4 Expensive Mistakes To Avoid When Using DFS Optimizers

DFS lineup optimizer is a cornerstone of any serious DFS players process. The ability to build hundreds or even thousands of lineups lets you scale up your DFS skill, letting you apply your skill to more contests at once. Now this sounds like a win win, but your optimizer might actually be the reason that you are losing at DFS.

So I want to uncover a few hidden, easy to make mistakes that you might be making with your optimizer that is killing the profitability of your lineup. And the last hidden mistake I'm going to talk about here is particularly tricky because it feels like you're doing the right thing, but without an understanding of how your optimizer actually works, you'll kill any potential your lineups had without actually realizing it.

My name is Jordan. I'm the head coach here at SaberSim, where we equip DFS players with the knowledge and tools they need to beat DFS. I've been using DFS optimizers for over 8 years, and I've tried almost every single one out there. I know how damaging these mistakes can be to your bankroll because I've made them, so hopefully this video lets you learn from my mistakes so you don't make them yourself.

Now, mistake number one is not making data driven decisions. I got into DFS because I thought I had an edge. I knew the sport I was playing well enough. to beat others picking the right players for my lineups. And it's great to trust your gut. You probably have a tuned intuition for the DFS sports that you play, but it is important to check your stands with data.

Backing up your decisions with data lets you mitigate any personal biases that you might have. And it also gives you the confidence to act on your gut when you're right. Now the risks here with this mistake are huge. Hitting that lock button or the exclude button on a player might leave you massively over or under leveraged on that particular player, even if you're directionally right with your stand.

And bumping up player projections based on who you like or dislike might be double counting data, That's already factored into the projections you're using. When you're looking for data driven reasons to play or fade a player, look for the stats that will put players into the game, and then give them opportunities when they're in that game.

Projected minutes, snaps, innings pitched, are all stats that will indicate how often a player will be in the game, And usage, target share, strikeout rates, things like that are all stats that will indicate the opportunities that players will have in those games. In general, if you're taking a stand on a player, you should be able to point to data that says you expect a player will be in the game more or less than the market assumes, or that they will do more or less than the market assumes with their time in that game.

If you can't back up your takes with data in that way, it's probably a good time to check yourself before you make this mistake. Part of the reason that this is particularly important is because DFS is a high variance game. You're not going to get the necessary most likely outcome or the outcome you were counting on every single night.

When your biggest play of the night fails, it's easy to just write that off due to variance. Backing up your stance with data gives you real world, quantifiable reasons to make your decisions. And it lets you check your work after. Did the player get the opportunities that you expected, regardless of how many fantasy points they ended up scoring?

Now, even if your decisions are well researched and backed up with data, your optimizer that you're using is probably still making assumptions about about players that have no basis in reality, and that brings me to mistake number two, which is making bad assumptions. Now, most traditional optimizers out there start you out with a list of players and a fantasy point projection of what they are expected to score on average that night, and that's it.

And when you hit build lineups, you get a bunch of lineups that are the top lineups by the sum of the average projection of players in that line. And if you click build over and over and over again, you will just keep getting the same exact lineups back. Now, what exactly is the problem with this? If you've played a DFS sport before, or a DFS slate before, you know that players don't always score their projections.

There is a range of outcomes associated with each lineup. Player. But what are these ranges of outcomes? Well, optimizers try to solve this particular problem with randomness. They make a bell curve, basically, of possible outcomes around the mean projection for a player. So if you've seen a bell curve before, the distribution basically says the mean curve.

is the most common outcome here, and then it is a bell looking curve on either side. Now, this is a very costly mistake. Player distributions do not look like bell curves all the time. Just to show you an example, this is a typical player distribution for a golfer in a tournament event where there is a bell.

a cut. So you can see for Sung Jae here, his average projection is about 64 points, which is somewhere around here on this curve, a bell curve would assume that the most common outcomes were all around the averages on either side. You can see that is very much not the case and see particularly how bad of an assumption a normal distribution or a bell curve is.

For this particular player. Now, different players playing different positions in different sports all have these different looking distributions and optimizers assume distributions are all exactly the same, that they are all this bell curve and these fake distributions that most optimizers come up with are not based in reality.

This means that you'll get DFS lineups building with a traditional optimizer, assuming players have upside that simply not there, or they'll assume players have way less downside than they actually do. Either way, you're getting lineups that aren't based on what actually can happen. So how do you fix this mistake?

Well, your DFS lineups need to be built using the real ranges of outcomes of players, not made up fake normal distributions, but an understanding of how often a player can score a different amount of fantasy points for that particular slate. Now, the most accurate way to do this is with a simulation.

Instead of starting with an average projection, and then trying to come up with a distribution of possible outcomes from that, a simulation starts with all the ways that slate could play out from the beginning, and then calculates the average projection from that. Building lineups with simulations means building your lineups With data that reflects what actually happens in the real world.

But that's not the only problem with bad optimizer assumptions. Optimizers also assume that all players playing in the same game are independent from one another. And if you've ever stacked your lineups before, you know that's not true. Brings me to mistake number three, which is stacking using rules of thumb.

Just like optimizers have to make up fake distributions for players to get a sense of what could be the real world outcomes, they also don't really know how players interact with each other in game. Sports are not played on spreadsheets. Players playing in the same game And again, if you've ever stacked your baseball hitters together or paired up a quarterback with a wide receiver in your NFL lineups, you already know this.

But the problem is that optimizers don't know how each player's relate to each other on their own. They require you The user to program in combinations of players that should be combined. And over time, the DFS culture and industry has come up with these rules of thumb about how certain players should be combined in lineups.

But the problems with rules of thumb is really the problems with rules of thumb anywhere. They are simple and they can't capture the complexity of the way that players actually interact with each other. When your optimizer picks a point on the distribution of a player to build the lineup, so let's say you've got a player that is projected for 17 points and it builds out that already fake distribution that we know is not accurate and it says in this game this player actually scores 21 points.

It ignores the correlation that player has to other players in the same game. Let's assume that we're talking about Patrick Mahomes here in this example, and the optimizer picks an upside outcome for Patrick Mahomes. It isn't any more likely now to include Travis Kelsey in your lineup. Because your optimizer doesn't know that those two players are playing on the same team and have a strong positive correlation.

But creating a rule to always include Travis Kelsey in every single one of your Mahomes lineups is oversimplifying the problem. The real question you're probably asking here is if you're playing Patrick Mahomes in your lineup as a team. quarterback. How much more likely should you feel to put Travis Kelsey in as your tight end?

Take baseball here. As another example, the rule of thumb is on draft Kings to create a five stack of your hitters to use five hitters from the same team in each one of your draft Kings lineups. Now, immediately, there are dozens of questions that pop into your head. Is it any five players? Is it any five players anywhere in the batting order?

Should you always be making a five stack no matter what team it is? If it's the Oakland A's with a 2. 9 run total, but you want exposure to that team? Are there players that you should only use in stacks? Are there players that you should only use as one offs that should never be stacked? Do the rules here change for different contest types or sizes or slate sizes?

At best, if you start to answer these questions or hear a lot of the conventional answers to these questions, they are just answered with new rules of thumb. Things like have one maximum batting order gap in your lineups or only stack players on teams with a certain team total or limit one offs to players that have a certain projection.

And it's this endless spiral of stacking more and more rules of thumb on top of each other, which ultimately ends up with you having to either spend a ton of time programming in all of these rules and accounting for all of the different situations, Or, oversimplifying the way that you're handling correlation and using blanket rules of thumb that attempt to account for a huge range of nuance and complexity with a very simple rule.

The best way to stack in DFS is to use a simulation. A sim understands exactly how often lineups with Patrick Mahomes and Travis Kelsey are successful. but also how often you can play either player without the other. Rather than oversimplifying your stacking rules with broad rules of thumb, or spending a ton of time trying to program all of these different rules for every single possible combination or correlation into your lineups, sims go straight to the source, creating the combinations of players that make sense based on how often those combinations of players are optimal in DFS.

Now, you probably already know that typical stacking rules and randomness rules lack some of the nuance of a simulation. But this last hidden mistake I'm going to talk about here is particularly tricky because it feels like you're absolutely doing the right thing when you're making this mistake. But without understanding how your optimizer really works, you'll kill any potential your lineups had without even realizing it.

That brings me to mistake number four, which is diversifying with bad lineups. Now you've probably had the experience of pressing build on an optimizer to build your lineups for a DFS slate. and immediately getting back 100 percent exposure to a bunch of different players. And it doesn't take many slates of losing all of your entry fees doing this to make you want to start diversifying your lineup portfolio, making sure that your lineups are different enough from one another so one bad game doesn't completely kill you.

The problem here is that the tools optimizers give you to do this makes your lineup significantly worse in the name of diversification. So let's talk about what some of those tools might look like. Option one is you set minuniques, requiring that each lineup you build is x players different than each other lineup that you've built so far.

Now the problem is that this constraint will force your optimizer to build increasingly worse lineups. By lineup 150, if you're building for a 150 max, say your minuniques is set at three, that lineup must be three players different than every other lineup that you've already built so far, meaning that the optimizer has a lot less options of what kinds of lineups to build, and the lineup 150 in that case is going to be much, much worse quality than lineup one you built in your pool.

Option two, for example, is to start setting minimum or maximum player exposures. And that's it for now. Requiring that your total exposure, your percentage of lineups that include a particular player does not exceed or at least meets a certain threshold. The problem with exposures here as diversification tools is that you will end up with lineup bunching.

You will have big bunches of lineups that all look very similar to one another. For example, Your f First, lineups built will have no restrictions, but after your 75th lineup built, if you have a 50 percent maximum exposure set and you're building 150 lineups, all of your players that have that max exposure of 50 percent will all get restricted at the exact same time, meaning that your 76th lineup will look dramatically different than your 75th lineup and will be a much worse lineup overall.

Now, the fundamental Problem here either way is that we don't want our desire or need to diversify our portfolio to ruin the quality of our lineups. In other words, desire that we have to diversify our DFS portfolio should not affect the way that DFS lineups themselves are built. We want to take a set of great DFS lineups and find a diversified set from among them to play.

And you can think about this the same way you think about your investment portfolio. You would never want to make straight up bad investments purely to diversify your portfolio. But at the same time, it's not smart to put all of your money into just one good stock. What you want to do is start with a set of great profitable DFS lineups and then find a diversified set from among them to play in your contest.

Diversification itself should not be in consideration in how your DFS lineups are constructed, but only in how they are selected. This is a very subtle difference, but one that is extremely important to your DFS results. Now on SaberSim, we build your lineups using our play by play simulations first. And then allow you to diversify after when you go to select which ones you want to play into your contests.

But in other optimizers, the most important thing you want to do is make sure you are not letting diversification heuristics affect the quality of your lineups. It's clear that understanding proper diversification principles is crucial to success in DFS. But most DFS players vastly underestimate the importance of this part of their game, which can lead to players that would otherwise have been profitable going broke far before they ever gave themselves a chance to realize their profit in these games.

So next up, watch this video to learn why diversification is so important in DFS and how to do it right.

‍