Are MLB Teams Getting Better at Drafting?

The draft has become a data-driven process with more information available than ever. Has that information led to better decisions?

Photo by Chris Chow on Unsplash

It’s draft week!

Even before I worked in baseball, I loved the Major League Baseball draft. It’s a singular event unlike any other in sports. The players aren’t household names and they don’t affect franchises immediately. And the players can come from anywhere. Selections stem from high school, junior college, and all divisions of college athletics. If you can play, someone will find you.

It’s also a process that has evolved over the last 25 years. Beginning with the release of Moneyball, teams began to incorporate statistics and other data sources into their evaluations. What once was a scout-driven art form became a much more scientific endeavor. By the time I entered a draft room (circa 2014), Division I statistics and the little pitch-level data we could get our hands on were built into draft models. In 2026, teams have access to stuff grades, bat speed, exit velocity, defensive metrics, and athletic testing for nearly every prospect in the draft, regardless of level.

The explosion of data has changed the way teams talk about players. But I’ve sometimes wondered if it’s actually helped them draft better. It doesn’t always feel like the additional information has led to better decisions. So I decided to investigate.

To do so, I pulled every draft from 1990-2019, and whether picks reached the majors and how much WAR they accrued within six years of their debut. I looked at the trend over time with an eye on three distinct eras: pre-Moneyball (1990-2002), post-Moneyball/pre-bonus pools (2003-2011), and modern times with pitch tracking (2012-2019). The assumption is that the talent available in a draft class is relatively fixed, putting the focus on whether teams are improving at ordering the best players in the draft.¹

What does that data tell us? Armed with more information than ever, teams have become better at avoiding busts and finding more useful players in the middle rounds, but in the area that really counts — identifying stars — they’re not appreciably better.

---

Let’s unpack this a little more.

The first thing I examined is the relationship between draft order and future value. I used standard approaches (R² , Spearman Correlation) and also developed an ordering efficiency metric that measures how close teams came to perfectly sorting a draft class.² By any measure, teams have improved at least modestly over time.³

Interestingly, the improvement comes almost entirely in the middle rounds. When we limit the analysis to first round selections, the improvement disappears entirely.⁴

Given the success rates of draft picks, players that make any positive major league contribution are considered good selections. But bench players and middle relievers don’t swing pennant races — and they can be acquired cheaply through other means. What we’re really chasing are star players.

To focus on stars, we can look at the percentage of a draft’s total WAR that’s captured at various cutoffs. If teams are getting better at ordering the stars, more value should be concentrated at the top. Instead, we see little to no improvement among the top-100 selections.

We can also evaluate improvement through a ratio — WAR from the top picks divided by the maximum WAR available had the team made the “correct” selection. The ratio tells us how much value teams have left on the table with their picks.

If you want an argument that the draft remains largely a dice roll, the fact that teams only capture about half of the available value in the top-100 selections is a pretty strong one. And again, there’s no identifiable trend or meaningful difference over time. The improvements don’t hold, and given the sample sizes, the differences suggest year-to-year variance rather than signal.

The best evidence that teams are improving where it counts is how early the best players are picked. From 1990-2003, the top player was selected on average 138th overall. From 2012-2019, that number dropped to 37, a finding that’s the same when you extend to the top three, five, and ten players in a class.⁵

But when you take the players as a group — like all the 10+ WAR players in a class — we see again that the difference comes after the top 30 selections. Teams aren’t finding value any more frequently in the early going.

Sliced several different ways, we’re presented a similar conclusion: despite more information to evaluate players, teams are no better at ordering first round selections than they were 25 years ago.

---

It has to be asked. Why have teams not improved anywhere near the rate that information has expanded?

There’s no way to say for sure, but we can come up with several plausible reasons.

First, just because teams have more information than ever, it doesn’t mean they’re using it correctly — or using it at all. One of the most challenging things about the explosion of data is figuring out what’s useful and what isn’t. To provide value, new data streams need to be understood, tested, and weighted appropriately. But on the amateur side, there’s often no history to lean on. Teams are forced to make assumptions and educated guesses if they want to make use of information immediately. And that assumes they’re using the information at all. Nearly every team in baseball has a draft model. What goes into those models — and more importantly, how they’re used to make selections — varies much more team to team.

The models themselves also contribute to the types of improvements we do and don’t see. My assertion that the goal of the draft is finding stars isn’t necessarily the way boards are built. Most models are designed to maximize expected value. The predictions combine a player’s probability of reaching the majors with an expected range of outcomes. Those models might very well output a similar order as something that predicts probability of becoming a star, but it’s a fundamentally different approach. An approach probably has a lot to do with the lower rate of busts and the more consistent ability to find value in the middle rounds.

In many ways, the same can be said for player development. While identifying players and developing them are distinct skills, they’re tightly linked and it’s difficult to parse out credit when players succeed. And recent advances in coaching tools and techniques are disproportionally likely to make fringe players into usable major leaguers or help players selected later in the draft.⁶ Most of the innovations in player development aim to increase velocity, add spin weapons, or generate more damage in the batter’s box. Players who already have those skills come off the board early in the draft. Coaches are also less empowered to make adjustments or innovate with top picks. The development strategy with those players has largely stayed the same: don’t screw them up.

The influx of data also hasn’t added much information around health, physical development, or psychology. The most critical component to determine success is player skill. If a player can’t hit a curveball or consistently throw strikes, there’s not much chance for them to provide value. For those that can, things like availability, physicality, aptitude, competitiveness, and ability to respond to failure become separators. And what teams are able to access in those spaces looks much more similar to 1990 than the on-field metrics.

But the simplest reason why improvement has been slow is probably that the draft is hard. The very reasons I love the draft — players don’t make an immediate impact and they come from all levels of competition — make it hard to beat with any consistency. Teams are predicting how human beings will adapt, mature, and develop. No amount of information will alleviate those challenges. In some ways, the slow and incremental progress shows that teams set a decent bar and it’s difficult to improve upon that.

---

One of the dirty secrets about working in a front office is that an inordinate amount of time gets spent on player acquisition processes that never come to fruition. The draft is different. While there’s some of that (given the hundreds of prospects evaluated), you’re guaranteed to come away with 20ish new players.

The makes the draft one of the most exciting and important things that teams spend their time on. It’s hard. It’s unpredictable. It’s taxing. But it would be somewhat boring if we knew who the best players were ahead of time. The fact that we don’t makes for more debate, better stories, and a rewarding feeling when everything comes together.

And still, the limited gains in the past 25 years hang over things. There is so much data available today. Teams are paying for it, talking about it, and in some cases, acting as if it’s revolutionized the way they draft. The improvements, while real, don’t meet expectations given the amount of information and the discourse around it.

That creates an opportunity. The lack of meaningful improvement can be both frustrating and invigorating because a ton of value remains on the table in the early rounds. Teams can continue on the same path, doing what they always do, or they can try something new — asking different questions and trying to find ways to leverage their information more effectively.

---

---

The Missed Opportunity in the MLB Draft

What Scouting Reports Say That Grades Don't

1. There are 43,462 picks across the 1990–2019 draft classes. I would have liked to include later years, but I stopped with 2019 because that was the last year before the draft shrank (2020 was five rounds, 2021 moved to 20 rounds) and wanted enough time to feasibly evaluate MLB impact. There are players from the later years who have not finished their full six years which was accounted for as best as possible. As I’ve done in the past, Major League WAR below zero was removed as I don’t think it’s right to treat players who reached the majors as less valuable draft picks than ones that did not, regardless of performance. I wanted to order selections by signing bonus rather than pick number, but that data is only complete for the later years. All data comes from Baseball Reference.

2. Ordering Efficiency is essentially a concentration index. It ranks players by an external variable (draft order in this case) instead of by outcome, then measures how concentrated WAR is at the front of the ranking. It’s computed as CI = (2/mean(WAR)) x Cov(pick_rank, WAR)/n per draft class and then normalized against the best possible CI. The more concentrated, the closer to 1.

3. Another way I tested things was to look only at college players. If the question is how much additional information has helped, the college ranks provide the most information and they’re where teams really have in formation edge over teams of the past. Those results didn’t change the results. The concentration index went down overall and it was better in the middle era, showing no trend over time.

4. Restricted to first-round selections, the correlation between draft year and ordering accuracy was essentially zero (r = −0.03, p = 0.89).

5. The draft used to be longer — 50 rounds but teams could pick until they passed — which does give some additional opportunities for the best players in a class to be found late. The probability of those players coming after round 40 are low enough that I didn’t worry about it, but if you control for that, the effect we see here is effectively cut in half.

6. There’s an argument that the improvement we see stem more from developments in PD than from actual gains on the scouting side.