Measuring Occupancy at Live Events

Measuring occupancy at live events is a remarkably fruitful application of people-measurement and it's one of the very few analytics use-cases I've found with a direct path to ROI. In most cases, revenue is by far the hardest part of the ROI equation to get, but it is only part of the equation. In addition to the return, you need to understand the investment. Whether you're interested in freeing up tickets in real-time, in capacity planning, or in both, you need to understand what it will take to get the measurement you need - and that measurement boils down to area and event occupancy.

How hard is capturing event occupancy? Though it's far from the most challenging of people measurement applications, it’s not as trivial as you might think and the type of measurement you need for intra-day ticketing is a little different than what you want for capacity planning.

For intra-day ticketing, what you’re looking for is site-wide occupancy and there are three different ways to determine site occupancy. One way is to count all the people in every area. The second way is to count ins vs. outs at all entrances. The third is to count outs vs. ticket scans.

Method one – universal counting – is very difficult at big live events which tend to be spread out over large and complex areas. When it is possible, it’s a reliably accurate but by no means precise measure (though this is not a problem that requires real precision). It’s usually not possible.

Counting ins-outs to calculate occupancy is the way we’ve always done it at retail stores, casinos, university cafes, etc. It works quite well in all those cases. Modern people counters are very accurate – often achieving 99% or better in measuring line crossings. Yet in the live event world, it’s hard to get it right. The large entrances at most live events get absolutely slammed at open and close – and large crowds make for a lot of occlusions – which makes for worse counting. In many cases, it’s not possible to get top-down mount points either. That means you’re relying on high-angle mount points using video camera and ML or lidar. Either way, achieving 99% accuracy isn’t easy.

And even if you do achieve that accuracy, it’s important to realize that 1% error can be meaningful. If you’ve got 50,000 patrons in a day, your 99% accurate in-out counting may still leave you with an error of 500 people in occupancy. That might be significant and even if it isn’t significant, it can just look bad when you’re reporting 500 people in an empty stadium or 0 people with 500 people still streaming out.

The problem is made worse, of course, by ins-outs from workers and accredited folks who aren’t part of the capacity plan. In-out counting can be significantly skewed by these groups – certainly enough to matter for intra-day ticketing.

Method three is to use scan data (ticket and accredited) for ins and use people-measurement for outs. This is usually our preferred method even though it requires some real-time integration of scan data. It has several advantages as a practical technique for in-day occupancy measurement. First, it provides the most reliable in-count there is, and it does that regardless of configuration. Second, it cleanly separates our accredited ins. Third, it uses the actual asset (tickets) and provides for seamless de-duplication of reentries against the true asset pool (how many unused tickets are still out there). Finally, and this is the most subtle point, it maximizes the accuracy of the out-counting.

Line cross counting of the sort used for ins and outs is least accurate when an area is most crowded. That means your error rates will be very low during slower times and will be highest during the crush of initial entry and final exit. From an in-counting perspective, that’s an unavoidable problem and it’s partly why ticket-scans work better for in-counts. But for outs, the huge crush is at the end of the day when counting no longer matters for intra-day sales. For most of the day, outs are a relatively low-volume stream with very high accuracy. Combining high-accuracy ins with high-accuracy outs nets high-accuracy occupancy.

Pairing this data up with a predictive model will produce highly-accurate intra-day occupancy forecasting – accurate enough to make sure that if you’re creating additional entries, you’ll never exceed your capacity.

?

Getting site-wide occupancy is half of the problem. But what about understanding usage across key areas of the event? That, unsurprisingly, is a very different beast. In most cases, the best way to understand usage of a key area is to count the people in it on a regular basis (like twice a minute). That’s what tools like PMY’s OPTIC do for almost any kind of large event space – from a grandstand to an open-field. Using video camera and ML-inferencing provides a robust and low hardware path to counting even really large areas. And while the answer you get is never exact, it’s almost always accurate enough to tell you what you need to know.

On the other hand, if you want to measure usage of a lounge or restrooms, that’s not the way to do it. For smaller and mid-sized areas, you can measure occupancy and usage either with measurement camera at chokepoint or lidar sensor overhead. It’s the same for measuring flow through a space (e.g. a corridor or connector). Both lidar and camera are potential solutions. For the most part, you can use the PMY Sensor Technology Decision-Tree to decide on the best approach – and you’ll always be navigating the top part of that tree – counting people or flow through a choke-point.

Then there’s the high-value areas that demand or reward more detailed measurement. Your merch stores, your key experiences, your premium lounges – these are best measured with full-journey techniques (usually lidar) because in those spaces, you want to know how many people there are, but you also want to know the detail about what people did.

Finally, there’s the measurement problem presented by large, complex facilities like an arena inside an event space. Here, we tend to duplicate the methodology for the entire event – using scans for the in-count and people measurement for the out-count.

The main point here is that you’ll probably have to deploy a range of different sensors and techniques to build the complete operational picture of crowding and capacity. If you do the work, you’ll be able to make a compelling argument for increasing event capacity and you’ll be in the even more enviable position of having the operational intelligence to make sure it works.

?