PredictX's Choe discusses:
- 'Furnishing houses' for AI in corporate travel
- The importance of prompt engineering
- Building a case for AI investment
PredictX emerged as the winner of this year's Business Travel Innovation Faceoff at Business Travel Show Europe, impressing the judging panel with its newly launched AI workspace Cogent. During his presentation, PredictX CEO Keesup Choe highlighted several use cases from early adopters, including a performance and savings manager, an air contract manager and tool, called Rubicon, that can create, evaluate and manage requests for proposal with AI. Choe spoke with BTN executive editor Michael B. Baker during the recent Global Business Travel Association Convention about Cogent's potential and how travel buyers should be preparing as AI power and capabilities increase. An edited transcript follows.
BTN: Where does Cogent fit in the marketplace?
Keesup Choe: Cogent is not just an AI application, because there are so many of them here [at GBTA]: there's a chatbot, there's a booking agent, things like that. We built all of those things in the beginning, and then Cogent is an entire platform, an agentic platform where you can create these agents and deploy them on their own or as an aid to an application. It took a long time to get there, a lot of successes as a result of failing as much as possible. We're rolling it out to everybody, and every one of our applications are enabled.
BTN: What key use cases are emerging?
Choe: The most valuable ones are the ones that are autonomously doing things, like audits, where we have AI agents identifying fraud and so forth. People are in different phases of the learning cycle, so they're most comfortable with things that look like ChatGPT, that they can talk to, but instead of generic, they can talk about their own program, and it will reply back with all the insights you wish you could get from an expert.
BTN: Is the idea, as you've said, to start looking at AI not just to do tasks we don't want to do but also tasks that no human beings could do?
Choe: Or things you could never afford, to hire 100 people to do this. We have the equivalent of probably 1,000 people researching companies every day. There's no way you could do that manually. It will take some examples. All human beings like to buy houses that are furnished, so we need to furnish a few houses so they can see how they live in it.
BTN: How can travel buyers keep their skill sets up to date as AI technology advances?
Choe: They need to have a tool they can actually use for their work. Microsoft Copilot, which helps you with Word documents or Excel, it's useful, but it doesn’t help you directly with operations. We have them, but I don't think there are enough of them yet. We've been working on this for a long time. Our error rates were pretty bad in the beginning, and now, they're really, really good. Depending on where they are, they will probably have some issues with errors. It's not quite mature enough yet. Even the biggest [travel management companies,] they're not going to lure an AI scientist [from the tech industry]. So, how do you get beyond that? What we did is you work at it for a long time. We have AI scientists, but obviously, we don't have the caliber like at Google or Meta. The other part is to teach the sector prompt engineering. I ask [when teaching], "How many of you can speak Spanish, just a little bit?" Just because you can say, "¿Dónde está la playa?" doesn't mean you are comfortable transacting business in Spanish. Just because you can use ChatGPT a little bit, you don’t know enough to actually transact business with an AI tool. To do so, you need to learn prompt engineering, and that's what I've been teaching. The more they feel comfortable, and they know how to communicate with AI, and the tools get better and more germane to their sector, then you'll get more adoption.
BTN: What's the best way to learn? Trial and error?
Choe: Do you say that if you go to Spain, you're going to do trial and error? No. Either you get immersed, which is hard to do, or you learn. You can attend one of my classes, download our e-books on prompt engineering. When I teach them, I'm hoping they walk away [thinking] it doesn't seem that hard, then they try to learn it. I'm not going to make them experts in a 30-minute session, but I have done intensive classes. If you go from "¿Dónde está el baño" to legal documents, you'll get frustrated, so you need some training.
BTN: Will it be a challenge for us to maintain creativity as AI advances?
Choe: It's changing so fast that it's hard to predict. The performance of the models are doubling every 10 months. It is the fastest growth of technology, of any technology, you can think of. What does that mean? At this pace, in five and a half years, it will be 100 times more applicable than today, and today it's pretty cool.
BTN: Will the energy consumption required for AI be a concern?
Choe: There are two separate energy [considerations]. One is the training, that's where the energy intensive thing is, and that's why Microsoft has bought a power plant. I think the U.S. should be concerned that China has almost four times their energy capability today. The inference part, as a user, is very efficient. There have been a lot of studies about every time you use ChatGPT, you're wasting this much water. That's completely skewed research, because in data centers, there's a fraction of the data centers which are closed water [meaning they reuse/recycle water rather than continuously requiring new water for consumption], but the legacy ones are open. Now all the big players are trying to get rid of the legacy ones, where 90 percent of the waste is coming from. Hopefully, within five years, there will be none of them left. Also, [graphics processing units], which AI uses, all of the latest ones are liquid cooled, which is 100 times more efficient. Otherwise, you have to cool the atmosphere of the data center. So, there's a lot of subtlety and detail in a scare piece you're not going to hear. The training is very intensive, and that's where the power goes.
BTN: We've heard some buyers say it's a challenge to convince their companies to invest in data and AI for travel. What's your advice?
Choe: A lot of companies have set aside budgets for AI transformation. You can tie that initiative together with a corporate aim. There's going to be funding already allocated, so it's getting a piece of that. The funding is sometimes hundreds of millions of dollars, so even if you can get 1 percent, it's well enough to cover it. Then, you want to tie it into a function. Like Rubicon, it was a no-brainer. A 10 percent improvement of the prices paid as a result of the RFP translates to $3 billion by savings. They do about $30 billion of sourcing, so it becomes a no-brainer. With the other example, air contract management, if you can extract an additional set of discounts, that's quantifiable.
The Microsoft Copilot, if you had asked people two years, "Would you like an AI alongside writing a document," I'm not sure it would have sold, so maybe things are changing. Think of this as a co-pilot for my travel. We definitely have clients who are reducing their spend on contracting. A lot of travel managers have [embeds from TMCs], and we're seeing that they don't need them anymore.