WEBVTT

1
00:00:00.391 --> 00:00:01.151
Hello, Internet.

2
00:00:01.171 --> 00:00:03.792
This is your captain speaking, Lester Nare.

3
00:00:04.112 --> 00:00:10.874
Today, we have a very special interview episode to end our World Cup coverage.

4
00:00:11.314 --> 00:00:13.615
It has been a phenomenal six weeks.

5
00:00:13.655 --> 00:00:20.197
We've done several segments covering the intersection of science and the world's beautiful game.

6
00:00:20.357 --> 00:00:27.499
And we will be joined by two guests today who are members of FIFA's data science team,

7
00:00:27.999 --> 00:00:37.651
to discuss one of the new visualizations that has been featured very prominently in the 2026 World Cup, Match Momentum.

8
00:00:38.091 --> 00:00:45.320
And over the course of the conversation, we'll get to learn a little bit more about what does it mean to be a data scientist at FIFA?

9
00:00:45.640 --> 00:00:47.743
How does it fit into the larger organization?

10
00:00:48.243 --> 00:00:49.665
What is match momentum?

11
00:00:50.146 --> 00:00:51.188
How did they come up with it?

12
00:00:51.728 --> 00:01:03.866
How do they use all of the complex and different options for data to derive and ultimately get the graphic on our screens as we watch the world's best team compete?

13
00:01:04.567 --> 00:01:05.768
for the World Cup.

14
00:01:06.108 --> 00:01:24.657
And the guests joining us today will be Juan Buso, who is a senior football data scientist based out of Zurich, Switzerland, alongside Aaron Ackerman, who is team lead for football performance analysis out of the United Kingdom.

15
00:01:25.078 --> 00:01:31.261
As always, we are going to talk about the science from the ground up today because this is From First Principles.

16
00:01:47.554 --> 00:01:50.736
Juan, Aaron, thank you so much for joining us today.

17
00:01:50.756 --> 00:01:54.939
I'm super excited about the opportunity to get to have the chance to talk to both of you.

18
00:01:55.620 --> 00:02:02.225
For those who are fans of the pod, we covered many aspects of the World Cup over the last couple of weeks.

19
00:02:02.345 --> 00:02:08.830
And we ended up putting out a short explainer on the match momentum visualization, which was fantastic.

20
00:02:09.450 --> 00:02:16.856
many people found very fascinating but also raised a lot of questions about how you could actually make something like that happen.

21
00:02:16.936 --> 00:02:31.248
So we are very blessed today to have two members of the data science team over at FIFA to help us walk through and better understand how the match momentum visualization works and how that integrates with the rest of the graphics and broadcast team.

22
00:02:31.268 --> 00:02:35.912
So before we dive in, gentlemen, I just wanted you guys to see if you could tell us a little bit about

23
00:02:36.432 --> 00:02:43.343
your role within FIFA and how it sort of works inside the larger ecosystem where there's a lot of moving parts.

24
00:02:43.463 --> 00:02:46.248
Again, graphics, broadcast, live.

25
00:02:46.548 --> 00:02:47.209
It's global.

26
00:02:47.249 --> 00:02:48.431
It's not just the World Cup.

27
00:02:48.691 --> 00:02:51.556
And maybe we can start with you, Juan, and go to Aaron.

28
00:02:52.924 --> 00:02:53.965
Thank you for having us.

29
00:02:54.065 --> 00:02:58.911
It's a great chance to also share with the rest of the world what we do.

30
00:02:59.111 --> 00:03:08.362
I mean, other than the World Cup, I think that it's nice to have a chance to explain a bit further the details of the inner workings of our job.

31
00:03:08.502 --> 00:03:12.146
So, yeah, I'm a data scientist at FIFA.

32
00:03:13.487 --> 00:03:28.591
My job mostly basically relates to trying to convert the principles of football into some statistical and mathematical component or algorithm that allows us to extract objective insights from the match.

33
00:03:43.495 --> 00:03:51.617
bring the, let's say, the most helpful insights to the teams, the fans, and the rest of the audience that we have.

34
00:03:55.407 --> 00:03:56.347
Thanks, Juan.

35
00:03:56.667 --> 00:03:59.248
So, yeah, so my role is slightly different.

36
00:03:59.288 --> 00:04:13.192
So I look after the performance analysis team here at FIFA, and I probably would say in terms of analysis, whether that be data or video, we're probably the closest to the grass and the coaches and the technical staff that we have.

37
00:04:13.252 --> 00:04:19.974
So on a tournament by tournament basis, we always bring a group of football experts in to work alongside us.

38
00:04:20.034 --> 00:04:21.374
So in the World Cup, we have Jurgen Klint.

39
00:04:21.754 --> 00:04:22.475
for example.

40
00:04:23.556 --> 00:04:25.217
We have Michael O'Neill from Northern Ireland.

41
00:04:25.237 --> 00:04:31.663
We have Tobin Heath with us who are working, living and breathing this every day with us to learn as much as we possibly can.

42
00:04:31.763 --> 00:04:41.171
But one of the most fascinating things for us is, and biggest challenges, as you've already experienced, Lester, is how we translate into the...

43
00:04:41.431 --> 00:04:45.432
that into information that can be understood on the most basic level, right?

44
00:04:45.632 --> 00:04:54.814
So that's where we then work with infotainment or stadium entertainment, we call it, or TV broadcasting, for example.

45
00:04:54.874 --> 00:05:03.296
And every year, every few months, we go through the process of identifying potentially new metrics, potentially new stories to tell within the game, right?

46
00:05:03.776 --> 00:05:05.577
And then the challenge becomes, okay,

47
00:05:06.297 --> 00:05:07.937
how do we graphically represent that?

48
00:05:08.097 --> 00:05:18.500
So I work extremely closely with, with Juan, for example, we work in the same team, um, ultimately, and we then come together to go, okay, what do we have within the data?

49
00:05:18.520 --> 00:05:27.322
We have a huge amount of data that both tracking event, skeletal, um, as you can imagine for, for all of our competitions.

50
00:05:27.362 --> 00:05:32.603
And then it's about, okay, well, we know we use the data for this and we can tell stories, uh,

51
00:05:33.164 --> 00:05:41.872
around the data in this way that may be understood by a data scientist or maybe understood by a coach, but how do we bring those two together to really tell a story to the audience?

52
00:05:41.912 --> 00:05:48.478
And that's where I guess momentum has come in and played quite a big kind of TV role over the last six weeks during the World Cup.

53
00:05:50.177 --> 00:05:53.038
That's so fascinating and makes total sense.

54
00:05:53.078 --> 00:06:13.723
I think coming from a background in software for more of a tech perspective, but having been a footballer myself, seeing kind of the intersection of these two worlds is I continue to find fascinating as data has become more and more integrated into the game over the last decade or two.

55
00:06:14.203 --> 00:06:19.044
And so if we kind of dig in here with maybe the first question, which is,

56
00:06:20.906 --> 00:06:34.387
Given that answer, Aaron, you just brought up about storytelling right through data and analytics, what problem are you trying to solve with the concept of match momentum and kind of what motivated the development around it?

57
00:06:35.218 --> 00:06:35.478
Sure.

58
00:06:35.618 --> 00:06:37.859
So I'm happy to give a bash at this.

59
00:06:38.920 --> 00:06:46.282
Look, we want to be able to tell the most honest and truest story of a match that's happening.

60
00:06:46.342 --> 00:06:46.543
Right.

61
00:06:47.143 --> 00:06:50.304
And we know that each team has their own style that they want to play.

62
00:06:50.444 --> 00:06:54.986
We know that teams have nuances and variations of the things that they do on the pitch.

63
00:06:55.526 --> 00:06:59.330
We can measure them in fairly basic ways by counting things, right?

64
00:06:59.770 --> 00:07:04.275
We can measure them based on positionally where players are positioning themselves.

65
00:07:04.335 --> 00:07:09.440
And commentators, for example, and experts on TV will identify these anyway.

66
00:07:10.161 --> 00:07:10.881
But actually,

67
00:07:11.702 --> 00:07:22.785
What is it that we can use to really try and identify and show that, okay, maybe this team doesn't have the ball, but when they do have the ball, they're creating a lot of threat and they are becoming really dangerous, right?

68
00:07:23.306 --> 00:07:31.388
And that's where, you know, myself and Juan work together really closely on trying to take knowledge from the experts about what does it look like?

69
00:07:31.408 --> 00:07:40.951
What does it feel like with regards to momentum, with regards to threats, identify what that potentially looks like in the data and then represent that graphically.

70
00:07:41.331 --> 00:07:52.282
So I know obviously previously on the pod you talked about another kind of data set or data company that have their version.

71
00:07:52.422 --> 00:07:56.766
And our approach was slightly differently, which I'll let Juan explain in a minute.

72
00:07:56.806 --> 00:08:04.133
But one of the key and most important things to us was it's not necessarily just about having the ball in the final third.

73
00:08:05.134 --> 00:08:07.997
And that's probably what most people think about momentum.

74
00:08:08.377 --> 00:08:10.019
You can use two extreme examples.

75
00:08:10.059 --> 00:08:11.440
These are the two examples I always use.

76
00:08:11.821 --> 00:08:16.045
Why is it always that Atletico Madrid win games when they don't dominate the ball?

77
00:08:16.065 --> 00:08:19.408
Well, it's because they create threat in the right moments, right?

78
00:08:19.728 --> 00:08:23.271
As an extreme example, we saw that in this tournament with Carbo Verde.

79
00:08:24.112 --> 00:08:28.135
Why do they, for instance, create great chances without having a lot of the ball?

80
00:08:28.215 --> 00:08:31.136
For instance, against Spain, who completely dominate the ball, right?

81
00:08:31.617 --> 00:08:48.727
And although our model will identify naturally that Spain have a lot of threat because they do, that isn't to say that at no point other than when Spain don't have the ball that then, for instance, Cabo Verde or in lots of cases, Atletico Madrid have the ball, right?

82
00:08:49.588 --> 00:09:03.300
That was a huge important piece for us to try and overcome that was beyond the basic kind of looking way of looking at it where it's just, if you have the ball in the final third, therefore you control the greatest amount of chances.

83
00:09:03.540 --> 00:09:06.443
That is true in most parts, but it isn't always true.

84
00:09:06.503 --> 00:09:10.567
And how do we make sure that we can represent as close to the truth as possible, Juan?

85
00:09:11.788 --> 00:09:15.771
Yeah, I think that you explained quite well that principle, I think that the,

86
00:09:16.912 --> 00:09:30.442
The idea was to have a metric that can tell a story and include the most contextual information about football that we can find and we have in our data sets, let's say.

87
00:09:30.582 --> 00:09:35.986
So, yeah, momentum was derived from another metric that is called threat.

88
00:09:36.366 --> 00:09:40.429
And this metric, you could think of it like a bit of a layer cake with different layers.

89
00:09:41.030 --> 00:09:54.108
One of them is like a kinetic pitch control that takes into consideration the acceleration and speed of the players and the direction they're moving in order to calculate on the pitch where the control from different teams is.

90
00:09:54.989 --> 00:09:56.811
Then on top of that, we have another layer.

91
00:09:58.633 --> 00:10:16.684
pertaining the ball and this is also with the speed, direction, and height of the ball to see you calculate the trajectory and then you can know if a player would intercept the ball if it would go out of bounds or the ball would go to a teammate and then create more threat.

92
00:10:17.324 --> 00:10:40.199
Then we have another layer that is basically the pitch, she would say, danger underneath, which is basically not just how close you are to the opponent's goal frame, but also how the setting is of the opponent, like the distance between the defenders and the goal frame, the goalkeeper and the goal frame, the attackers and the defenders, and all these interactions, which is quite rich and very dynamic.

93
00:10:40.520 --> 00:10:43.702
Football, I think that this is one of the things that makes it so...

94
00:10:44.442 --> 00:10:55.050
interesting is that there is so much context and so much happening at the same time that brings these challenges for us to represent the metrics, but also makes it interesting.

95
00:10:55.590 --> 00:11:03.656
And then we have another layer on top of that, that is basically the side of the players looking at how clear is the view of the call frame.

96
00:11:04.096 --> 00:11:13.843
So if they have a direct view or there are some players in between and how far, you know, because a player on your face will block much more than a player 20 minutes away.

97
00:11:14.484 --> 00:11:20.874
And then we bring all of this together and we have a threat calculation for each frame of the match.

98
00:11:21.695 --> 00:11:23.939
And then momentum, basically what it does is

99
00:11:25.161 --> 00:11:28.103
feeds on this threat and metric.

100
00:11:28.603 --> 00:11:38.628
And depending on the amount of attacks, the length of the attacks and the strength of these attacks or the threat of these attacks will spike or go down.

101
00:11:38.668 --> 00:11:45.732
The more that you see, the more frequent, the longer, let's say the more threatening these attacks are, the metric will go up.

102
00:11:46.392 --> 00:11:51.295
And then when the team basically subsides a bit, then it starts to slow down and so on.

103
00:11:51.715 --> 00:12:00.200
So it's a very, very contextual metric that in a very short glimpse of time allows you to have the context of the history of the match.

104
00:12:00.380 --> 00:12:03.121
And I think that that was the entire purpose of this.

105
00:12:03.241 --> 00:12:08.464
Try to tell a story in one graph of what has been happening until this moment in the match.

106
00:12:10.038 --> 00:12:10.959
This is fascinating.

107
00:12:10.999 --> 00:12:16.503
So if I'm understanding you both correctly, there's almost like two layers to this, right?

108
00:12:16.804 --> 00:12:23.629
The initial layer is taking, you know, multiple pieces of context, which is being identified as the threat metric.

109
00:12:24.750 --> 00:12:32.376
And that's being taken every frame, meaning basically every some time period, you know, 500 milliseconds a second, what it might be.

110
00:12:33.181 --> 00:12:35.343
It's 50 hertz, the tracking data that we have.

111
00:12:35.823 --> 00:12:36.143
50 hertz.

112
00:12:36.203 --> 00:12:36.403
Okay.

113
00:12:36.864 --> 00:12:37.404
Incredible.

114
00:12:37.444 --> 00:12:52.455
And then so you use that as sort of the base, and then that change over time and sort of basically a little bit of an algorithm on top of that is how we then create what we sort of define as momentum, which is a combination of all these contextual layers combined.

115
00:12:52.636 --> 00:12:53.516
Yeah, exactly.

116
00:12:53.796 --> 00:12:57.279
And how they are sort of ebb and flow over a time period.

117
00:12:57.299 --> 00:12:57.599
That is...

118
00:12:58.760 --> 00:12:59.663
That is very interesting.

119
00:12:59.703 --> 00:13:04.395
I think one of the things, Juan, especially you brought up there as well is –

120
00:13:06.395 --> 00:13:10.458
that there are so many points by which the data can be captured.

121
00:13:10.518 --> 00:13:24.668
So can you guys actually talk a little bit more about between passes, player movements, other match events, which have to be tracked in real time, which is an incredible processing challenge, independent of just the actual data analysis challenge.

122
00:13:24.968 --> 00:13:26.489
But is it computer vision?

123
00:13:26.529 --> 00:13:28.911
Is it optical sensor data?

124
00:13:28.971 --> 00:13:33.474
How does this sort of mesh of sensors really work together?

125
00:13:34.282 --> 00:13:38.745
So I'll give you for momentum, first and foremost, it's purely tracking data.

126
00:13:39.206 --> 00:13:46.211
There is no event data in it whatsoever, which I think, again, it's really cool in its own reason.

127
00:13:46.271 --> 00:13:46.451
Right.

128
00:13:46.511 --> 00:13:51.895
Because the only information that we know from that data is who has the ball at every frame.

129
00:13:52.575 --> 00:13:56.859
Outside of that, we just have the player location 50 times a second.

130
00:13:58.419 --> 00:14:07.209
But when it comes to the other kind of data ecosystem or the full data ecosystem that we have, we have a joint venture with Hawkeye, for example.

131
00:14:07.690 --> 00:14:15.880
So when it comes to TV, what you will see is all of our data and our data model that we've built over the last couple of years are...

132
00:14:16.320 --> 00:14:21.124
which is a lot different to most other providers, for example, but then that's automated.

133
00:14:21.504 --> 00:14:27.629
So that is automated through a number of computer vision processes and algorithms to identify.

134
00:14:27.649 --> 00:14:35.755
A lot of machine learning has gone into that, to identify what the events actually look like and then to automatically categorize them.

135
00:14:37.176 --> 00:14:40.418
So Juan, I don't know whether you want to give a little bit more detail on that process.

136
00:14:40.919 --> 00:14:45.042
Yeah, so as I mentioned, we have, I would say,

137
00:14:46.929 --> 00:14:48.930
four different sources of data.

138
00:14:49.030 --> 00:15:04.875
One is the tracking data from the players that, as he mentioned, comes every 20 milliseconds and provides a lot of context and information just by looking at how the dynamic of the players is on the pitch.

139
00:15:05.395 --> 00:15:14.018
Then we have the event data that it's also, we have a version live and then we have a version postmatch that it's,

140
00:15:15.178 --> 00:15:23.671
basically scrapped and cleaner and following what we have developed, which is the FIFA football language, which is what Aaron hinted before.

141
00:15:24.652 --> 00:15:30.301
And this data is extremely precise to the frame where the players make contact with the ball or...

142
00:15:31.203 --> 00:15:35.804
perform a particular action and very detailed also of how these events are.

143
00:15:36.204 --> 00:15:42.845
Basically, not only if it was a left or right foot, but what kind of pass, who was the receiver, how and etc.

144
00:15:42.865 --> 00:15:47.766
So each of the events has its own set of traits and characteristics.

145
00:15:48.326 --> 00:15:56.128
We also have a lot of limb tracking data that it's also used for other components in football, like VR, for example.

146
00:15:57.048 --> 00:16:03.093
And then we also have the ball data, which is very, very high resolution.

147
00:16:03.694 --> 00:16:09.919
It's 500 frames per second, and we get a lot of information from it.

148
00:16:10.419 --> 00:16:21.468
It's a very complex data set because it has a lot of, let's say, components, but it's very enriching as well.

149
00:16:21.708 --> 00:16:25.551
Not only the ball position, but also like the rotational accelerometers, etc.

150
00:16:27.954 --> 00:16:29.797
So those are the main data sets.

151
00:16:30.177 --> 00:16:39.329
But as Aaron mentioned, for Momentum, we focus mostly on the, I mean, only on the tracking data, because that's where we get all that context of the players.

152
00:16:39.369 --> 00:16:40.150
What are they doing?

153
00:16:40.190 --> 00:16:42.133
How are they running, interacting, and so on.

154
00:16:44.638 --> 00:17:01.204
That's very, I mean, there's so many interesting pieces here because I think there's sort of two separate conversations here, right, which is the entire pipeline of data that informs a variety of different end products, real time and post match is quite vast.

155
00:17:02.285 --> 00:17:11.428
But within that context, the match momentum feature is actually a very small, just a very narrow set of the available context there.

156
00:17:13.289 --> 00:17:15.574
which it's so interesting.

157
00:17:15.854 --> 00:17:21.025
And so when, you know, I kind of want to dig in a little bit about this idea of momentum.

158
00:17:22.127 --> 00:17:23.409
And why is it that

159
00:17:24.970 --> 00:17:33.335
Given the vast array of data that's available, having a confidence level that that minimum data set is the correct approach?

160
00:17:33.355 --> 00:17:36.337
Can you help me understand the philosophy behind that a little bit, if that makes sense?

161
00:17:37.138 --> 00:17:44.723
I'm going to come back at you for a second because I think if we simplify it for a moment, we have all of this...

162
00:17:45.483 --> 00:17:51.167
data, which gives us loads of different other opportunities.

163
00:17:51.647 --> 00:17:54.329
But the reality is, is those are very new things, right?

164
00:17:54.590 --> 00:17:56.311
So the ball data is very new.

165
00:17:56.331 --> 00:17:57.812
The limb tracking is very new.

166
00:17:59.133 --> 00:18:03.356
It's very heavy in terms of its size and its processing rate.

167
00:18:03.856 --> 00:18:19.224
Which is also, that's not an issue in many ways, but the big thing for us is the most telling when it comes to momentum and dynamics of the game actually is not how fast the ball spins or the location of the ball or where it goes to.

168
00:18:19.604 --> 00:18:23.326
That is an outcome of what the players decide to do.

169
00:18:23.866 --> 00:18:28.369
Now, the next iteration may be that limb tracking supports players

170
00:18:28.949 --> 00:18:32.511
movement, it supports body position, it supports body orientation.

171
00:18:32.931 --> 00:18:46.779
But the reality is, is the dynamics are all in players' movements, which positions they hold, the spacing between each other, the spacing between them and the opposition, their reaction, their timing, the real dynamics of the game.

172
00:18:46.839 --> 00:18:57.565
So that's why we're so confident that just with this single data source that we get the majority of what we need to really understand how we define momentum, right?

173
00:18:57.585 --> 00:18:57.745
Right.

174
00:18:59.938 --> 00:19:01.299
No, that makes a lot of sense.

175
00:19:01.339 --> 00:19:17.488
And so maybe another interesting question to kind of touch on this is you brought it up earlier where simple possession, like the idea of simple possession as a value, is unique and distinct from how momentum is defined here.

176
00:19:18.009 --> 00:19:24.733
And quantitatively speaking, when we say momentum as unique from possession, like what are we really like?

177
00:19:24.873 --> 00:19:26.033
I just want to dig deeper on that.

178
00:19:26.313 --> 00:19:28.755
What are we saying in the difference there in the storytelling?

179
00:19:28.775 --> 00:19:28.895
Yeah.

180
00:19:30.926 --> 00:19:32.551
Juan, do you want to go or I'm happy to go?

181
00:19:33.252 --> 00:19:33.473
Go.

182
00:19:33.674 --> 00:19:34.917
I'll jump later.

183
00:19:36.181 --> 00:19:36.521
Sure.

184
00:19:36.702 --> 00:19:37.003
So...

185
00:19:38.182 --> 00:19:49.510
quantitatively, what we're saying is that we already have a good understanding of what good looks like, let's say, right?

186
00:19:50.390 --> 00:19:54.133
Because we have a lot of data that shows us what good looks like.

187
00:19:54.333 --> 00:19:59.076
So we work backwards from what good looks like, right?

188
00:19:59.096 --> 00:20:01.758
And we work backwards or in a sense that

189
00:20:02.538 --> 00:20:11.905
just because you don't have the ball or just because you're not close to the goal necessarily in this moment, you do not have a potential threat on the opposition.

190
00:20:13.046 --> 00:20:16.068
So quantitatively, you have a really high line.

191
00:20:16.769 --> 00:20:21.632
You have a player moving at an acceleration pace of X, Y.

192
00:20:22.973 --> 00:20:34.555
which we know is getting closer to goal at a certain rate, which we know causes a level of threat, that you also see a dispersion of the other player.

193
00:20:34.595 --> 00:20:42.917
So quantitatively, we know that all of these movements are creating space and space is what ultimately creates the greatest amount of threat.

194
00:20:42.937 --> 00:20:50.458
And then when you take space creation closer to goal at a speed, that is when you get the most threatening moments.

195
00:20:51.058 --> 00:20:52.978
forget about the ball for a second.

196
00:20:53.158 --> 00:20:54.539
That is ultimately what you then have.

197
00:20:54.739 --> 00:21:08.161
So when you are then without the ball, you're sat in a really deep position, for example, and you have multiple players going and pressing the opposition who, yes, we have a one to say that that opposition team are in possession of the ball.

198
00:21:08.701 --> 00:21:17.543
We quantitatively know the speed at which they are closing that space down and therefore reducing the threat of opposition, but also increasing their own threat.

199
00:21:17.803 --> 00:21:19.003
Should they then get the ball?

200
00:21:19.883 --> 00:21:21.925
All of a sudden, okay, bang, there you go.

201
00:21:22.085 --> 00:21:22.305
Right?

202
00:21:22.706 --> 00:21:33.697
So if you, if you don't take all of those things into consideration, what you just have is you have possession based metrics, which is not what momentum is.

203
00:21:33.757 --> 00:21:37.140
Momentum team having a momentum isn't a possession based metric.

204
00:21:39.622 --> 00:22:01.437
And so on a follow-up, you know, as you guys have been going through this development process, right, around kind of honing in, you know, to really make sure that the match momentum reflects the flow of the game, you know, as you went through that process, was there anything surprising or something that you didn't expect as you worked through that process?

205
00:22:02.858 --> 00:22:06.700
Yeah, Juan, I'll give you some of the surprising things that we found.

206
00:22:06.720 --> 00:22:07.681
Yeah.

207
00:22:09.253 --> 00:22:19.941
To be honest, I think that when we were developing these principles, it was very, I would say, for me, interesting to see.

208
00:22:20.241 --> 00:22:25.245
My background is in biology, so I come from a place where there are no boxes, basically.

209
00:22:25.285 --> 00:22:30.089
So everything is in a distribution, everything is kind of their probability.

210
00:22:30.109 --> 00:22:36.253
So this is a bit, for me, the surprise that in football, like...

211
00:22:37.154 --> 00:22:53.450
bringing this into football and trying to transform the box concept into these probabilities or basically continuous distributions generated a lot of interesting situations, outcomes during the validation process where the...

212
00:22:55.812 --> 00:23:04.220
there would be a high, let's say, threat value or high momentum, and then the validators would go and check, and they would be like, oh, true, now it makes sense.

213
00:23:04.540 --> 00:23:20.875
But it would be like something that sometimes situations would pop up that were not perceived before, and when you actually look at that, and because of what Aaron said, that you have all these contexts with the tracking data of creating space and the moving players,

214
00:23:21.455 --> 00:23:27.357
then all these things come to life and all the things come to, let's say, resurface.

215
00:23:28.077 --> 00:23:31.738
So I think that there was a lot of situations in that sense.

216
00:23:32.118 --> 00:23:45.061
And also in the other sense, I would say that it took some tweaking in order to see, okay, so this moment was considered, let's say, more threatening than what the metric actually was, what was happening.

217
00:23:45.681 --> 00:23:51.047
So again, having the discussions and trying to incorporate, OK, what do you see footballistically here?

218
00:23:51.547 --> 00:23:57.833
And how can you transform that into a mathematical model that basically reflects the perception?

219
00:23:58.174 --> 00:24:04.740
And I think one of the other things that I think we both found was quite cool was...

220
00:24:05.821 --> 00:24:08.303
actually identifying the value of being offside, right?

221
00:24:08.944 --> 00:24:10.525
That was a huge discussion.

222
00:24:10.645 --> 00:24:12.726
So, you know, you're holding a position offside.

223
00:24:13.107 --> 00:24:17.630
Well, technically, you receive the ball and you are not threatening in that moment.

224
00:24:17.690 --> 00:24:24.035
But actually, holding your position offside for a certain amount of period of time does hold some weight to the value of threat.

225
00:24:24.095 --> 00:24:25.376
Because, again, this...

226
00:24:26.076 --> 00:24:34.778
how smart players are in terms of using the offside to then create space for themselves, to then come onside, receive the ball, and then all of a sudden the threat goes up.

227
00:24:34.818 --> 00:24:45.781
So, you know, taking some of those things into consideration was, you know, was a huge thing to understand and overcome and, you know, discuss that with football experts in terms of,

228
00:24:46.721 --> 00:24:50.947
what they're coaching their players when they're talking about how do they use the offside.

229
00:24:50.967 --> 00:25:01.982
So there's so many fascinating things that just understanding space and valuing space and movement was just fascinating in itself.

230
00:25:03.532 --> 00:25:26.614
And I think we've seen even in the discussions around the games, even as we've gotten deeper into the knockout stages, these discussions from football fans, when it passes the eye test, when you see what looks like and feels like to be, oh, the shifts in the game and the teams that are able to identify spaces in ways that you sometimes we at home are like, how did...

231
00:25:27.315 --> 00:25:30.939
you know, X, C, Y in this fashion or in this way.

232
00:25:31.460 --> 00:25:40.509
And it's just, it's really fascinating to hear you discuss what is really a blend of sort of the art of football, right?

233
00:25:40.710 --> 00:25:44.614
And the science of data and finding a middle ground between those two.

234
00:25:45.194 --> 00:25:46.476
Because it is, you know,

235
00:25:48.493 --> 00:25:51.455
there's such a long storied history, right, in this sport.

236
00:25:52.215 --> 00:25:55.377
And there's sort of our ebbs and flows and changes tactically.

237
00:25:55.817 --> 00:26:00.980
You know, there's eras where the way in which the game is played changes.

238
00:26:01.120 --> 00:26:09.305
And I think we're sort of in a very fascinating era right now for the rise of so many of these different analytics, such as match momentum.

239
00:26:09.325 --> 00:26:13.868
I mean, just looking at it as an observer on TV at home, for example,

240
00:26:14.944 --> 00:26:18.845
The momentum, like it passes the eye test when we look at it from home, right?

241
00:26:19.705 --> 00:26:36.890
And so there's clearly the validation process and the work that you guys have done has gotten to a place where folks who have been, you know, football watchers of any league, you know, for some period of time can kind of intuitively see that there's this feels, has this feeling that's correct.

242
00:26:37.370 --> 00:26:39.671
I will ask just a few more questions here.

243
00:26:40.751 --> 00:26:54.017
In that same vein, there are aspects of the game, you know, that are new that it would be interesting to get your take on how you may have seen it affect match momentum.

244
00:26:54.437 --> 00:27:07.743
Most prominent of which is the introduction of hydration breaks during the World Cup specifically, because correct me if I'm wrong, the match momentum has existed prior to the World Cup.

245
00:27:07.783 --> 00:27:09.964
It just happened to be featuring prominently at the World Cup.

246
00:27:09.984 --> 00:27:10.464
Is that correct?

247
00:27:12.303 --> 00:27:24.600
So did you notice, you know, as you've been looking through, has there been any differences between the non-hydration break data sets versus the hydration break data sets as it relates to match momentum?

248
00:27:24.640 --> 00:27:27.944
Or is it not really the right way to think about this metric?

249
00:27:28.605 --> 00:27:33.028
Yeah, no, there are some signs.

250
00:27:33.488 --> 00:27:35.550
I would say we haven't fully analysed it.

251
00:27:35.590 --> 00:27:38.572
It's one of the questions that we've been asked to analyse and we will do so.

252
00:27:39.252 --> 00:27:45.497
But at the moment, I wouldn't say necessarily that momentum necessarily reflects...

253
00:27:46.277 --> 00:27:47.458
any big changes.

254
00:27:47.978 --> 00:27:58.124
For sure, the reasons why we're talking about them is because of the tactical decisions the coaches are making as a result of having those hydration breaks.

255
00:27:58.465 --> 00:28:06.870
But it's really hard, especially, for instance, when a team already has momentum and they further increase their momentum, right?

256
00:28:08.205 --> 00:28:20.648
What does that actually mean and how do you actually measure that versus when you see a momentum switch as a result of the hydration break?

257
00:28:20.708 --> 00:28:31.411
So I think first we have to really understand what we're trying to use momentum to measure in that context before we then can go, okay, well, this is...

258
00:28:32.551 --> 00:28:39.655
this is the consequence or this is what's happened as a result of the hydration break on momentum.

259
00:28:39.855 --> 00:28:47.920
I think it's actually the hydration breaks are more complex than we actually probably realize just watching it.

260
00:28:48.640 --> 00:28:58.751
And what it does to teams both psychologically, tactically, and also the state of the game that you're in when that hydration break comes, right?

261
00:28:59.832 --> 00:29:04.537
There are so many different scenarios that you can use those hydration breaks to then overcome.

262
00:29:04.918 --> 00:29:09.703
And I wouldn't say match momentum is the single thing that necessarily will reflect that.

263
00:29:11.827 --> 00:29:13.187
That's fascinating.

264
00:29:13.387 --> 00:29:31.093
And it is, you know, as we see sort of the tacticians in managers, it's been interesting to see how managers have utilized the introduction of that and how quickly sort of, you know, strategy around using those to...

265
00:29:32.653 --> 00:29:56.896
approach the games differently and having that opportunity to have a reset again whether it's psychologically whether it's tactically and so i would be interested to see how that continues to evolve both from how do you how do you look at it from an analytics perspective and what story is it telling um and you know does it continue to exist in the same format you know or does this

266
00:29:58.237 --> 00:30:05.562
does this become something that's perceived after the data is a little bit more digested, that its impact on the game is outsized in a way that's

267
00:30:06.709 --> 00:30:09.710
maybe not intended, but has become the reality in its implementation.

268
00:30:10.751 --> 00:30:24.977
Yeah, I think there's no indication that we will see hydration breaks again in the near future because of the way that we measure the reasons why we have a hydration break, right?

269
00:30:25.177 --> 00:30:30.779
It's pretty complex, the measurement, the calculation for the reasons why we have them.

270
00:30:31.180 --> 00:30:31.320
Yeah.

271
00:30:32.320 --> 00:30:36.803
So I think the best way for us to actually look at it is in our next competition.

272
00:30:37.544 --> 00:30:37.764
Right.

273
00:30:37.824 --> 00:30:46.850
So you understanding it once our next competitions have, when our next competition has happened, actually looking back, what impact did they have?

274
00:30:46.910 --> 00:30:47.951
Because the other thing that...

275
00:30:49.125 --> 00:30:56.171
We obviously had coming into the tournament where we didn't have hydration breaks or, you know, we did have them in the Club World Cup, but only on certain matches.

276
00:30:56.651 --> 00:31:01.154
There wasn't a clear kind of principle around tactically how we can use them.

277
00:31:01.955 --> 00:31:11.903
So, you know, we've got to maybe a stage where everybody has some form of tactical decision to make during or uses that opportunity to make a tactical decision.

278
00:31:13.084 --> 00:31:17.747
And now we'll go back to not having that opportunity or the same type of opportunity.

279
00:31:17.808 --> 00:31:19.909
So I think it's probably in that way.

280
00:31:20.149 --> 00:31:23.712
Previously, the tactical decisions were based on substitutions, right?

281
00:31:24.393 --> 00:31:30.738
Now you've got an extra layer that you can use in the current tournament to make some tactical changes.

282
00:31:33.297 --> 00:31:45.382
I want to end on sort of a forward look here because I think you guys really helped, I think, us understand some of the philosophy behind what is the story that Match Momentum is trying to tell.

283
00:31:46.422 --> 00:31:53.805
The layers that are used to structure it as independent from threat, as an example, or the wider context.

284
00:31:54.125 --> 00:32:16.146
tool set and data sets that are available which again the like data processing aspects of this make my head hurt just even thinking about it with that number of different sources and doing it in real time So that that's a whole separate conversation But as we look forward, you know, and I think Aaron you were just mentioning this, you know, there are a variety of new and

285
00:32:17.107 --> 00:32:22.852
sources of data, the limb tracking, you know, the introduction of ball tracking.

286
00:32:23.793 --> 00:32:29.819
The part of the mandate for the team is to continue to find ways to tell different stories about the game using data.

287
00:32:29.839 --> 00:32:32.821
Do you guys have any thoughts about, as you look forward now,

288
00:32:33.903 --> 00:32:47.266
with match momentum and you know seeming to be a success other things you're thinking about if you have brainstorms or ideas kind of on the dartboard or just even sort of a wish list of as we look forward this would be great if we could start to do these types of things

289
00:32:49.252 --> 00:32:49.672
Should I go?

290
00:32:49.772 --> 00:32:50.872
Juan, I'll let you go first.

291
00:32:51.313 --> 00:33:14.839
Okay, so I think that one important pillar that we are working on and we have already some work, it's just that we want to further develop things before sharing with the rest of the world, is the analysis of the physical data and trying to develop metrics that reflect a bit more the physical demands of the game.

292
00:33:14.879 --> 00:33:16.580
I think that in football there is a

293
00:33:17.560 --> 00:33:24.229
I would say a conservative approach that has been applied for a certain period of time, which is good and functional.

294
00:33:25.170 --> 00:33:32.419
But I think that it leaves out of the table a lot of the effort that the players are putting on the pitch and, um,

295
00:33:32.839 --> 00:33:50.467
Sometimes, because reaching a particular spread threshold is not how the particular context of the match allows you to perform, then this performance is undervalued or sometimes not included in the effort, but players do a lot of effort.

296
00:33:51.108 --> 00:33:55.530
So I would say that there is a new wave of metrics or

297
00:33:56.971 --> 00:34:12.364
that would come in this direction in order to try to reflect what the actual physical effort or approximate more with a more appropriate proxy, the physical efforts that the players are putting on the pitch and how much work they are actually doing.

298
00:34:12.805 --> 00:34:20.812
And sometimes we are not perceiving it because you only get in your head the players that are running at top speed and you're like, wow, that was a fantastic run.

299
00:34:21.372 --> 00:34:25.493
But there is a lot of work in the very small, short distances that they are doing.

300
00:34:25.953 --> 00:34:33.434
And that's what we want to try to reflect and bring the audiences to basically to value as much.

301
00:34:34.494 --> 00:34:49.217
So, for example, Lester, one of the things that we've been working on is actually the value of the energy that players contribute and how that actually links with the threat that they then support with the team.

302
00:34:49.917 --> 00:35:05.094
So, for example, to use Juan's example there, we see a lot of the times that people talking about, OK, maybe Kylian Mbappe's speed or Lise's speed or, you know, these players who are super, super quick, who are extremely dangerous.

303
00:35:05.334 --> 00:35:07.016
We know this, right?

304
00:35:07.356 --> 00:35:07.656
Right.

305
00:35:08.297 --> 00:35:10.958
or it may be the total volume that a midfielder runs.

306
00:35:11.138 --> 00:35:14.999
Everybody talks about 12K for a player, right?

307
00:35:16.200 --> 00:35:18.301
A World Cup match, for example.

308
00:35:18.641 --> 00:35:25.963
But the reality is of what the 12K for the left centre midfielder versus the right centre midfielder looks completely different.

309
00:35:27.404 --> 00:35:34.007
based on their role and the actually the moments that they're putting all of their energy into that is completely different.

310
00:35:34.267 --> 00:35:42.711
Or you have vice versa, is a centre back, okay, maybe only runs 10K, but actually he puts in more energy and effort than somebody that does run 12K, right?

311
00:35:43.271 --> 00:35:50.975
And being able to, okay, we have this information now, myself, I'm kind of really confident that we've really got our heads around this and we've come up with some really nice metrics to

312
00:35:51.555 --> 00:35:56.776
to tell this story to maybe the scientific audience and be able to measure that in a way that makes sense.

313
00:35:56.836 --> 00:36:03.257
But now it's a case of, OK, how do we bring that to life for coaches, for fans, for TV, for example?

314
00:36:03.277 --> 00:36:06.538
But I think for us, that will be the next frontier.

315
00:36:06.618 --> 00:36:16.820
And, you know, what you may see in the Women's World Cup next year is a lot more information around energy that players are putting down on the pitch, which I think is really fascinating and cool in itself, right?

316
00:36:18.184 --> 00:36:30.771
if as a former number six, uh, who was doing, who's doing 12 K every game, I, I can appreciate, um, you know, your point just about, um, um,

317
00:36:31.557 --> 00:36:36.602
there are moments of explosiveness, there are moments of top, how frequently are you reaching top speed?

318
00:36:36.622 --> 00:36:44.730
So just because you have the same total distance, how you're managing each component part of that total distance is a very different thing.

319
00:36:45.892 --> 00:36:49.455
And that's such a really important point here.

320
00:36:49.475 --> 00:36:49.696
And I'm,

321
00:36:50.276 --> 00:37:17.428
really looking forward to seeing what you all do with that because I think there's, you know, especially because we live in a world of connected devices, streaming, mobile applications, there's a variety of end destinations for this data that may not necessarily be in the broadcast but can be a part of this larger ecosystem of how people engage with the game and there's a lot of different opportunities and contexts by which to provide that.

322
00:37:17.448 --> 00:37:17.608
I know

323
00:37:18.668 --> 00:37:20.229
a culture of we want to watch the game.

324
00:37:20.249 --> 00:37:25.112
We don't want all the extra stuff, but I think there are, there there's a time and place for everything.

325
00:37:25.232 --> 00:37:28.334
And that changes based on geographical audiences as well.

326
00:37:28.694 --> 00:37:33.177
For example, the U S audience this year has been craving it because they're so used to it.

327
00:37:33.557 --> 00:37:37.039
When you look at the NFL or you look at the NHL or the NBA, right?

328
00:37:37.079 --> 00:37:40.862
So I think with every region that we go to with the world cup,

329
00:37:41.482 --> 00:37:44.771
the demands and the appetite for this information changes.

330
00:37:44.811 --> 00:37:50.304
And naturally, as the years go on, people's appetite for this information is constantly increasing anyway, right?

331
00:37:51.927 --> 00:37:57.028
100%, which is, as you can see from the level of engagement on the short we put out on Match Momentum.

332
00:37:57.128 --> 00:38:04.449
I mean, people were very curious not only about it in general, but understanding how it works.

333
00:38:04.669 --> 00:38:16.211
And I think going into next year's World Cup, which, again, we, you know, now that we have an open line of communication, would love to continue the discussion as you continue to put some of these concepts out.

334
00:38:16.491 --> 00:38:18.832
Because I think there is an audience that is hungry for...

335
00:38:19.672 --> 00:38:27.154
you know, a bridge between a general audience and a little bit more of a technical understanding of the work that gets, you know, done behind the scenes here.

336
00:38:27.574 --> 00:38:32.515
And I'm grateful to both you, Aaron and Juan, for joining us today to have this discussion.

337
00:38:32.535 --> 00:38:35.435
I mean, it's, the beautiful game is continuing to grow.

338
00:38:36.196 --> 00:38:38.076
It's already the world's most popular sport.

339
00:38:38.156 --> 00:38:39.336
It will continue to be so.

340
00:38:39.836 --> 00:38:47.498
And I'm looking forward to you and your teams continuing to push the envelope to help us better understand and tell different types of stories.

341
00:38:48.078 --> 00:38:51.522
about this game that means so much to so many people on the planet.

342
00:38:53.023 --> 00:38:54.765
No, it's been an absolute pleasure, Lester.

343
00:38:54.825 --> 00:38:59.050
Thank you for having us first and foremost, and enjoy the remainder of the games.

344
00:39:01.911 --> 00:39:02.231
Thanks.

345
00:39:02.471 --> 00:39:17.659
And also I really appreciate the platform to also be able to share with the rest of the world as well, like the intricate work and concepts that go behind the things that we try in a simple line, try to show.

346
00:39:17.719 --> 00:39:19.920
But there is a lot of thought and process behind it.

347
00:39:21.041 --> 00:39:21.681
Absolutely.

348
00:39:21.781 --> 00:39:25.423
We'll be having both of you back here hopefully soon.

349
00:39:25.443 --> 00:39:26.944
Thank you, gentlemen, so much for your time today.

350
00:39:27.965 --> 00:39:28.485
Thanks, Esther.