Franco De Bonis (00:00:05.26)
We’re going to talk about the industry’s dirty secret. We’ve been working on this for the last 10 years in AI. That’s since the beginning of time. It’s a secret they work really, really hard. This is providers work really hard to hide, and they jump through a lot of hoops to give numbers and stats that make sense, but the reality is a little bit different. So firstly, what we do is we deliver Visual-AI, also known as computer vision. And the way we use it is to expose and extract insights that are locked into visual media. So that’s brand data, text, objects that are burned into the visual media, and we extract that, which then allows that data to be interrogated and manipulated and sliced up. We don’t deliver these services. We simply extract the data that allows those companies to deliver the best outcomes. Okay? So the last five years, we focused on sports sponsorship, and we developed a tech stack specifically designed. So this is not some Somebody else’s tech stack, and we’ve manipulated it. This is developed by us to deliver the best results in this area. Our key partner is Vision Insights.
Franco De Bonis (00:01:25.20)
We’re going to hear from Zaheer later on how they make use of this within their platform and their service. In the last three years, we’ve basically been processing every major US sport, a whole bunch of fringe sports, smaller sports, and international sports. What I’m talking about here is not a theoretical thing. It’s not like, Well, we think all this. This is stacked up against other data that you will see. It’s about delivering that meaningful, accurate data. If we look at any sponsorship campaign, it’s made up of a number of elements. Within that campaign, this is just some of them. Within that, we see the ones in red here are the ones where our tech has a role to play. And that role is about measuring the exposure, not just the impressions, but the overall exposure within that. So a lot of people They talk about exposure and impressions as being not very meaningful. They discount it. On its own, I agree. On its own, impressions and exposure data is not super relevant. But as part of the overall chain of data that you get that takes you from the spend side all the way over to the ROI side, it is absolutely critical that that core data is accurate.
Franco De Bonis (00:02:54.08)
The decisions made at the end of the day are talking about millions of of Euro, millions of dollars that may be spent on the sponsorship. You need to make sure that that decision makes sense. What is this secret that we’re talking about? The bottom line is not all the data is being processed. So when they are reviewing videos and they’re extracting those insights, they are not looking at all the data. And that means it’s not really accurate, it’s based on assumptions and inferences, and it’s not transparent. Now, I’m going to take you through this journey so we understand why we say that. So when we look at this here, an average video is 30 frames a second. In some cases, it may be 60. This is what it looks like when you watch that video at one frame per second. You’re seeing a jumpy stuff here, and this is very fluid. Now, what does that translate to? Well, hang on, let me click the button again. What it translates to is this. You take the first frame, you throw away the rest. You don’t look at it. So, no, ignoring that. Then you take the next frame of the next second and you review that and you throw away the rest.
Franco De Bonis (00:04:12.16)
You’re not looking at every frame. You’re looking at just one single frame from each second, and you throw away the rest. Now, that’s the majority of cases, it will be one frame per second. In some cases, it might be two or three. In other cases, it might be less than one frame per second. So We’re getting less data. Why is this an issue? Why is it a problem? What you see here is a short four-second clip, processed at one frame per second, and we’re looking at Spectrum brand. In this example, processing at one frame per second, you get 2 seconds of exposure. But if you go at 30 frames a second, you see 2. 36 seconds of exposure. That’s an 18% difference. Okay? It might not sound like a lot, but 18%, when you add it all together, starts to become critical. In this other example, same thing. We see a 36% difference in the exposure, plus 36%. It’s not always a plus, it’s not always positive. In some cases, it’s less. In this MLB example of something quite static on a home plate, here it’s reporting 5 seconds, and down below, we would see 4.
Franco De Bonis (00:05:27.28)
43 seconds, which minus 11%. Now, this is just three short clips, three short examples, where we see an average variance of 33% and a swing of 11. 4 to 36%. What we see in an average 2-3 hour video is the swing can be anywhere from minus 25 to plus 25. That’s a big difference. When you take it across a whole season and you start making important decisions on that, it can be quite critical. So I want to do a thought experiment here. I don’t know how many of you are from brands or even rights holders, but if you’re running a campaign, I don’t have an answer to this, but if you’re running a campaign and you spend 2 million and you see your intent to buy go up by 10%, and you make a 5 million increase in sales. I don’t know if that’s a great campaign or not. That’s not for me to argue. I’m just picking numbers. How would that decision, your future decisions on a sponsorship change, if your AVE was one of these three numbers? How would your decision change? If the answer to that is, Yeah, I might make a different decision, whether to spend less, spend more, what I would do, how I would activate it, then the answer is clear.
Franco De Bonis (00:06:51.20)
Accurate data is really, really important because that’s the swing that we’re talking about. So if there is a better way, why don’t they do it? Why don’t they implement that? Well, it comes down to three things: scale, an AI that’s not really built for purpose, all in order to reduce their cost. So let’s look at an example here. An average MLB video, for example, has 270,000 frames. You’re talking an average brand exposure is of six. This example shows 16. For every detection, you’ve got to take… Sorry. You’ve got 1. 6 million detections, and for every detection, you’ve got to do two actions. What’s the brand? At least two actions. What’s the brand? What’s the placement? The placement is the object. Is it a jersey? Is it an upper tier, lower tier? Is it on the pitch? Where is it? That’s 3. 2 million actions per game that have to be taken. It’s a lot of work. That’s just one game. If you wanted a human to do that, you’re talking about 667 days, one person to do all those actions. That’s making two actions every 12 seconds. Two years. To do it in five days, for one game, you’d need 133 people.
Franco De Bonis (00:08:23.07)
So clearly, you’ve got to implement some form of automation, some form of computer vision. So what’s the impact? What do they do with that? So yes, they reduce the workload by doing one frame a second, and they also use some form of computer vision to meet the market cost demands, to meet acceptable time scales, but still, you’ve got these issues here. Lack of transparency, lack of accuracy, and it still takes a long time. We could really ask, Well, why don’t they just improve the computer vision? Why don’t improve the system. The reason for that is it’s really tough. The more you automate, because computer vision, AI, as anyone who’s used AI knows, it’s not infallible, it’s not perfect. You can improve and improve and improve, but it will make mistakes. So if it’s not a great system, the more you automate, the more errors, the more humans need to fix those errors. So you end up with the same situation. But if you have a system built for purpose, you get these benefits: faster results, deeper, more granular data, accurate and transparent results, and optimised cost. By now, you should be thinking, I need to check the data I’m getting.
Franco De Bonis (00:09:50.16)
If that’s what you’re thinking, I’m going to introduce some questions that you need to be asking your providers to make sure that you’re getting the right data. Firstly, accuracy. What level of quality can you guarantee? So 95% accuracy is a good number that we’ve identified. Anywhere around that zone or above, Any errors really don’t move the needle. So you might move by half a percentage point or something like that in terms of exposure and AVE. So 95 is a good number. How How transparent is your data? Don’t just accept ranking scores or, Oh, you’ve got nine out of 10 for your AVE, or whatever it may be. You want really detailed data that you can interrogate and validate to be accurate. What turnaround do you need? You might be happy with 2-3 weeks, but if you’re making, think about how many digital assets are now happening and are being invested in. You might want to know very, very quickly within a few days, are those assets from one week, one fixture, one weekend to the next fixture, I invested in this asset. Was it worthwhile? Did I get the value from that asset? If you’ve got the right system and the right data provided, you can make those decisions and change the asset if that’s the case.
Franco De Bonis (00:11:25.08)
If you have to wait till mid-season, end of season, that’s a lot of lost opportunity. And then what data points were important for you? I talked about brand and placements, but there’s way more than that. You can have objects tagged, you can have player tagging, you can have verbal mentions, for example, where you roll in any audio data into the video data to expand that exposure. And when you do this right, in real terms, what you end with is a whole bunch of data that supports that, and you should be seeing. If we’re talking transparent data, every exposure should come with a clip and a snapshot. You can validate for yourself Am I getting the quality that I’m being told? Is the valuation correct that is underpinned by this core data? What that looks like is a file that typically has 20,000 plus rows of data, all of which is supported by all of these columns, which can be a CSV, a JSON, which is then imported by the provider and exposed to you and allowing you to slice and dice that. So in short, in summary, if you’re a brand, don’t settle for inaccurate data when you’re making crucial decisions on that.
Franco De Bonis (00:12:58.11)
If you’re an agency, Why deliver inaccurate data slowly when you can deliver accurate data quickly? Okay. So I’m going to hand over now to Zaheer, who’s from Vision Insights. We work very closely, a great partner, and he’s going to talk about how this data is then integrated and exposed to clients.
Zaheer Benjamin (00:13:19.20)
Many thanks, Franco. So yes, my name is Zaheer Benjamin. I am the President of Product and Strategy at Vision Insights. By way of introduction, we are a research and insights agency. We’re headquartered in the United States. We work with sports teams and league across all the major US sports. We’re just starting to expand internationally to the UK, Europe, Asia. We have two digital products. One is around syndicated fan insights research. We survey hundreds of thousands of fans in 22 markets all around the world. We provide subscription access to that research. Everything you’d want to from a sports marketing perspective about a fan’s, their demographics, their psychographics, their purchase habits, their impact of sponsorship, their fandom. We put that in an online web application called Decoder, and provide subscription access. The other module within Decoder is media analysis. We’re really pleased to have been working exclusively with Visua over the past three years as our data provider on the media analysis side. To give a bit of more colour on how we leverage Visua’s data and to continue with Franco’s theme about the best kept secrets, I think one really interesting insight, certainly in North American sports, is that For example, a typical NBA game, when it’s broadcast, will have multiple feeds produced.
Zaheer Benjamin (00:14:50.28)
So the home team will produce a broadcast feed, and then the visiting team will produce another feed. And these feeds will have different camera placements, different replay angles. But what typically happens is that many of the providers in this space, when a brand or a property asks for the total exposure generated across all feeds, they’ll take one feed, typically the home feed, and they’ll just multiply it by two. They’ll measure one feed because it’s costly, it’s expensive to measure these feeds. They’ll measure one feed and they’ll double it, and they’ll give that as the total exposure. Obviously, if you think about this, this is less than ideal. Especially given the level of accuracy and rigour that the brands and properties are demanding. So there has to be a better way. And what we found is that through leveraging visualist technology and measuring both the home and the away feed separately, we’re able to get much better level of insight, accuracy, and quality to provide to our clients. So here’s a quick example of NBA game showing both the home feed and the away feed. Home feed on the left, the away feed on the right.
Zaheer Benjamin (00:16:01.25)
This is the same segment of the game. So you see the score is the same, the clock is the same. But you’ll notice on the left, the home feed, you’ll see a large sponsor local, MedStar Health, clearly visible just inside a three-point line. Now that we are in the era of digital overlay and brands are superimposing their advertising directly on the broadcast feeds, that same brand is not visible on the away feed. If you were to take just one feed, double the exposure, you’re either going to severely over-count or under-count the value of that exposure. So again, I think when we’re looking at delivering on the accuracy and the level of insight that brands and properties are demanding, the ability to scale, as Franco has described, is critical for us. So through this technology, we’re able to answer some of these key questions, right? What is going on in the away feeds? What is going on with digital assets that are often virtual? When they travel? Are these exposures and values and durations real? We think in terms of the value proposition, again, partnering with Vizua, we think there’s three key components of our value proposition.
Zaheer Benjamin (00:17:17.08)
One is the all-encompassing analysis. Again, we’re measuring all feeds, all games, 30 frames per second. So there’s nothing that we don’t capture. There’s nothing that is missed in our analysis. Our valuation is therefore rooted in accuracy. So every brand is captured. We have an extensive library of logos and brands. In fact, even if there are sponsorship opportunities that are not being leveraged as yet, so perhaps there is a particular asset that is used by the team or the property IP. We’re measuring everything so we can actually do white space analysis on an automated basis and think about what would be the commercial potential of a digital asset or a broadcast asset that is currently not in a partnership. So we’re able to do Whitespace analysis in a very automated and rapid manner. And then finally, proof of performance. So we’re able to deliver not just the numbers, but we’re able to be an open book, deliver all of the clips, all of the snapshots to actually prove the broadcast exposure for every single brand, for every single match. So this is really helpful on a match match basis. But again, the fact that we are covering all the matches, all the games across all the league allows us to do some general benchmarking across properties and across brands.
Zaheer Benjamin (00:18:40.21)
So for example, this is an example of one brand, FanDuel, that has sponsorships in many different arenas. We’re able to compare the exposure they get from similar assets. In this example, the tunnel covers across two different arenas. The arena on left has the tunnel cover on the extreme left-hand side the screen, and the arena in the middle has the tunnel covers in the middle of the screen. And even though you’d think in the middle, it probably has better exposure, when we calculate our vision value, we see that the placement in the left is actually a better exposure better, better broadcast value because it’s less obscured by the backboard. So once we aggregate across the entire season, we’re able to benchmark across different teams, different arenas, and provide value to both the brands in terms of how they allocate their their investments and the teams in terms of how they optimise their inventory, their broadcast visible inventory. So that’s an example from the NBA. We can do the same thing in Major League Base. So we looked across all of our teams, all of the matches, all of the games, and we looked at the outfield wall signage.
Zaheer Benjamin (00:19:49.17)
We were able to identify which outfield wall signage performs best on a broadcast exposure perspective and come up with some best practises for teams and brands. For example, we can say that the clear large logos with either very dark or very bright backgrounds tend to perform the best. You may say that as intuitive, but we’re actually able to quantify how much better these types of creatives perform and put a dollar value in that. And these are aggregated across entire season. These differences can be significant in hundreds of thousands, even millions of dollars. And it’s not just for matches, any media. We can run through visual technology. So we’ve worked with brands here in the UK in terms of valuing the exposure from integration in popular TV shows, Gavin and Stacey, and Ted Lasso as well. So hopefully that’s a helpful demonstration of how we can leverage visual technology to answer some key questions about accuracy, true market value, and transparency. Thank you.
VIDEOS 
