Madden 17 Deep Dive - A Look Inside CPU Playcalling

[DeuceDouglas]

So over the last couple weeks I've come across a couple of great write-ups over on Reddit that took a look at player development as well as the effectiveness of OL ratings. I really enjoyed reading those and it's inspired me to create my own write-up about one of the more under the radar issues I've had with Madden over the past several years and that's CPU playcalling. This won't be nearly as deep as those partly because the issue isn't really that difficult to identify and it doesn't take much digging at all to find the root of the problem or the solution to fix it. It may seem like a small thing and with the increasing emphasis on the PvP aspects of Madden, it's not that surprising that it has been something that has been an afterthought year after year but I still believe that there is a large chunk of Madden players (especially CFMers) that prefer the alternative experience and would greatly benefit from this.

Background

If you're unfamiliar with how the CPU makes their calls, Madden 11 introduced the Game Plan feature. The feature was originally meant to get the user through games quicker by selecting 20 plays per a given game situation (1st down, 2nd & long, 3rd & short, etc.) and assigning them ratings based on a 5-star system. The game would then act as a "coordinator" and call the plays for you from what you had put in each situation. The higher rated the play, the greater liklihood that play will be called in a given situation. As I mentioned before this was something initially intended for the user but it ended up giving us a look behind the curtain into CPU playcalling as well because this is also what the AI uses to determine what plays are called. Over the years the feature has pretty much been phased out in favor of alternative playcall suggestion methods which include things like community suggestions along with strategy and frequent play calls. Unfortunately, despite beneficial changes to make user playcalling quicker and deeper, the same luxury has never been provided to the CPU and sadly their playcalling is really no deeper than it was when the feature was introduced six years ago.

Don't Judge a Playbook by Its Cover (Or GamePlan)

If you were to take a peek into the custom playbooks section of Madden and start messing around, you'd see that each teams offensive playbook carries around 450 plays. The playbooks themselves max out 500 plays so almost every team has a nearly filled playbook so quantity isn't really the issue. At first glance you look at it and think that there is a lot of plays for the CPU to use but once you dig deeper you see a far different story unfold. I picked five random playbooks (Jaguars, Colts, Packers, Vikings and Dolphins) and decided to look into all the plays in their book to see how they were rated.

The first thing that becomes immediately noticeable is just how many plays will never even have a chance of being called because they have no star ratings for any given situation. I went through each of the five playbooks and counted how many of these types of plays there were in each, here are those numbers:

Team	Unused Plays	Total Plays	%
Jaguars	265	429	61.7%
Colts	248	399	62.1%
Packers	235	394	59.6%
Vikings	260	421	61.7%
Dolphins	268	420	63.8%
Total	1276	2063	61.8%

The first thing you'll probably notice is that I mentioned before that each playbook had around 450 plays but in the Total Plays column the numbers are more close to 400. This is because the total plays includes special teams, hail mary and goal line plays (which are also virtually nonexistent in most teams books and unable to even be added to situations including goal line). Just looking at the Jaguars book there's 30 additional plays with special teams, hail mary and goaline included but still nine extra phantom plays so I'm not entirely sure where the total play number comes from.

Regardless, looking at those numbers you'll see just how much of each teams playbooks will never even have a chance to be called. The CPU isn't utilizing even 40% of their playbook and if I didn't include plays that only had star ratings for situations like goaline, 2pt, 4th downs, etc. that number would become even higher. To go along with that, a lot of teams playbooks and situations include many of the same concepts and types of plays within the same situation which just adds to problem with variety and utilization which I'll get to more later.

The obvious problem here is the amount of plays per situation. I said earlier that sadly playcalling is no deeper for the CPU than it was when the feature was introduced six years ago and it is entirely true. When GamePlan was first introduced there could be no more than 20 plays for a given situation, it was a hard limit. However, with the bump to current gen that hard limit disappeared and the amount of plays allowed to be added to a given situation is seemingly endless. But take a look at any teams playbook for a given situation and how many plays will you see? You guessed it, only 20. And sometimes it's not even that many but that you will never see more than 20. 400 offensive plays at hand and at most only 5% can be used at a given time? That is a huge problem.

The Four Vertical & Curl Flats Conundrum

As I mentioned before, a lot of situations include many of the same concepts and play types within them. This is no more evident than with the Verticals as well as Curl Flats. Go into any playbook and you're bound to see some form (and even multiple forms) of curl flats in most situations. To the same extent, go into any playbook and look at the 3rd and Long or 4th and long calls you'll notice that the amount of Four Verticals plays or concepts ranges anywhere from 5-10 plays out of the 20 total which further narrows down the already limited playcalling. The main issue behind this, as well as Curl Flats, is the AI's general incapability of running those plays. I can't explain the amount of times I see these plays called and the QB just stands around waiting to get sacked. Or with curl flats or slants the QB will take the snap and either throw it away immediately or stand in and get destroyed. They're immediately recognizable and often well covered which results in a large amount of throw aways and sacks.

The similar and arguably more important aspect of this is that virtually every team seems to play the same. You go into most teams playbooks and you're likely to find the same types of plays in the same situations. Curl Flats, Sticks, Slants, Corner Strike, etc. Of course these are plays and concepts every team runs but when you only have 20 plays and give or take ten pass plays, you can't have these as abundant as they are for almost every team. The end result is that pretty much every team lacks a true identity and a unique feel to them. You might have certain teams throwing in some unique plays here and there but by and large everybody feels pretty much the same.

I decided to take a look at a few teams and see how their gameplans (1st through 3rd down) stacked up with their overall playcalling from this past season. The teams I chose were the Saints (2nd in pass attempts, 20th in rush attempts), Panthers (27th in pass attempts, 1st in rush attempts), Packers (18th is pass attempts, 13th in rush attempts) and Vikings (32nd in pass attempts, 4th in rush attempts). I tried to pick teams with a large discreprency in ratio as well as a more balanced team to see if that was represented in their GamePlans. Here's what it looks like:

Saints

	1st down	2nd & Long	2nd & Medium	2nd & Short	3rd & Long	3rd & Medium	3rd & Short
Run	13 (3.5)	8 (2.5)	11 (2.5)	13 (3)	0	0	14 (3.5)
Pass	7 (1.5)	12 (2.5)	9 (2.5)	7 (3.5)	20 (5)	20 (5)	6 (1.5)

Panthers

	1st down	2nd & Long	2nd & Medium	2nd & Short	3rd & Long	3rd & Medium	3rd & Short
Run	14 (4)	11 (2.5)	12 (2.5)	14 (3.5)	0	1 (.5)	16 (3)
Pass	6 (1.5)	9 (2.5)	8 (2.5)	6 (3.5)	20 (4)	19 (5)	4 (2)

Packers

	1st down	2nd & Long	2nd & Medium	2nd & Short	3rd & Long	3rd & Medium	3rd & Short
Run	13 (4)	6 (2)	11 (2.5)	12 (3.5)	0	0	12 (3)
Pass	7 (1)	14 (3)	9 (2.5)	8 (1.5)	20 (5)	20 (5)	8 (2)

Vikings

	1st down	2nd & Long	2nd & Medium	2nd & Short	3rd & Long	3rd & Medium	3rd & Short
Run	14 (4)	9 (2.5)	12 (3)	15 (4)	0	0	18 (3.5)
Pass	6 (1)	11 (2.5)	8 (2)	5 (1)	20 (5)	20 (5)	2 (1.5)

So on the surface, not too bad at all. The only issue I really see here is that each team has a very similar, run heavy presence on first down which creates a lot of that predictability that I mentioned earlier. Maybe some oddities for a couple situations but just looking at those numbers, it seems as if the distribution and ratings is overall not too bad. So the biggest issue within this still remains the lack of variety when it comes to plays in these situations. On first down you only have six or seven pass plays from an entire playbook for each team and they all seem to be sharing at least two or three of them leaving on four or five unique pass plays on first down. And as you get deeper into the downs it becomes even more common. Slants, Strong Flood, Curl Flats, Mesh, Stick, etc. make up a large portion of each teams pass plays and that would be perfectly fine if there were more to choose from and the AI was capable of executing these plays better than they currently are. Just looking at the Vikings on third and long and 12 of their 20 passes are some variation of four verticals, the Saints have eight and both the Packers and Panthers only have five. For certain situations I can understand it a little bit more because I can still hear John Madden circa 2002 talking about how there isn't much in the playbook designed to pick up this many yards on 3rd down. But when it comes down to other more variable situations, there has to be more available to them to keep things interesting.

The Solution

Short Term

So every problem obviously needs a solution and luckily for something like this the solution is pretty easy. If you haven't seen it already, there are great things going on in this thread. But it's a short term fix that requires a workaround and is also time consuming and can tend to be somewhat unstable but it works and it works well. I can attest to just how much better the CPU plays using custom playbooks that do nothing more than expand their base playbook for every situation. I don't know if it's possible with 17 but they could completely eliminate the need for the workaround by either expanding books themselves via patch or figuring out a way to have custom books load into CFM or even implementing the GamePlan functionality into CFM. If we could set each teams playbook at their scheme screen (which would still take 32-team control) it would make this workaround far easier and much less of an annoyance.

Another similar workaround is to use the generic playbooks that the game provides as it will provide a somewhat similar result because the vast majority of the plays in each situation are rated a half-star so you'll get a bit more variation at the expense of a more generic book and ultimately more generic playstyle if you choose to use them around the whole league. This is something that was used heavily back in Madden 13 when there were GamePlan issues that caused the CPU to run the ball way too often. However, the same issue is present with these books as the defaults with only 20 max alotted per situation.

Long Term

The long term is essentially the same as the short term but it would come from their end. First and foremost, the immediate fix has to be a large increase to plays available to the CPU in a given situation. Whether they bump it up to 60, 80, 100, 120, whatever, the CPU simply has to have more of their playbook at their disposal. You just can't have every team leaving over 60% of their playbook completely unused and untouched, especially if you're playing multiple seasons of CFM. It makes teams feel stale, predictable and ultimately lifeless.

That's the easy solution, all it takes is time really. As for the hard, more complicated solution I've got a few of those as well:

I wrote about the idea of Dynamic GamePlans in the thread I created about rebuilding CFM and I think that would be a great solution but it would have to be a fully fledged feature. The idea being that players would come paired with traits that would either boost or decrease the star rating of certain concepts of plays based on their player type. I go into more detail in that thread and I encourage you to check it out but the basic gist is that something like a great power back would come with a half or full star increase to all inside runs and decreases for all outside runs. Conversely a speedy back would have just the opposite. A pocket QB would have a huge decrease on any QB runs while a mobile QB would have increases for those and read options. The idea being that each teams playstyle would change along with their roster and they would truly play to their strengths and away from their weaknesses over the course of several years in a CFM.

I'm not sure how difficult it would be but taking a similar approach to how they've done commentary this year could also be something they could look into. Expand the base playbooks like I've mentioned and then at points throughout the year add and alter plays and star ratings for each team depending on how they're playing throughout the year and based on those teams tendencies. It could be something done monthly and it would keep the game a bit more fresh throughout the year for those that play CFM.

I plan on tweeting this to the devs so that it will hopefully garner some more attention and possibly be something they look into moving forward because I definitely think it'd be something that would greatly change the game for the better. I know there have been a few threads on this since release but I just wanted to consolodate the true issue and problems with playcalling that cause this into one place. If you have anything else to add to this feel free to comment on your experiences with CPU playcalling. Also, if there's any ways you think that playcalling could be improved within (or even beyond) the current system I'd love to hear those as well. Thanks for reading and I'll leave you with this comical playcall by the CPU that would probably make Pete Carroll proud:

[rich8682]

Honestly this and QB AI are the top of my wishlist. Deeper franchise would be nice, but it's good enough that with those 2 fixes, I'd probably play the full 30 years.

Sent from my SM-G386W using Tapatalk

[KershawsCurveball]

Changing schemes to alter the AI pb to a generic one does not require 32-Team control (at least offline).
Just create a coach,change the scheme and pb and retire. It will stick,even throughout the offseason.

Gesendet von meinem SM-A510F mit Tapatalk

[JohnDoe8865]

Quote: Originally Posted by KershawsCurveball

Changing schemes to alter the AI pb to a generic one does not require 32-Team control (at least offline). Just create a coach,change the scheme and pb and retire. It will stick,even throughout the offseason. Gesendet von meinem SM-A510F mit Tapatalk

What about changing them to custom user playbooks? Doesn't stick?

[LionsFanNJ]

Quote: Originally Posted by JohnDoe8865

What about changing them to custom user playbooks? Doesn't stick?

Custom playbooks aren't accessible in the coach scheme screen. only the default.

[Sgexpat]

great stuff. I would point out though that many NFL teams play for field position rather than 1st down on 3rd and long and do run draws or designed screen plays that may not be expected to get the 1st. It is unrealistic for the 3rd and long book to have no draws / and in most cases no screens (although throwing underneath to the HB on 4 verticals is 'kinda' like a screen, but not really.

[jaiel26]

The simple solution is to add coaches to the roster file like in ncaa '14... You could assign names (like Bill Belichick) scheme (3/4 4/3, Hybrid) , playbook, custom playbook (if you wanted to customize situations), and formation subs prior to starting a cfm #ProblemSovled

[jaiel26]

Plus the formation subs would help with the forever broken auto sub system where RB2 has been useless and always on the trading block.