Personalized Ad Insertion: The Key to OTT Success
As consumer demand has rendered over-the-top (OTT) video provision a necessity, providers have been forced to look at how to monetize it. The market is highly competitive, the audience is fickle, and of course ...
As consumer demand has rendered over-the-top (OTT) video provision a necessity, providers have been forced to look at how to monetize it. The market is highly competitive, the audience is fickle, and of course, monetizing digital content is almost as complicated as delivering it in the first place. In addition to subscription, OTT providers have also become reliant on advertising for revenue. But consumers are not particularly willing to sit through irrelevant ads, and this poses a problem for a business model reliant on advertising to generate revenue. In fact, a report from IPG's Media Lab states that 65% of people skip online video advertising. The trick is to ensure a worthwhile experience for the viewer, as well as added value for the advertiser.
An effective way of doing this is to better target the ads to end users based on who they are, where they are and what device they are using. It is also important to ensure that the content surrounding the ads (including other ads) is contextually relevant. The "icing on the cake" scenario occurs when the advertiser can also understand what level of engagement the ad received. For example: Was it watched? Did the user interact or "click through"? Did they buy what was advertised?
Metadata is Critical
Once upon a time, OTT was pretty much solely based on on-demand programming. Ads were added at the beginning - known as "pre-roll" - or in one or more "breaks" in the middle of the content - "mid-roll." If the on-demand content is recorded from a live/linear transmission, the ads that are already in place may be retained for a few days (in the United States, Nielsen counts the ratings of content viewed after the original air date for three days). This doesn't do anything to personalize the ads, of course.
In order to make that happen, "metadata markers" are placed in the appropriate places in the video file during processing, and when the viewer reaches that pre-determined point in the video, a call is made to an ad decision server and the returned ads are dynamically inserted on the fly. Ads can be served based upon each user's demographic profile, their location, the device they are using, as well as information contained in the metadata markers, such as what the content is, what the genre is, how far along in the content the viewer is. As each ad is played, tracking URLs can be called to report that it has been (or is being) viewed, and some of the more sophisticated systems allow reporting of click-through engagement as the viewer goes off to look at a shiny new BMW. The video will wait, of course, for the viewer to interact before resuming. This is a key advantage of on-demand viewing.
This model has been in place for several years already, generating material revenues for content owners and distributors.
Today, though, it is more complicated, as existing linear TV broadcasts are being made available online more or less at the same time. Online live/linear video is typically delayed by about 30 seconds to a minute from the broadcast because of the protocols used to get continuous video through a medium designed for file delivery (i.e., the Internet). Hence, there are significant differences between on-demand and linear delivery when it comes to advertising.
Firstly, the timing of exactly when to insert and the length of the advertising breaks are much more critical in a live/linear situation. If the ad is inserted too early, it will cut off programming, and if it is inserted too late, the viewer will see whatever is to be replaced (filler promos, other ads etc.). This isn't a problem in on-demand as the metadata markers described above are inserted manually on an edit station, offline. In live/linear, it must be executed automatically. To make this possible, metadata markers signaling the exact start and end of each piece of content must be inserted as the linear programming is originated.
Secondly, televison is scheduled in advance with breaks full of ads that the TV broadcaster has sold. Not all the ads scheduled on broadcast TV can be delivered online because the advertiser may not want to deliver an ad designed for TV over the Internet (or they may not have the right to do so). This means that online advertising breaks need to be scheduled in advance, too.
So, in addition to tailoring the ads for individual user demographics, location and device type, the ads need to be scheduled in advance and be the same length as in the original broadcast. Individual broadcast TV ads must be capable of being removed and replaced with new ads. Without the right metadata markers in the video, as well as advance knowledge of the schedule, it would be near impossible to enable this functionality seamlessly. The only way of delivering an online version of a broadcast TV channel without metadata would be to completely duplicate everything, "from soup to nuts." This would be exorbitantly expensive.
Where the Ads are Inserted Matters
Up until now, we have discussed the decisions regarding what ads are replaced/inserted. Another significant shift since the start of OTT advertising in the on-demand-only days is where the ads are physically inserted into streams being delivered online.
In the early days, dynamic insertion almost always took place on the device itself. The principal advantage was that it enabled the online delivery of the content to be completely uniform to every device. The device managed the insertion directly and communicated with the ad decision server directly. It also made it quite straightforward to call the tracking URLs as the ads were viewed, since this could be built directly into the viewing app on the device.
However, there were significant disadvantages. Firstly, hackers could quite easily interrupt the ad insertion process and essentially steal the ad inventory. Although this has been mitigated somewhat more recently, it is still a concern for advertisers and content providers alike. Secondly, switching video seamlessly so that not only the timing is right but the quality of the ads also matches the content is hard to do consistently on the vast array of consumer devices used today.
As a result of the deficiencies in client-side insertion (as the above is known), providers started to insert ads on the server. Server-side Insertion makes it much easier to insert ads seamlessly and virtually eliminates the ability of hackers to steal inventory as there is no easy way to detect when the content ends and the ads start - there are no calls to ad decision servers and no metadata markers. It also simplifies the app on the device somewhat, though at the expense of making tracking URLs much harder to deliver, though not impossible.
Given the complex advertising market that OTT now is, it is likely that both server-side and client-side will be used together, broadly split (though not exclusively) with linear/live using server-side, particularly because of the advance scheduling requirement and high value of ads likely for major live events, and on-demand using client-side because of the interactive capability.
Putting it All Together
Because delivery over the Internet requires individual streams for each viewer, it is theoretically possible to deliver completely different ads (and content for that matter) to each viewer. In practice this is unlikely to happen. However, the ability to target specific ads to people based on (A) their demographic profile and maybe past viewing or even buying habits (personalization); (B) their geographical location (regionalization); or (C) on the content surrounding the ad (contextualization), greatly increases the efficiency in terms of cost per thousand people (CPM). This is because the advertiser is reaching a thousand people they do want to reach rather than a thousand people, some of whom they don't intend to reach.
The broadcaster essentially gets to re-sell the same inventory multiple times since the same time slot can now be reused multiple times for different ads. These ads are also more relevant to the viewer and so are less likely to be a chore to watch. So everyone wins.
This sort of targeting can only be achieved with the right metadata both from the broadcast side - timing, content information, schedule of upcoming ads - and from the client side - demographics, device, location. Fortunately, obtaining the broadcast metadata and inserting it into the outgoing video stream is something several vendors can do automatically quite easily today.
Alan Young is the chief operating officer of Crystal.