Kambi’s Massive Bet on AI: Interview with COO Erik Lögdberg
How B2B sports gaming provider Kambi is taking on the sportsbook world with artificial intelligence.
As artificial intelligence and machine learning models have advanced, sports gaming companies are increasingly moving to automate their operations. Kambi Group Plc, a B2B provider of technology and services for consumer-facing sportsbooks, is moving aggressively in this area.
The company recently announced a new “Third Generation” AI approach. We sat down with Kambi COO Erik Lögdberg to talk about what this is and what his views are on the industry.
The new technology is the product of an ambitious plan to rethink the technology behind sports gaming. The idea is to make machine learning models that can not only run sportsbooks autonomously but also learn on their own, allowing the company to build new products extremely fast.
Kambi has set some ambitious long-term financial revenue targets, and this new AI technology is key to its growth plan and cost savings. Despite losing DraftKings and its expected loss of Barstool, Kambi has customers such as Bally’s, BetRivers and Parx. Other sportsbook operators are watching what Kambi is doing—as they are intensely competing and seeking to build, partner to access, or buy their own automation technology. Kambi is betting it can move faster than other operators, especially consumer-facing ones, by focusing on the underlying technology behind sportsbooks.
New approach
Kambi actually created its first algorithms to generate probabilities—then known as Unibet—in around 2005, says Lögdberg, who also joined Unibet in 2005. (Kambi was officially formed in 2010 as the B2B unit of Unibet and spun off in 2014.) That technology later became critical to Kambi offering in-play markets, he said. But the technology has advanced tremendously since then. It has gone from “machine assisted human trading to human assisted machine trading.”
The idea behind Kambi’s third generation system is to build underlying technology that can quickly adapt to new products and sports. Instead of building specific machine learning models for each sport, Kambi has built a model that can switch to different sports, Lögdberg says. Kambi announced the new third generation product for pre-game soccer earlier this year and now aims to roll out it out for in-play soccer this fall, he said. Then it will roll out to other sports.
“What if we could automate that and let that, over the years, more so evolve with the data (rather) than actually rebuilding bespoke stuff?” he said.
Previously, Kambi built separate models for different sports to understand specific aspects of a sport such as corner kicks or yellow cards or red cards. But that was time consuming and costly. So the new version can essentially learn new aspects of a sport or learn new sports on its own. It requires new data and some adaptations, but with the new system, “We can make these adaptions far fewer, faster and less costly than in previous systems and this drastically speeds up product development of the sportsbook.”
Here’s how he explains it:
“In the second generation, how we did it—this was very costly—you start from your core soccer product. Now we're going to understand how corners work. We need to study the statistics and build a corner model. Okay, we've done that. Now we're going to understand how cards work. And in the long run now we're going to understand how basketball works. What we believe we're on to here is that we invest a lot of years in one core model, and that model is not so much about whether it’s goals in football or points in basketball. It's more how we crunch this data. Of course, there will be some specific things—there's different rules. But the big investment is rather into one core—then scaling to others. So what our AI/ML now is doing is: we don't really tell it anything about what to expect from how corners in the football match works. We don't tell it anything about that. We more or less use the same model that we then use for cards—we just feed it with data.”
This also saves resources from investing in trading, Lögdberg notes. “And if it does a good job setting the probabilities of course then you need to invest less in the trading systems as well. Because you don't need so much of that because it is automated to a much higher degree.”
Kambi already had automated trading before this new launch, but he says some human traders are still needed now—though perhaps less than before.
“We, for many years, have had a quite automated setup. A live betting match will have 50 different markets. And odds need to change on just a possession in an NBA game. So it needs to be automated to a high degree. (But) a really skilled trader can change input parameters to these models. Now as external and internal data gets richer and machine learning algorithms more powerful, of course, they can do more of that job that the trader does. But I think for many years (going forward) even with the best machine learning models now on a big game for live betting: it will be possible for traders to add value as well.”
In the long run, Lögdberg believes this new product will make it much faster to roll out new products and new sports. “This will generate over many years, we believe, a much faster throughput of product development.”
In addition, the new models will give a boost to products which were aging, he said. “We felt we were reaching a limit for how much more exciting a product we can do for the end user with this whole setup.”
Lögdberg notes that there are still aspects that need to be proven as it grows. “I mean, I wouldn't say it's totally proven and to what degree it scales. but I think the signs we're seeing are certainly very positive.”
In-house incubator
Kambi has been interested in developing this type of AI product for a number of years but faced a challenge many companies do: it was busy building and maintaining its existing products.
“We talked about it for many years. We knew what had to be done, but it didn't really happen. Why? Because we have an existing big business. We have the deadline of NFL starting in nine months. We have to get the product up to par. We’ve then got to set aside the five best mathematicians to explore something new. And we're okay not evolving the NFL product now.”
The reason to finally do it: there’s intense competition in the industry so Kambi was trying to figure out: “How do we future-proof our business—in five years from now, and onwards?”
Eventually the company started discussing the new AI system in 2019 and formed a group of just three people to build it in summer 2020, Lögdberg said. The team didn’t have to worry about day-to-day current product development. In other words, it set up a kind of in-house incubator to try to build something fast, like a startup. The group eventually expanded to a much larger team.
“We said: we have to think differently. We have to think much more long-term. We set up a team. We asked them what they need. And we didn’t come with deadlines or short term product requests. We at the same time continued to invest in the medium-term to keep our NBA and NFL products leading. So then (we could) break out a proof of concept when it starts to prove itself.”
By focusing on the AI/ML models, it was able to push the technology further, he said.
“What we did—a key point—was asking them, more so the mathematicians than the traders, or product people: starting from an algorithm-first approach, what do you need? And we gave them everything they needed. We didn't put boundaries of how our platform works, how a trader works. We asked them to start and then let us know, how can the traders contribute?”
BetCrunch is sponsored by InPlay Innovation:
InPlay Innovation is the leader in AI-powered sports betting. Fully automated micro-markets, odds origination, and risk management.
Live vs Pre-game
In-play or live betting has become popular in the U.S. in recent years, and that should only increase in the future, Lögdberg said. But there are nuances if you look overseas.
He noted that unlike the U.S, the dynamics are different in Europe, where live betting has been available since around 2015 and is popular with tennis as well as soccer. However in Europe there are many more live games going on around the clock—since soccer is a global sport. So it’s easier to find live betting any time you want it in Europe, as opposed to targeting a specific team in the U.S.
On a random “Wednesday, at least in Europe, it's natural to go to the live betting product. You can say: ‘I don't necessarily know of any game going on right now but I'm more in it to get some action.’ Sort of like using it like a casino,” he said.
Kambi offers about 300,000 live events per year globally, while the “Big Six” U.S. sports have 50,000 if you go deep into college sports, he said. For live betting to really take off in the U.S. there needs to be a growth in interest in other sports or international leagues of the major sports, he said.
“Will (US consumers) be interested in European basketball or South American basketball? Maybe to some degree, but so far I think we see a much more of a limit to what is going on in the US,” he said.
But globally pre-game has come back in the last three years, he said. This has been driven by new markets, player props, the ability to create bets across sports, and new ways to cash out.
“Where live betting for many years kept growing…and in some markets would have gone up to maybe 70% of total bets—not revenues because margins are lower in live betting. But since these newer markets—bet builders, play props, cash out, as well—they have made the pre-match product more interesting again. So we're seeing globally a bit of a shift back to pre-match.”
Meanwhile, as the U.S. market has developed, more U.S. operators are focusing on retention instead of just customer acquisition fueled by bonuses. That opens up more opportunities for third party providers, Logdberg said.
“It's a positive development, but it is really positive for any company that is involved in creating the product. Because if it becomes less about big volume and marketing and bonus spend, and when it focuses on retention, it has to be more about a sticky product. So I think that's positive for us as a sportsbook supplier.”
The new technologies also make it harder for operators to keep up and give opportunities for Kambi, he said.
“The big change here was when the same game parlays and the bet builders—when they came around, it was very tricky for any sportsbook to adapt to be able to do that,” he said. “It was a big technical and algorithmic change when the bet builders came and certainly smaller suppliers were way faster.”