IBM's IOD/Business Analytics Forum event was held recently in Las Vegas. If you weren't able to make it and want to catch up on some of the action and video highlights, please click here.
Quick story from the event: A group of us decided to play poker one night after our conference duties had ended. There was one spot left at the table and a non-IBMer sat down. I felt bad for him. A few hands were played, and our new friend immediately lost them all. And, he lost a few more. Finally, he asked how we all knew each other. We kindly told him that we were there for a conference and that we all worked at IBM on the SPSS brand and told him about predictive analytics.
He sat there long enough to not appear to be running as fast as possible from us, but did quickly gather his chips and bid us adieu.
It reminded me of a quote from the movie Rounders, "If you can't spot the sucker in the first half hour at the table, then you are the sucker."
Now, there was no cheating involved, but we all held a simple advantage -- the ability to gather data, quickly analyze it and make smarter decisions.
As the 2010 World Series of Poker finals come to a close tonight in Las Vegas, I thought it was a good opportunity to talk with Marcus Hearne, an IBM product marketer for the SPSS predictive analytics brand (and avid poker player), to discuss the comparisons and benefits of using analytics in poker and business:
When/how did you start playing poker?
It was at the start of my career at SPSS when a colleague invited me to a friendly dime-and-quarter game he had with some mates. Skip forward a few years and I found myself at some interesting high stakes games and tournaments. It all came to a happy end when I got married. The interesting, yet obvious, point is the parallel between my improvement at poker and becoming far more immersed in the SPSS product family as I shifted from marketing into product management. I realized that using data in decision making was critical to my being a better poker player.
What is the foundation of any good poker player?
Math. You have to know the odds. And it goes well beyond the 52 cards in the deck and the likelihood of drawing a king, or a heart to make a flush. It comes down to being able to quickly calculate how much is in the pot, the bet you have to make to call or raise, where you are in the betting order, and the ratio of my stack to the pot, the pot after my bet, and the expected total pot at the end of the hand. A good poker player can also figure out what their implied odds are given the hand, the pot, and the opponents (particularly their stack of chips). If you watch poker on TV, you can occasionally hear them breaking down the odds and ratios quietly and quickly while sitting at a table.
As I've learned over the years, good poker players hate to gamble unless they suspect it's a coin flip, or a 50 percent chance of winning. A good poker player will always try to play a hand where they see not just the odds of winning being high, but the cost to compete making sense in terms of the rewards available. For example, if I offered to flip a coin with you, but said you would have to pay me $100 each time and would win back $105 when you won the toss, you'd rightly refuse. However, if I then offered $1,005 each time you won, it would be an excellent investment. Given over time there is an equal distribution of heads and tails, then for 100 tosses you could assume to win 50 times. You would have handed over $10,000 and won $50,250. Shift that to the poker table, and then adjust the reward to be the pot in the middle and your chance of winning based on your hand, and you can then determine the amount you are willing to "invest" in the hand. But that's it for gambling in poker. Everything else is data and analysis.
What about luck?
Sure, being lucky is a part of gambling, but it doesn't make you a good poker player. Luck is by definition the ability to win against long odds, such as a number coming up in roulette, the hard six (3 and 3) being rolled in craps, or being dealt blackjack three hands in a row. In the end it's not really luck at all, but simply a 1 percent chance of an event happening. So I ask, would you rather bet on the 1 percent event, or the 50 percent event? Stick with math and put your money where you see an advantage. Sure, a 99 percent likelihood of success means you're going to occasionally lose to someone who gets "lucky" and hits the 1 percent event against you, but you will succeed the 99 other times. The same goes for business. This is exactly what marketers are trying to do by reaching out to customers who have the highest probability of responding to their offers, and doing so with limited resources. A marketer who simply sends out mass amounts of direct mail to every contact they can will find it is akin to a roulette player putting a chip on every number on the board. Yes, they'll hit the winning number, but the cost-to-reward ratio is awful. Any marketer will tell you that having insight into which customers are most likely to respond is money in the bank, and the first step is collecting data.
Why is data collection so important?
It's absolutely essential. Everyone knows the Hollywood movie version of poker where the master player looks for physical tells in his opponents, such as shifting eyes, the bead of sweat or the shaky hand. It's actually rubbish. Well, most of the time. The reality is that the top poker players sit there continually gathering data to build a picture of how their opponents play. Eventually the real "tells" are how a player bets, from which position, into what size pot, at which point in the hand, and so on. Ultimately, these better players build a model that maps out patterns of behavior and can whittle down the possible cards their opponents are holding. For example, I might observe an opponent's betting patterns and the results for 20 hands. At the start of the 21st hand, an opponent might raise which would allow me to then narrow down their possible starting hand to five possible combinations. As the hand progresses, and the community cards are exposed, I can further narrow down the cards to what I believe they have to a high degree of certainty, and then act accordingly. Using additional data, such as a player's adversity to risk, or the fact that they might play loose with more chips in front of them, I can either maximize the damage to their stack, or minimize the damage to mine. This is what allows really good poker players to not even bother worrying about their own cards and simply play their opponents.
Are there other types of "tells?"
Sure, sentiment, attitudes and opinions. It's what we call "unstructured" data, and is what usually tips my predictions of a player (and their hand) into a higher grade of accuracy. It's as simple as listening. Say at some point in a hand I think you have one of two hands - one of which beats me and the other I beat you. Suddenly, you start talking about the Cubs. In past hands I've noticed that you talk about the Cubs when you are starting to think that you might lose. This sort of information brings about a whole new dimension to my model and I now know that regardless of what cards you have, I can scare you off the hand with a huge bet. In poker this can be tricky. People are constantly trying to act, which is why this data is so hard to capture. It's critically important that if you're going to collect unstructured data you correctly identify it's true sentiment and rank its worth. Thankfully, in business it is much more straightforward. People are usually quite frank and forthcoming regarding their opinions of a product or service. Using unstructured data to better target customers is far easier as a marketer than it ever was for a poker player.
Seems that poker has changed quite a bit since the days when cowboys played in a saloon?
Back in the day, a good poker player probably instinctively conducted analysis of data they gathered as they played. What has really changed in the last few decades is that a few of the poker greats actually captured very detailed explanations of the odds, the analysis required, and which data is the most important. This has armed so many people with the skills required to play poker that the result was a boom in poker's popularity on a global basis. Once again, and I hate to belabor this point too much, but there is an amazing parallel to business. Take for example two insurance companies - one using predictive analytics to identify fraudulent claims, and the other not. The one with analytics can now more quickly identify those claims to investigate and fast-track the safe claims. This improves customer service and better targets budget dollars to investigators for the fraudulent claims. The insurer not using analytics is like the dusty cowboy in the saloon, still holding onto the past. As far as I'm concerned, I'm "going all in" with analytics, to use a betting term.
Is it as easy to use analytical skills in poker as it is in business?
It's easier in business, honestly. In poker you have to do it all in your head, you're prone to emotion and gut instinct, and you're on your own. In business you have the tools and data at hand, right in front of you, and you can slice and dice away to your heart's content until you have what you need (in lieu of deadlines of course). IBM has spent many years trying to make customers as successful as possible. Everything we do is aimed at giving people the tools to draw the intelligence from their data to make the best decisions possible.
Care to test what you just talked about and play a hand?
Sure, do you want to borrow some of the money I took off you in Vegas?