the amount of revenue a particular opportunity can generate. That is determined by the characteristics of the auction around that distinct placement — such as which point of sale you’re competing for, the length of stay, day of the week, which hotel [the consumer] is looking for, and the competitiveness of the rate at that moment.
Each iteration of these attributes resolves in a different prediction. Machine learning is great for these situations — such as when calculating the answer requires processing large data sets, the environment is constantly fluctuating, and where a large amount of fine-tuning among options is required.
SkiftX: Can you give an example of this?
Degara: Many of our clients actually manage their budgets on a per-hotel level — but not all hotels and points-of-sale benefit from the same volume of data. That makes knowing how much to bid challenging. When our algorithm identifies that we don’t have enough data to make reliable predictions for a particular hotel, it will start grouping together hotels that behave in a similar way by looking at their common characteristics. We can then use that cluster’s data to inform a more refined prediction than would otherwise be possible. This is done dynamically and automatically, which greatly improves performance when managing hotel-funded programs at scale, or when bidding for newly opening hotels.
SkiftX: How does machine learning help drive cross-channel campaigns and cost allocations?
Callow: Clients will often be inclined to go with the opportunities they have the most experience with, which locks budget into familiar silos at the expense of other opportunities. By managing your budget with a channel-agnostic approach, with a focus on your commercial objectives, you can allow the algorithm to learn the unique mix of placements which is appropriate for every brief. This also allows the algorithm the freedom to dynamically adapt bids to shifting demand, market forces, and auction changes in line with that brief.
In addition, every channel has a slightly different approach to how they display ads and how they allow their advertisers to manage those ads. This results in placements behaving differently to the same bids — learning the characteristics of each channel, for each hotel, and bidding appropriately. This is key in maximizing the efficiency of cross-channel performance. Without an automated solution, finding the right mix of placements and maintaining that efficiency over the length of a campaign would be near impossible.
SkiftX: How is DerbySoft anticipating the evolving needs of hoteliers when it comes to metasearch?
Callow: The potential of metasearch is continually increasing with new multipliers and more granular geographic targeting being made available. We’re also seeing new ad types being introduced — complementing the established metasearch placements — which allow hoteliers to push their properties up the destination search results page and drive incremental upper funnel demand. This will lead to better ROI, both at the chain level and for individual properties — but only for those advertisers well-positioned to take advantage of the opportunity.
To ensure our clients are in that position, we’re spending most of our efforts on automation and machine learning. We’re making it possible for clients to focus on their commercial needs, so the only thing they need to input is their performance goal, budget, and campaign dates. Everything else is effectively automated by default — although you can get more granular as needed.