Corporate travel buyers and suppliers have been at the same impasse for a long time. In short, how do they negotiate pricing and agreements and still accomplish goals? The situation presents problems on both sides. Buyers need to ensure availability and service levels for their travelers at good rates; suppliers need to protect inventory to ensure goods go to the highest bidders and that the companies¡ªand shareholders¡ªmake money. We¡¯re not just talking about corporate business, here; meetings, groups and leisure travel factor into the equation, too.

Corporate relationships matter. But, especially in revenue management circles, they don¡¯t necessarily translate into optimized quarterly performance. In the most difficult circumstances, promises made on the sales side can¡¯t be fulfilled on the revenue management side and may not be communicated at all. It leaves buyers and suppliers in a tough place. Could artificial intelligence change this most fundamental issue? Well ¡­ maybe.

Supply Side: Building Better Pricing Models

Revenue management, at its most basic level, adjusts prices based on demand, suppliers¡¯ current booked business and competitors¡¯ prices in the market. The complexities of the practice have led to the proliferation of fare classes in the airline business¡ªfar beyond first, business and economy; any given cabin has multiple seat ¡°types¡± that look identical to any passenger. Only a certain number of seats are available at the lowest fare class, and they come with the most restrictions. As those are filled, prices rise until close to the departure date, when in some cases the demand breaks and seats are released to lower fare classes to drive final sales. Revenue management strategies most likely are to thank for that great airfare you found at 8 o¡¯clock Sunday morning but that you lost by 4:30 that afternoon after confirming plans with a friend.

Airlines sharpened the practice, but the hotel industry has picked it up with fervor, carving out a percentage of rooms and types eligible for the lowest corporate rates and reserving other blocks for higher-paying customers like groups that will drop some coin on food-and-beverage events. What percentage to reserve versus how much to contract and how to price it all has been described as more art than science. As more data science enters the picture, however, that may be changing.?

Just over a year ago, Best Western Hotels & Resorts launched new revenue management technology, BestRev, that crawls markets to search for pricing benchmarks 24 hours a day and weighs those benchmarks against current demand and booked business. The hotel chain allows certain hotels to set the system on autopilot, which means it makes changes to best available rates based on its own calculations and without human intervention. Hotels not set to autopilot get the same recommendations, but an employee had to approve the rate changes before they were reflected in the market. The initiative has now delivered a full year of results. ¡°Ending 2017, the hotels that gave the system authority [to change rates automatically] saw a 3.1 percent incremental increase in revenue compared to those that did not,¡± said Best Western SVP and COO Ron Pohl, adding that the machines continue to learn from themselves and that the algorithms improve with human tweaks. ¡°We continue to evolve this, and revenue management is one of those things where computers, over time, can really determine the best way to sell hotels.¡±

While Best Western has seen good results from its first foray into machine learning and revenue management, American Airlines director of revenue management and operations research Marcial Lapp isn¡¯t so sure that price determination is the right way to use AI. ¡°We create on the order of a half billion forecasts every night for all different flights and all different subsegments of those flights,¡± he said. ¡°We build some pretty sophisticated algorithms in there. However, they are not what I would call auto-adapting, if that¡¯s a way to represent machine learning technology. These algorithms need feedback, and that¡¯s where it gets tricky because it¡¯s not always clear if certain pricing, per se, was good or bad.¡±

Airlines are not selling widgets, Lapp continued, and this applies to hotels, as well, especially in high-demand business markets. ¡°I¡¯m not necessarily looking to increase conversions by 10 percent or whatever. The problem I have is that I have customers wanting to buy a ticket and I have to make sure I can say, ¡®Hey, customer, I¡¯m not going to sell you this seat because I¡¯ve got another person lined up and ready to go who¡¯s willing to pay more.¡¯ These algorithms aren¡¯t traditionally designed to do that. I¡¯m not saying we won¡¯t ever get there, but we¡¯re not there now.¡± Lapp said he needs human analysts to provide market context for events like the Boston Marathon that ¡°might look like a blip to a machine¡± but that are very meaningful to price forecasting.

Yet Noodle.ai CEO Stephen Pratt said that is exactly where AI-enhanced revenue management for airlines and hotels is going. ¡°The amount of data [that hotels and airlines are analyzing] with learning algorithms to help set prices is very small,¡± he said. If he were a revenue manager, he¡¯d want AI to quantify the effect that weather has on hotel demand or that other events are having in the area. Not to mention macroeconomic conditions, consumer sentiment and what customers are saying about the brand online. ¡°What¡¯s the tone that¡¯s set? What are my competitors doing? What specials or deals do they have?¡±

These types of data sets, Pratt said, can fuel the AI fire when it comes to pricing. Plus, he said, ¡°There¡¯s a whole new world of analytics, of mathematics that are now possible to be applied to revenue management [with] breakthroughs [in] super computing power.¡±

Buy Side: Staying Competitive

Pratt¡¯s soothsaying probably sounds scary to corporate travel buyers struggling to manage just their traditional T&E data. ¡°It would be an interesting exercise to take a stopwatch and track what activities a travel manager actually spends time doing,¡± said Keesup Choe, CEO of business intelligence tool PredictX. ¡°They spend a huge amount of time chasing after data, checking data, correcting data, generating reports, sending reports, correcting and resending that report ¡­ and I bet none of their business cards say ¡®data collector¡¯ or ¡®data generator.¡¯¡±

Yet at the same time that buyers are chasing the data they can get, they are keenly aware of how much they are missing, especially when they sit down to negotiate with suppliers. ¡°Every organization that goes to the negotiating table has been reliant on supplier data,¡± said Advito senior director of intelligence and analytics Lexi Honohan, who was formerly a travel buyer for Swiss bank UBS. ¡°It would be great to have conversations [with suppliers] where the buyer could say, ¡®You canceled your flights in that market 15 percent of the time,¡¯ or, ¡®Our travelers know that your airline cancels every time there is a weather issue,¡¯ or with a hotel supplier, ¡®Our rates are not available 70 percent of the time.¡¯¡±

We¡¯re getting there, slowly, according to a handful of travel business intelligence executives. The emerging best practice for corporate travel data, said Choe, is the smart data lake. ¡°You just bring in every single piece of data: pre-trip, post-trip, social media or anything. Then you can create models that generate alerts, information based on that data without having to be bogged down at that database level.¡± Big companies like UBS?and Microsoft are building on this concept for travel already, but it doesn¡¯t have to be out of reach for the masses.

Honohan has built, on a Domo platform, a business intelligence tool and dedicated advisory practice at Advito that she calls ¡°user friendly and economical.¡± In addition to traditional travel management company, credit card and expense data, it digests data on market rates and data from FlightStats, QSI, traveler sentiment and other sources. It also can integrate with other systems in the organization like sales, HR and finance.

Data Visualization Intelligence president Brian Beard has been democratizing machine learning-enhanced business intelligence for travel management, and he feels the concept gaining traction as travel managers start to see what is possible, particularly for predictive volume and pricing. ¡°One of the most intriguing areas for us is to look at how much it costs, for example, to go from New York or London or New York to Los Angeles with everything all in. We¡¯ve used machine learning on the historical data to create total trip cost with airfare, hotel, meals, ground transportation, etc.,¡± he said.?¡°Then, you start to look at economic data, development in the market, [airlift] capacity changes.¡± At that point, Beard said, buyers will be able to model the data to show suppliers ¡°for this airline or for that route or for that hotel chain: ¡®This is where our travel will be in the next year and possibly even the next two years.¡¯ Can you imagine going into negotiations with that kind of predictive data? It¡¯s becoming doable.¡±?

Conventional wisdom might hold that buyers armed with this level of data would make suppliers nervous. Not so, said Beard. ¡°When I talk to suppliers about it, at least today, they want buyers to come in with better information. They can set their revenue targets and contract targets at a more acceptable level and figure out terms that really make sense. I can¡¯t tell you how many companies set targets they can¡¯t possibly meet.¡± That sets a bad tone for everyone.

?Back to the Supply Side: Making Good on Promises

¡°The biggest problem right now is that the revenue management department and sales department in a traditional system cannot communicate,¡± said Marco Benvenuti, chief marketing officer and co-founder of cloud-based hotel revenue management technology provider Duetto. ¡°Even if you have data in the system, you have an inherent problem in the incentives between the two departments. You have to fix that alignment and get to a revenue strategy. You can¡¯t just look at revenue management. You have to get revenue, sales, marketing and analytics under one umbrella.¡±

That¡¯s going to take better data storage strategies to pull the pieces together internally, said Benvenuti, plus more advanced algorithms to find the data patterns that will shape the right strategies. If hotel companies can get that far and buyers can confidently model their future volumes, however, the horizon may open up for new agreements.

Buyers should get ready, though. Benvenuti said pricing agreements have to be dynamic; static pricing agreements won¡¯t work. ¡°The flexible pricing model is the last component if you have the data and the AI and the culture together. It will change the way we segment and manage. The travel manager gets rational pricing and access, and the hotel makes money¡ªnot necessarily by charging more but by charging the right prices at the right times. If we can¡¯t move to dynamic pricing, I can put the most advanced AI in the picture but the answer is nothing.¡±

Negotiating Beyond Pricing

The data disconnect is pretty much the same for airlines, according Lapp. ¡°It¡¯s not a lack of data. Airlines have always had a lot of data¡ªrevenue management data, loyalty data, marketing data, network data¡ªbut it¡¯s been sitting in siloes.¡± And he agreed with Benvenuti that bringing the data together is key. ¡°Integrating all this data to make better decisions is a goal. It requires a new platform, which we now have, but it also requires new skills, which we are learning.¡±

Where they differ is in pinpointing the opportunity. The pricing model for negotiated airfares already comes down to a percent off market rates¡ªi.e., it¡¯s already dynamic. There could be an opportunity to fine-tune it, but there¡¯s a point at which you are diving for pennies. Lapp is concentrating American Airlines¡¯ machine learning efforts on merchandising and personalization with ancillary offers.

¡°This relates to my previous point in that [ancillary purchases] are much more about increasing conversions,¡± he said. ¡°If you bought a fare on AA, for example ¡­ and you also bought priority access, I can use that in my algorithms to then say, ¡®Here are a bunch of other things you may be interested in. If you are, great. If not, it¡¯s still good because you as a human being have given me information that allows me to tune my algorithms based on what works on your segment and what doesn¡¯t.¡¯¡±

Translating that into the corporate world, he said, is challenging even if travel managers came to the table with amazing traveler segmentation data and behavior stats that could enhance agreements. The reason: distribution technology. Managed travelers generally are not booking direct, and therefore technology limitations restrict what can be done from a merchandising perspective. But in theory, this is the future, as standards like New Distribution Capability penetrate deeper into the industry.

Right now? We¡¯re not quite there yet. As Benvenuti said of the hotel side, the data is too dispersed on the airline side to apply a true AI strategy, Lapp said. ¡°I don¡¯t have the data in one place so that I can run these algorithms, so that¡¯s where we¡¯re focusing our energy right now: making sure that we have all the data in one place. And then we can figure out what actually matters for personalization. Machine learning and AI will be the icing on the cake.¡±?