In two recent interviews, Jarred Wilichinsky, Paramount’s SVP of ad operations, explained the story behind the Paramount ad experience. He discussed what it took to oversee three different properties’ ad-tech stacks, create a comprehensive strategy, and deliver on requirements for the Super Bowl and Paramount+ launch.
The unforgiving world of ad tech means any mistake can cost millions of dollars. Ad tech is the key to a successful AVOD/FAST monetization strategy, but there are so many moving parts that it can be a challenge to get it right. Wilichinsky provided practical insight into what works, what doesn’t, and how ad-supported delivery compares to subscription offerings.
This interview has been edited for length and clarity.
Nadine Krefetz: What does Paramount offer in terms of subscription versus ad-supported viewing?
Jarred Wilichinsky: We’ve always had a subscription model where users have a choice, a lighter ad load, or an ad-free experience—that’s the Paramount+ side of the house. Then there is Pluto TV, one of the original FAST platforms.
Nadine Krefetz: Why is EyeQ important?
Jarred Wilichinsky: EyeQ is our offering to the ad market that helps consolidate and commercialize our entire footprint. Whether it’s owned and operated, distribution, TV Everywhere, SVOD, or FAST, we have the largest premium-content, scalable option out there. As of the first half of 2022, we reach 80 million unique streamers on a given month. You could hardly touch the internet without running into our content in some form.
EyeQ growth year-over-year
Nadine Krefetz: How did you plan to consolidate ad tech across Paramount?
Jarred Wilichinsky: Previously, everything was independently run with different teams, different processes, different tech. I’ve been at CBS for almost 20 years, and Q4 2020 to Q2 2021 was the craziest and most intense period of my career. We were planning for NFL playoffs and the Super Bowl, we had Paramount+ coming up, plus March Madness and the Masters, and we were trying to develop our ad stack migration plan. The ad platform is very broad, including CRM, order management, audience data, and the actual ad server. So, we consolidated it and tried reducing the impact of third-party data.
How Paramount’s ad-tech operations have evolved, 2020–2022, as it has implemented its ad stack migration plan
Nadine Krefetz: What is your focus for consolidating old and new technology?
Jarred Wilichinsky: The ad server logs are some of the most important things. It’s not just like, “You’ve got to do ad serving and figure out a process.” You need to get all that data in a central place. You’ve got to normalize and operationalize data. It’s such a huge focus to help us scale and do things correctly. Then, of course, you’ve got to bill. So, we had different billing elements of finance, and we have to consolidate on that.
We capture everything in our logs from every single endpoint and every single app, on-site or off-site. It’s all about getting it to one place so we
can make intelligent ad decisions across every single thing we have from a demand standpoint: direct I/O, programmatic guaranteed, private marketplaces, indirect lanes.
Nadine Krefetz: Why was updating your ad server so critical to this process of merging ad tech at Paramount?
Jarred Wilichinsky: When you’re serving inventory into different endpoints, it’s just exponentially harder to do anything to execute, report, and bill on a campaign. CBS was on Google, Viacom had their own instance of FreeWheel, and Pluto had a different instance of FreeWheel. We wanted to sell it all as one.
As part of this migration, not only did we have to migrate all of our owned-and-operated apps to a new instance of FreeWheel, we also had to do the same work with every single endpoint we distribute to: Amazon, Hulu, Roku, YouTube TV, Comcast, Charter, Cox, Altice, and Xumo. Pluto has distributed channels on Samsung TV and Sling. This also included all of the TV linear content from CBS to all the Viacom brands—MTV, Nickelodeon, Comedy Central—distributed to every single MVPD and virtual MVPD. There are 50 different things to fill in, and we needed standard naming conventions. There are also a lot of things we don’t even directly control because we can’t make the changes on Dish; they need to do it.
Nadine Krefetz: How long did this take?
Jarred Wilichinsky: We spent a couple of months on it in Q4 2020, but we really started doing app and partner work after March Madness. April to June was a condensed period to execute our plan.
Disney did a similar move from FreeWheel to [Google Ad Manager] not too long ago. We’re all friends, so we’ve talked about the pains they went through. I think one of the things they probably wish they spent more time on was the reporting and the billing element. Your ad server is the source of a lot of this information to automate all these other processes. We deliberately overpaid attention to our data hierarchy. When traffic started flowing on June 28 and we started seeing the logs come in, we knew we were in very good shape.
Nadine Krefetz: What are the key metrics you look at to make those kinds of assessments?
Jarred Wilichinsky: One of the most important things that we look at is our supply. Having a solid grasp of the overarching supply and hundreds of endpoints and different apps is probably the most important. If the supply is sound and everything is working, then we can go one level deeper and see if our completion rates are good, if the frequency of ads is correct. But if the supply is broken, everything else below it is broken.
Nadine Krefetz: Where do things break?
Jarred Wilichinsky: Everywhere. You name it, it breaks. SCTE triggers, ad insertions, CDN outages, AWS outages—if it can go wrong, it has gone wrong. Literally, it can take one person putting a space into the code, and you can potentially bring down the whole system.
Nothing in digital is 100%. Latency is always a never-ending challenge, especially on live, high-concurrent stuff. You get blank VAST [Video Ad Serving Template] tags. To this day, people still think FLV is an acceptable video format. VPAID [Video Player-Ad Interface Definition] still comes up, even though it’s been deprecated by the IAB. We have to think about how do we normalize audio at scale.
Other things we look at are timeout thresholds, from an ad server and DSP. We make sure that things aren’t getting filtered, for whatever reason, [in SSAI]. The SSAI transcode elements also need to work correctly, or ads don’t get on screen. This is even more important in programmatic.
Nadine Krefetz: What other datapoints do you gather, and how to you use them?
Jarred Wilichinsky: With our order management system, we have very well-defined products. We’re able to see what’s selling, what’s not selling. From an operations role, we use Salesforce as a big part of our workflow and processes. We see workload by traffic and campaign manager. We know how many placements, creatives, and programmatic deals they have. So, we’re able to forecast our workload a bit and understand if we need more resources or if we need to move people around to accommodate things.
We also use data for monitoring. We need 2-hour problems, not 2-month problems. Being able to visualize data—especially with the breadth of our network and where we run it—is extremely important. Being able to quickly identify and troubleshoot that stuff is critical. There’s a dashboard for almost everything because there’s a lot that can go wrong. A lot of times, there’s nothing internally we can do, so we have to call a vendor to fix something.
Nadine Krefetz: What breaks on CTV?
Jarred Wilichinsky: CTV is no different from mobile apps. It’s an application. The tough part with CTV compared to mobile is that in mobile, there are Android and iOS, and that’s it. With CTV, they all don’t have the same chips, they all don’t have the same CPU power. You can do something here, but you can’t do it there. There’s so much more complexity when you’re trying to do wide support for CTV. Our stitcher works well on one, but not on another because of DRM requirements.
Nadine Krefetz: What are viewing patterns for subscription versus ad-supported?
Jarred Wilichinsky: They are more or less the same. It comes down to expectations going into it. Knowing that there will be ads or not ads really isn’t dictating what they consume or how they consume it.
Nadine Krefetz: Are you doing personalized advertising? Is it one-to-one audience-segment targeting?
Jarred Wilichinsky: Personalization can be done many different ways. We have a slew of first-party data. We have third-party data. Our clients
bring their own first-party and share their segments with us, so we help them find those users.
In digital, there’s a lot of small targeting, but in order to target small, you must be big. If you’re going to have a 1% match rate, 1% of a billion is better than 1% of a million when you’re dealing with match conversations. Consolidating it was key. We haven’t necessarily gotten to the point where we have an official approach to how we use this data to personalize ads because there are thousands of endpoints and thousands of potential variables to personalize an ad. If there’s a very small datapoint, that makes sense to personalize, but if you can only do it for 10 out of 80 million people, it’s probably not worth going down that path.
Nadine Krefetz: What’s included in contextual advertising?
Jarred Wilichinsky: Audience targeting is one thing, but contextual targeting is just as important—geotargeting, time of day, day of week, there’s a lot of different targeting options out there that we offer. Contextual is still extremely important, especially because we can be transparent in our content targeting and reporting where others can’t. This includes certain genres of shows, sports, or news, so we have a handful of content bundle rotations. We’re targeting that content no matter where it lives, whether it’s on our Paramount+, CBS, Hulu, or YouTube TV. We’re selling content, not endpoints.
Nadine Krefetz: How do you handle targeting?
Jarred Wilichinsky: This comes down to how we facilitate your privacy-safe identifier. In some cases, we know emails because Paramount+ is a subscription product. Or we know what someone is potentially consuming across the owned-and-operated landscape of Paramount, from CBS News to Pluto to CBS Sports. Our first-party data on what you’re watching is important, but we also partner and have access to third-party data sources.
Nadine Krefetz: How do you handle frequency capping to ensure you are not showing the same ad over and over again?
Jarred Wilichinsky: That’s all done within the ad server. There are a lot of features within FreeWheel or Google, where there’s metadata to frequency-cap off of. You have your record of Procter & Gamble, and it’s the Charmin brand, and you’re able to do frequency capping with that. [The AD-ID registry] is not very prevalent in the digital ecosystem.
In the ad server, you have a placement that you’re able to frequency-cap against. It’s tied to a user ID, so the system knows to show #12345 only once per hour, twice per day, or whatever [your frequency cap allows]. It’s not like I put in my AD-ID, and it tracks off of that. The systems aren’t there yet.
Even though we’re digital, we have to still act like TV and work like TV. I think this comes down to how we use FreeWheel and our frequency management. On Paramount+ or Pluto or any of our other endpoints, you’re not going to see the same ad five times in an episode. We know that’s a problem across the industry in general, where user experience suffers because of some of the issues we have in ad tech.
Nadine Krefetz: How has digital advertising influenced how ads are delivered in broadcast and cable?
Jarred Wilichinsky: We already have live linear, addressable at Paramount with a bunch of our MVPD partners. So, there are spots today that are using data to deliver either a much stronger optimized show or targeting. In broadcast or cable, you don’t yet have that individual household decisioning, but we can definitely do better planning on some of the one-to-many segments where we’re using very specific data segments.
Over time, things will migrate from the traditional “everyone sees the same ad” to “every single ad is individualized.” There’s a lot of hardware that needs to get updated. There are a lot of TV sets that can’t do any of this stuff. There’s a lot of the ecosystem at the broadcast endpoints—the stations—that needs upgrading. This is a long road to getting to a place where every single ad is addressable. Whether it’s using data or time or content targeting, we’ve got a long road before it all changes.
Nadine Krefetz: Are you using technology to check the ads, or are you actually eyeballing it?
Jarred Wilichinsky: We use technology, and we have partnerships in place to check the core specs—bitrate, file size, audio levels, frame rates, and release frames. The operations team still watches ads. We still make sure there’s not cursing or anything inappropriate; the whole standards and practices aspect is still important, even though we’re digital.
Nadine Krefetz: How would selection criteria work, where you have multiple advertisers targeting the same audience segment? How do you determine which advertiser would get preference?
Jarred Wilichinsky: We rely on the ad server ad decision system—FreeWheel, in our case. It’s their job to figure out how to do the actual ad selection, but we provide input to make sure we don’t hammer people with the same ad over and over and over again.
Nadine Krefetz: Is your approach to build or buy when it comes to implementing an ad server?
Jarred Wilichinsky: We’re not going to build an ad server [for most of our platforms]. We use Google DAI as the ad stitcher for the legacy CBS, Viacom, and Paramount+ properties. So that’s a buy. But on the Pluto side, they’ve built their own stitcher.
Nadine Krefetz: How do you make ad operations more profitable for Paramount?
Jarred Wilichinsky: To be more profitable in the long run, we need higher CPMs. We get higher CPMs with data targeting. Paramount+ is going to be in great shape because of login status. We have your email, and that opens the door to a privacy-safe approach for the future. Any time we can have better options for advertisers around data targeting, that’s one of the healthiest ways to increase your CPMs.
Nadine Krefetz: What are you offering users to improve their ad experience?
Jarred Wilichinsky: You have three options. With Pluto TV, which is free, you get a certain ad experience that’s a heavier ad load. For $4.99 a month, on Paramount+, you have a low-cost option, which gives you limited commercials. Or you could go commercial-free for $9.99 a month, and then you get no ads. I think it’s a slippery slope when you start saying, “Don’t show me this ad again.” If you say you hate every ad you see, after a certain point, what are we going to show you? We’ve got to show you something. So, I think that that level of choice gets tougher.
Nadine Krefetz: If you’re targeting people with ads, and they are not really engaging with the ads, that’s a problem. What can you do to get people more engaged?
Jarred Wilichinsky: We work with Innovid and BrightLine to build games for interactive ads. There’s some appetite for that from the buy side, and we have it available. But as much as we want every ad to be interactive and be a lot more engaging, the buy side isn’t necessarily ready for it.
It’s easy to provide 15-second or 30-second ads [the same length as ads that run on linear TV]. It’s more like the stock market: It just trades, versus something that is custom that you can’t trade on. We’ve recently launched Pause Ads on CTV devices for Paramount+; that’s a fairly well-known and accepted format. However, there are lot of nuances when it comes to some of these emerging interactive formats. Any time you add resistance to these things, they don’t scale.
Nadine Krefetz: Is any latency acceptable?
Jarred Wilichinsky: If it’s a live event, latency is the number-one enemy out there in dynamic decisioning. Pre-fetching helps mitigate that, especially in live with scale. If there are 1,000 or 5,000 people watching an event, that doesn’t tax the ad-tech ecosystem. If it’s hundreds of thousands or millions of people, you call it a minute or two in advance knowing a trigger is coming to give the systems more time to respond.
In a big-scale event, like a Super Bowl, there’s a ton of testing ahead of time where we’re simulating load, looking at performance of vendors, and determining who will be allowed and who won’t be allowed. Who has the capacity to handle millions of concurrent viewers at any given point? We always do that exercise before the game.
It’s really with AFC Championship or Super Bowl kinds of events where we’re getting close to seven figures or higher where we really have to do a lot of extra planning and be very proactive on capacity.
Nadine Krefetz: Is there ad-tech redundancy on these large-scale events?
Jarred Wilichinsky: No, that’s one of the scariest parts. If FreeWheel were to go down, there isn’t a backup FreeWheel sitting in a closet in L.A. somewhere. The same thing goes with Google. They have very, very strong uptime, but it’s a risk nonetheless. It’s not like linear TV where you have a backup facility somewhere else.
Nadine Krefetz: What does the future of ad tech at Paramount look like?
Jarred Wilichinsky: Automation is key to a lot of what is going to help Paramount get to the next level. We are looking into and developing AI and RPA [robotic process automation] to get us even further ahead on scale, reduce errors, increase time to execution, and automate instead of doing Control-C and Control-V. Self-buying is also going to be important. Google and Facebook are super successful because they empower their partners to buy on their own.
We will also be operating a bit more like TV with redundancy backups. In order to operate a scalable streaming business, those things are insanely important. The more things break, the more you impact end users, and the less they watch. It’s a downward cycle.
The goal is, you input something and then everything is automated: automated pushing, automated checks on specs. Instead of having people driving the entire process and moving things one to one, people can focus on making sure that the automation’s working. They can also spend more time on optimization and doing next-level management things versus “Copy this tag and put it in here” or “Do this and do that.” We’re really trying to automate the day-to-day tasks as much as possible so that our people can focus on the bigger picture.
Nadine Krefetz: Any final thoughts?
Jarred Wilichinsky: If you’re trying to watch high-quality TV-like content, almost 90% of it will be on a TV. So, we have to act and behave like we’re TV in terms of the user experience. We’ve seen data that people quit when they have a horrible ad experience. Whether it’s repeating ads, serving loud ads, or buffering—all of those are the kinds of things that can make someone quit, especially since it’s so easy to quit. Nothing goes into Paramount+ without us checking specs, making sure we have a mezzanine file so it looks good on the end screen.
I always give props to the ad-operation shops and what they have to go through to get to that final 30 seconds on screen. We’re that last mile, and delivering a pristine user experience is extremely important to us.
What are some of the biggest adtech challenges for live streaming at scale? Jarred Wilichinsky of Paramount talks about the ways his team works to mitigate technical issues, such as minimizing latency, load testing, and correcting audio levels, along with ensuring that the ads themselves meet acceptable legal standards and practices.
18 Oct 2022
FAST (free ad-supported television) is on the rise, but how can content distributors thrive in this very competitive landscape?
12 Aug 2022
Companies and Suppliers Mentioned