clock menu more-arrow no yes

Filed under:

How would analytics have fared in predicting Buffalo Bills’ 2020 draft picks?

New, comments

Would a data-based draft model have done a decent or horrendous job predicting the Bills’ draft choices?

Rumblings readers with better memories than me might remember that for the 2019 NFL draft I attempted to predict the Buffalo Bills’ choices using an analytics-based model. Wait, what? What data would you even use? Under the premise that even wealthy people have a finite amount of time, I used various sources to chart the frequency of visits among players and position groups to see where the Bills were focusing their attention.

Well I forgot to do it this year but hindsight is cool, right? The good news is that the data is still the data so we can see if it may have helped make draft-day predictions if I had, you know, made some.


By player

This year was pretty messed up as far as having visits and all that, so the data was also a little wonky. This might have been a year where multiple visits with a player just wasn’t going to be the standard.

That said, three players met with the Bills twice. Jon Greenard (DE), Javon Kinlaw (DT), and Denzel Mims (WR) were the only players that had more than one meeting with Buffalo according to chatter.

Greenard was projected as high as the second round, in line with Buffalo’s first pick. Kinlaw drew praise as high as All-Pro potential and went in the first round. Mims was widely felt to be a day-two player, and was picked only five spots after the Bills selected A.J. Epenesa.

Overall, I would have seen a model based strictly on this as moderately successful. It correctly would have guessed two of the Bills’ first three picks by position (DE and WR) and pointed toward plausible players as the Bills were finally coming to the podium for their first pick.

By position group

This is incredibly straightforward. Here’s a chart of how many visits the Bills had sorted by position group.

Based on the number of visits by position group I would have declared a wide receiver was the top target with a defensive lineman next up, with a preference toward end. And yes, a running back would be predicted to be in the mix. This would also have been moderately successful in that it would have predicted the first three position groups selected, albeit not in the right order.

By school

One of the things that’s often brought up is that teams will interview players who went to school with a target to find out how they’re viewed. What this means for data nerds is another way to sort player visits with the theory being that interviewing these tangential targets will increase the count by school.

Unfortunately, a chart here would be cluttered with ones and twos to the point of being nearly meaningless. To sort it out faster I can just say that only two schools had a count higher than two. Tied at four apiece were LSU and Utah.

Adding this to the model may have led to predicting Zack Moss as a target. Potential targets from LSU included Justin Jefferson (WR), Clyde Edwards-Helaire (RB), and K’Lavon Chaisson (DE). All three went in the first round, though there’s some wiggle room to think maybe one would have fallen based on pre-draft profiles.

Aside from Moss, John Penisini (DT) was the only other Utah player to meet with Buffalo at a position of interest defined as WR, RB, or defensive lineman. Ultimately a sixth-round player, it would be unlikely for the Bills to have poked around his teammates to any large degree to scout Penisini—which would have left Moss as the most likely Utah player of interest.

Who would have been my pick?

First off, I would have made sure to put a disclaimer on this one. Waiting until the 54th pick, the amount of volatility created from other teams’ first 53 choices are problematic to say the least. That said I would have went with Denzel Mims I think.

I would have and do think that Buffalo likely worked hard to put a valuation on Justin Jefferson, Clyde Edwards-Helaire, and possibly K’Lavon Chaisson to see how far they’d be willing to trade up. However, they thought enough of Mims to meet a second time and the drastically higher number of receivers on the list makes it dishonest for a truly analytic approach to pick any other position group.

Is the premise sound?

I think it went pretty well last year and even better this year so my answer is “yes.” To reiterate, though the order wasn’t right, the analytics by position correctly chose the three priorities for Buffalo. Sorting by school seems to have had evidence of a priority player. General manager Brandon Beane indicated on this year’s “Embedded” they were trying to trade up for Zack Moss for 15 minutes before he finally fell to the Bills.

Another thing to note is that none of the usual sources knew that the Bills met with A.J. Epenesa and Gabriel Davis at the Combine. That wasn’t discovered until “Embedded” was released. This means that Buffalo met with their top three picks. Now that we have a couple Beane drafts behind us, it seems clear that the Buffalo Bills don’t like to waste effort when it comes to the draft.

All that said, a model based entirely on analytics is unlikely to get it completely right when it comes to specific players—especially the later the rounds go. The discussion above focuses on the picks through Gabriel Davis and I’m not shy about saying that the model, while useful until then, is almost entirely useless afterward. At best you could argue Isaiah Hodgins isn’t a surprise with the amount of work put in on receivers.

The moral of the story seems to be that a person who is familiar with draft prospects could easily refine their areas of focus based on the model, but could not be replaced strictly by the data.