Data Co. Cerebri AI Dives Into Corporate Travel Arena

Jean Belanger

Cerebri AI’s Belanger discusses:

  • Launching a new travel solutions platform
  • Combining sources for accurate carbon data
  • Tailoring reporting to target audiences

Data science startup Cerebri AI in recent months has launched a new corporate travel solutions platform, which it says can track travel program costs, make trip cost predictions and provide analytics and predictions around companies’ environmental, social and governance needs. The Austin, Texas-based company has been around for a few years, providing customer experience data solutions for such companies as Ford, Verizon, Mercedes-Benz and Scotiabank, but following the Covid-19 pandemic, during which it “went through a near-death experience 1,000 times,” the company has turned to travel as a primary focus, co-founder and CEO Jean Belanger said. Belanger, who previously started up supply chain software company Reddwerks, spoke recently to BTN executive editor Michael B. Baker about his company’s approach to travel and ESG data and its progress in winning new customers. An edited transcript follows.

BTN: How did you land on travel as a focus?

Football Highlights

Jean Belanger: Coming out the of the pandemic, we had a decision to make: What vertical are we going to concentrate on? We had done auto, and we had done wireless, and we had done banking. 

We looked around to see who had really tough data problems, and where was growth? Travel has interesting, longstanding data issues, and we’re really high-end data engineering capable, and the data is the time series. What we do is we set up one traveler journey for each traveler, and we create one set of data for each traveler, no matter how many trips they’ve taken, so we have one source of truth for all the predictions we need to make and all the drill-down capabilities with regards to looking at travel cost by employee, by meeting, etc. 

We started in Q4 last year. We launched the product and showed it for the first time at [the Global Business Travel Association’s convention] in August, and now we’re out and about selling it. We have a different view on travel analytics than most. What we learned for the last 25 years, give or take, with enterprise is that they would rather have less applications. If you can consolidate different value propositions they have or need onto one platform, it’ll be better. We have travel, and we reconcile credit cards, expense reports, the [travel management company] and HR hierarchy.

BTN: And how did that eventually become a focus on ESG?

Belanger: We were doing a scale-up for PwC in London in Q1, and they asked if we could figure out geolocation of employees for withholding taxes. You have to do it quickly, near real-time, because you only have 10 days to file the taxes. If you work half a month in California at their more-than-modest state income tax levels, then come to Texas, where I live, where there isn’t any, then you have a different scenario. Someone asked about business traveler insurance, what countries you are in. If you can’t tell the insurance company that precisely, then they put a premium on your premium to make sure they cover any risk you haven’t disclosed. So, that’s part of ESG, governance. 

Then, someone asked, can you figure out the spend on minority-owned business that we do through the corporate credit card? And we said yeah, that’s in the vendor file from SAP or financials. So, one thing led to another, so ESG started to percolate to the fore. 

Then, the issue of emissions came up. We said, “That’s interesting, because it’s a prediction.” We’ve done probably 75 models across numerous verticals. So, we did a dive in that, and we’re going to have our carbon calculator available very soon, and it should be at least as accurate as the best that is out there, and I think we’re going to best others. We’re basically a science-based company. We have our data science team to solve the issues. We’ve delved into it pretty deeply. We’ve looked at aerospace engineering software to calculate fuel burn. So, that led to another problem. We said, “OK, you make a prediction, that’s great. You may want to do that, or at least give the information before a person makes a purchase decision, so pre-booking or pre-ticketing. But what happens if you cancel a flight?” About 5 percent of business flights get canceled, and 25 percent change, so you’re looking at 30 percent of travel. If you make a prediction when you book a ticket, it’s going to be wrong when you finish at the end of the year. 

So, we said, now we’re going to have to do two carbon calculations: the pre-ticket, pre-booking, and then a post-trip verification. How are we going to do that? Well, we already do the reconciliation – credit card, expense report, TMC—so, if I change the class of service, the flight and all of these things, when I finish the trip, I’ll be able to reconcile the data and do a final estimate and hopefully, if it works out on the post-trip verification, we’re going to use the actual distance, the route flown, not the great circle distance. There’s quite a discrepancy between the shortest distance between two airports relative to flying around weather and all other kinds of things.

One thing led to another. We probably got about eight modules now in the system. We didn’t start out thinking that way, just listening to the customers, and one thing led to another.

BTN: What are you aiming to solve?

Belanger: What we set out to do was to not only do AI but also to automate data engineering. We decided we’d fixate on time series and customers in particular—customer journeys and customer experience—and in order to do that, you had to get multiple data sources. Some were batch, some were streaming and others came in an [Internet of Things]. With mobile phone, there’s telemetry coming in off the phone, and that’s a big impact and interesting data from a customer use point of view, which is useful for prediction. 

We merged all of that into streaming data, which we used to build a platform. We didn’t want to do custom data engineering every time, so we prebuilt more than 30 transformers. As a simple example, if you’re taking data from stream No. 1 and it’s from Europe and stream No. 2 is from the United States, they don’t record dates the same way. So, a transformer will take the European date and transform it into a standard format. Transformer No. 2 will take the U.S. data and change it, and eventually that will merge, and we do all of this in a streaming environment. It’s very fast, we do all the [quality assurance] required. 

That creates a data set that is almost ready for AI, with one key step in the middle, which is creating engineering features. If you want to know what the average per-trip spend for all a traveler’s trips for the last six months, because that might be predictive on indicating what the next trip will cost, that’s an engineered feature, which is a fancy way of saying the data scientists do math and come up with a value. If you have 85,000 travelers, that’s 85,000 values, one per traveler, but every time the cost structure changes in their journey, it has to be recalculated. After you do that, then you have a model-ready data set. That’s what we set out to do for customers. 

What we do is we set up one traveler journey for each traveler, and we create one set of data for each traveler, no matter how many trips they’ve taken, so we have one source of truth for all the predictions we need to make.”

BTN: Where are you in terms of customers?

Belanger: We started out looking at enterprise, and when we added all these other things, [now] everyone wants to know the carbon emissions for their flights. We ended up at six or seven tiers in pricing. We think the utility of some of these value propositions is equally valid for SME as it is for enterprise. We have about 15 proposals in customers’ hands right now, so we expect we are going to be signing—we’ve had verbal commitments from several—them very shortly. We’re going to be actively involved with TMCs and other industry players, because we have to interject ourselves in certain points. We’re not going to wait until everything is done and then get the classic data feeds and do the reconciliation, because some of this requires us to intervene in different stages. We’ve been verbally accepted in the partner program from one of the big TMCs, so we’re going through that process now.

BTN: What functionality do you have in place?

Belanger: How do I use the software? How do I use the information provided? This is my second data science startup, so one of the problems we faced last time was we had a very sophisticated infrastructure, which is a way of saying you can report on nearly anything real-time. The only problem is that it’s hard-wired and it’s really hard to change the use. The CEO doesn’t need to see the same information as the travel manager. They have different needs. 

We decided we would do things a bit differently. We set up our [user experience], all through APIs, and we call it Answer the Question, ATQ. The reason we call it that is that I don’t have any patience. I just want the damn answer. So, we set it up that way so we have an inventory of pre-configured modules, we call them widgets, that you can easily put on the screen and configure a dashboard in one or two hours for a group of people, and an individual, and a CEO or CFO or a head of procurement. 

The widgets—some are graphs, some are pie charts, lists or tables—are all tied into the data backend, so the person who is building the dashboard doesn’t have to worry about where the data is coming from. The data comes in from all these multiple sources, and some of it is useful for modeling and making predictions, and some of it is just needed by the UX, but it all has to appear in the UX, and you’re not going to have to wait a half an hour for something to show up on the screen. We built a very high-speed backend to interface between the data and UX, which is easy to configure. You can have as busy a dashboard or as simple as you are required to do your job to the best of the capabilities. 

BTN: What are your data sources?

Belanger: Before we launched this, we decided that we were going to need some travel data experience and expertise. We hired several people who are very experienced in dealing with the travel data, who have incorporated almost all the TMCs, all the credit cards, all the expense reporting systems, especially Concur because it’s important in terms of the larger enterprises. One of the advantages in having invested heavily in the data engineering side is that when you get a new data source, it’s not nearly as daunting as it would otherwise have been. On the carbon emissions, for instance, we’re going to have to interact with multiple vendors who heretofore the travel industry has not seen that much—like, what’s the scheduled route versus actual route miles.

BTN: Do you expect travel will remain your primary focus in the years ahead?

Belanger: For the foreseeable future, if they don’t stop throwing problems at us, we’re not going to do anything else. …  In spite of all the aggravation, the system works pretty well. The question now is how to make it better. We don’t see anything else for the foreseeable future given the number of interesting problems we’ve been asked to look at.