AI vs. NFL Experts: Can it Deliver an Edge in Sports Betting?

It’s almost impossible to go an hour without hearing about AI these days. Its applications are seemingly endless. Whether you’re talking to a virtual assistant like Siri or using it for much more demanding tasks, you’ve likely had an experience with it.

It’s that same versatility that got us thinking about AI’s untapped potential. Naturally, we thought about sports betting. Obviously AI can provide simple picks for games, but would it be worth it to rely on those predictions?

Over the first four weeks of the NFL season, we put out AI sports betting assistant, OddsTrader Turbo, up to the task of making picks against the spread. We also tracked the public’s consensus, as well as picks from six NFL experts, to see how AI stacks up against the competition. The results may surprise you.

How it Works

Every week, our AI betting assistant provides betting predictions on our NFL picks page. These picks are rated on a scale of one to five stars, with five-star picks packing the most value. We recorded those picks every Thursday.

Additionally, we also tracked public picks made by NFL experts from various news outlets and publications. This includes CBS Sports, Bleacher Report, and The Athletic. 

For good measure, we also wanted to compare the picks made by OddsTrader Turbo and our pool of NFL experts with those made by the public. To do this, we tracked the public’s betting percentage for each spread pick and went with the majority. 

The Results: Experts

Correctly picking which team will cover the spread in an NFL game is no easy feat, and the experts proved that. With a combined win rate of just over 47%, the first few weeks of the season turned out to be quite unpredictable. In fact, only one of the experts we tracked had a win rate of 50% or better:

• Pete Prisco, CBS Sports: 36-28 (56.3%)
• Frank Schwab, Yahoo Sports: 30-31-3 (49.2%)
• Jamey Eisenberg, CBS Sports: 30-34 (46.9%)
• Brent Sobleski, Bleacher Report: 29-35 (45.3%)
• Gary Davenport, Bleacher Report: 28-36 (43.8%)
• Vic Tafur, The Athletic: 25-35-4 (42.2%)
• Combined Expert Record: 178-199-7 (47.3%)

Week 4 proved to be particularly tough for the experts. Prisco (10-6) was the only expert to go .500 or better, with the others ranging from three to seven wins. Chargers-Giants and Panthers-Patriots were especially challenging calls. 

With that said, this illustrates the fact that sports betting can be deceptively difficult. Even the most informed among us can have the tables flipped on them by unforeseen circumstances. Knowing which expert to tail – and when – only adds another layer. 

The experts are still experts, though, which was proven by following the public betting consensus over the NFL’s first four weeks.

The Results: Public Consensus

Preseason expectations often differ from reality, and there’s no better example of that than the public’s record against the spread in Week 1. The consensus was correct for just three out of 16 games as teams like the Colts, Packers, and Commanders greatly surpassed expectations.

The public caught on as the weeks progressed, but more often than not, they found themselves in the red. Just take a look at the weekly breakdown:

• Week 1: 3-13 (18.8%)
• Week 2: 9-7 (56.3%)
• Week 3: 7-9 (43.8%)
• Week 4: 6-10 (37.5%)
• Overall: 29-35 (45.3%)

There was also some evidence of overcorrection from the betting public. The Colts, Packers, and Commanders – the same teams that burned them in Week 1 – all failed to cover the spread in Week 4 despite strong consensus. Perhaps it’s best not to lead with your emotions when it comes to sports betting?

The public did outperform some of the experts, which is commendable. However, a 45% win rate (and only one week over .500) isn’t a strong argument to go with the consensus. In fact, you would’ve had better luck betting the opposite of what the public preferred. 

The Results: OddsTrader Turbo

So the experts did fairly well, and the public wasn’t quite as hot. How about our in-house AI?

OddsTrader Turbo turned in a win rate of 52.5% and went .500 or better in three out of four weeks. That’s significantly better than the betting public, and it also outperformed all but one of the experts we tracked (Prisco hit at a 56.3% rate). 

Turbo missed the mark in Week 4, getting burned by high-value underdogs like the Colts, Jets, and Cardinals. However, it racked up at least eight wins in each of the other three weeks and nailed nearly 67% of its picks in Week 3. 

• Week 1: 8-8 (50%)
• Week 2: 8-6 (57.1%)
• Week 3: 10-5 (66.7%)
• Week 4: 5-11 (31.3%)
• Overall: 31-28 (52.5%)

Let’s provide some additional context to these predictions. Not every betting pick is valuable or even worth making, which Turbo understands. That’s why our in-house AI assigns a rating to each of its projections. One-star picks are the least valuable, while five-star ratings provide the highest expected value.

Turbo skipped providing a spread pick altogether for a few games over the first four weeks, which is why not every record adds up to 16. Moreover, the least valuable picks (one- and two-star, in this case) were just 9-14 overall.

However, by skipping the low-grade picks and focusing solely on four- and five-star picks, Turbo compiled a 17-12 record, or a 58.6% win rate. This means that the predictions with the highest expected value were also the most accurate. 

• 1-Star Picks: 5-4 (55.6%)
• 2-Star Picks: 4-10 (28.6%)
• 3-Star Picks: 5-4 (55.6%)
• 4-Star Picks: 10-5 (66.7%)
• 5-Star Picks: 7-7 (50%)

Is AI Good for Sports Betting?

In this case, our in-house AI proved to be a strong option for NFL betting picks. OddsTrader Turbo significantly outperformed public betting consensus and finished with a better record than five of the six NFL experts we tracked for the first four weeks of the season.