UNU POSTS 9-1 PREDICTION RECORD AGAINST THE SPREAD IN NFL PLAYOFFS

As the 2007 New England Patriots could tell you, all winning streaks eventually end. So it was for Unanimous A.I. this weekend, as our perfect run of picking NFL playoff games against the spread came to an end when Green Bay wilted in the Georgia Dome against a determined, high flying Falcons squad.

Despite that, we were redeemed a bit later in the day when the Patriots cruised to an easy win and covered at home over the Steelers, setting the matchup for Super Bowl 51. With that win, the published predictions from Unanimous A.I. yielded a solid 9-1 record in picking NFL postseason against the spread.

No we weren’t perfect, but getting 90% correct predictions against the spread is a very impressive accomplishment that’s worth breaking down a bit. By doing so, we can demonstrate just how potent the predictions generated by a Swarm A.I. can be, especially compared to experts and individuals.

First we should talk a bit about our methodology. Using our UNU technology, we ran two swarm sessions for each round of the playoffs. When the two swarms agreed (which was most of the time) the prediction was clear. On the rare occasions when they diverged, we dug deeper into the data, assessing the relative levels of confidence each swarm had in its prediction. For that, we have lots to look at.

Each swarm was first asked to simply tell us who they think will win, and by how much. (We DON’T give them the game’s point spread here.) That gives us one confidence factor to examine. Then we show them the point spread and ask them to again make a pick, and to put confidence behind it. That gives us two more confidence factors: the expressed high or low confidence, and also the brainpower derived in arriving at the decision. Finally, we add one final layer by asking the swarm to provide a hypothetical wager, between $0 and $100. That provides a final confirmation as to how the swarm is thinking about the game. When two swarms come into conflict over the same game, we looked at all these confidence factors in recommending our plays. With that understood, let’s see how UNU did in each round.

ROUND BY ROUND — UNU vs THE EXPERTS

For the Wildcard Round, UNU strongly recommended three of the four contests. Both Swarms easily tabbed the Steelers, Seahawks and Packers to cover at home. The Swarms disagreed a little on the outcome of Oakland at Houston, however. Although our results suggested caution in betting anything for this game, we did come down favoring Houston in this matchup.

How’d we fare against the experts this round? For starters, since UNU is picking against a point spread, those are how we can best compare. Chris Berman at ESPN hit two and missed two against the spread for Wildcard week. At CBS Sports, all eight experts make their picks against the spread, and UNU still looks great, beating their 3-1 record by one game.

Moving on to the Division Round, UNU once again had some strong recommendations. Most notably, UNU took the Steelers as an underdog on the road and also tabbed the Patriots and a lopsided 16 point spread to cover. Both Swarms had milder, but firm recommendation on Green Bay, clearly anticipating a very close game but again picking a road dog. Our two Swarms diverged a bit on the Falcons, but while one swarm offered no recommendation at all, the other Swarm showed heavy confidence in Atlanta. That became our lowest recommendation, but a pick nonetheless.

The experts did better this round, but not as well as UNU. At ESPN, Chris Berman had a tough 1-3 record against the spread in the Divisional Playoffs. At CBS, their panel of experts did much better, with the majority of their picks catching all four games correctly against the spread.

Which brings us to the week of games just completed. UNU went 1-1 against the spread this week. While both Swarms predicted a big New England win, they split over the Falcons and Packers. It was very close, but eventually weighted for confidence and performance to date we ended up making a very mild recommendation for the Pack. If there’s any consolation to finally losing, this week was the first week we asked both Swarms to recommend a play on the Over/Under lines in both games. Our Swarms strongly recommended taking both overs.

This was a tough round to call against the spread and the experts fared no better. At ESPN, Chris Berman went 1-1 this round, missing the same game UNU did by picking the Packers. CBS’s expert panel had it even rougher. They also missed on Green Bay, and their even split in taking the Steelers put them at 0-2.

The final scorecard of UNU versus experts was no contest. UNU’s 9-1 record bested CBS’s 7-3 record by 2 games, and blew away ESPN’s Chris Berman’s 4-6. Sorry Swami! But head to head records against experts are only one way to show how smart UNU’s Swarm Intelligence really is. To give you a clear demonstration of that, let’s look at how UNU’s picks fared against the picks of the individual users who make up our Swarms.

UNU Vs INDIVIDUAL SWARM PARTICIPANTS

Part of our process is to ask our Swarm participants to register their own picks before they join the Swarm each week. Each Swarmer registers not only a pick against the spread for each game, but also assigns a high or low confidence to their pick. That high/low confidence measure is then duplicated in the Swarm’s own picks during the session. Once all the game results are in, Unanimous AI researchers tabulate the participant responses, giving correct answers with high confidence two points, low confidence one point, and deducting one and two points respectively for incorrect low and high confidence responses.

With that scoring method, it’s possible to score a maximum of 8 points in a playoff round when there are four games, provided you pick all four winners at high confidence. It’s also possible for a Swarmer to go negative by being wrong too many times, at too high confidence. We can also score UNU’s Swarm responses the same way, since those also express high or low confidence picking against the spread.

There were eight points possible across all three playoff rounds, for a total of 24 possible. That includes four games each in the first two rounds. It also includes two outcomes plus the over/under bets on both games in the conference championship rounds. Amazingly, we had at least one, and sometimes multiple perfect scores of eight points by individual users in separate rounds. But there was also a huge standard deviation across the samples each week as well, and no user repeated a perfect score.

In the Wildcard Round, the average individual score of participants was 2.26 points out of 8 possible. In the Divisional Round, the average score for a swarm participant jumped a bit, to 2.5 points out of 8 possible. In the Conference Championship round, the average score fell way off, to 1.28 points out of 8.

The Swarm’s score is much, much higher. Although the individuals making up the Swarm could only score a 2.26 in the Wildcard Round, working together they scored 7 out of 8 points. In the Divisional Round, it was a repeat of the previous week. The individuals scored 2.5, but UNU managed 7 of 8 again. In the Conference Championship Round, UNU scored 5 of 8, even though the individuals scored just 1.28 points on average with the same questions. Across all three rounds, the UNU Swarm managed 19 points out of 24 maximum. The individuals who made up those swarms could only muster a cumulative average score of 6.04 points on the same questions and scale.

UNU and RETURN ON INVESTMENT

There’s one other way to measure UNU’s success in the playoffs this year, and that’s to examine the ROI of UNU “recommended” bets in each contest. As we point out constantly, these are not meant to be taken as recommended wagers at face value. They’re something researchers at Unanimous AI use as an expression of confidence in a pick.

With that said, a calculation of ROI based on those stated confidences can express the success of UNU’s weighted pick probabilities. Our picks and wager recommendations were all made Against The Spread using online sports book numbers from Vegas, so we’ll also subtract the house take (ATS bets typically pay the equivalent of a -110 Moneyline wager) for our calculation.

As the chart demonstrates, the UNU playoff Swarms offered a tremendous return on investment percentage. Of the $629.00 we recommended to play, all but $64 ended up generating a positive return. And if anyone in our Swarms bet the $152.00 our Conference Championship groups recommended in playing the “over” bets on this past weekend’s games, you got back another $138.18 for your trouble.

We hope this has been instructive for demonstrating the power of Swarm A.I. We’ll continue to provide examples of how swarms make smarter decisions and forecasts in the coming weeks and months. And of course our picks for Super Bowl 51 next week will be a big part of that. If you’d like to get our weekly newsletter sent to your email, just drop us a line below.