The world of Daily Fantasy Sports (DFS) has evolved dramatically since DraftKings launched its first tournament over a decade ago. As the industry has grown, so has the competition, with fantasy gurus, mathematicians, data scientists, and your everyday DFS degenerates all vying for an edge. Last NFL season, a groundbreaking new strategy emerged that shook up the DFS community, offering a fresh way to gain that elusive edge.
For years, the best strategy was simple: use past performance to predict future outcomes, plug those predictions into a DFS optimizer, and pray that your "optimal" lineup would outshine the competition. At GridIronAI, we had a decent run using this approach, particularly in Showdown/Captain Mode single-entry cash game contests. In large-field $25 single-entry double-ups, I managed a 54% win rate, yielding just under a 10% ROI. While a 10% return over a 5-month period might not make anyone rich, any seasoned investor would jump at such a consistent gain.
But this year, we’re upping the ante. Instead of relying on traditional predictions to build a single optimal lineup, we’re taking a different approach—one that leverages the power of machine learning and simulations to create thousands of potential lineups and see how they stack up against each other in thousands of simulated contests.
So, How Does It Work?
We start by gathering thousands of data points on every NFL player, both on offense and defense. Some of the data points include team pass/rush rate in different game situations, player snap-rate, target shares in short, medium, deep, and redzone pass situations, and a whole lot more. This data is then fed into an NFL Play-by-Play Game Simulator that I’ve developed, which simulates each game on the slate 1,000 times (and we're working on boosting that number even higher). These simulations give us a range of realistic outcomes for every game. How do we know the sims are realistic? We ensure that the average of the simulated game lines (spread and total) match up close to the latest vegas lines so we know that the simulations are not “out to lunch”.
Next, we take these simulated outcomes and use them to “simulate a slate” by randomly selecting one simulation from each NFL game. For each simulated slate, we calculate the fantasy points for every offensive player and use those as projections to build the optimal lineup for that simulated slate. We repeat this process 20,000 times, yielding 20,000 potential optimal lineups. But we don’t stop there. We then simulate a DFS contest by pitting these 20,000 lineups against each other in 10,000 simulated slates. The lineups that consistently finish “in the money” across these simulations are identified as high-ROI lineups, while those that frequently fall short are flagged as low-ROI.
Why Is This Important?
This strategy offers a major advantage over traditional methods. The ownership and leverage considerations that typically require extra analysis are inherently built into our simulations. By letting the simulations dictate which lineups are most likely to succeed, we’re effectively outsourcing the heavy lifting to the AI. The result? A competitive edge in both cash games and GPPs that’s grounded in data, not just intuition.
What’s Next?
This season, we’re not only continuing to provide the optimal lineups based on traditional projections, but we’re also rolling out the top ROI lineups for both cash games and GPPs, straight from our contest simulations. These lineups have been battle-tested in thousands of simulated contests, giving you an edge as you build your DFS strategy.
So, are you ready to take your DFS game to the next level? Follow along as we dive deeper into this innovative approach throughout the season. With machine learning and simulations on your side, you’re no longer just hoping for the best—you’re strategically positioning yourself for success. From here on out, we’ll give you a fighting chance in the competitive world of NFL DFS.