Preparing for AI Video Ads

Last week, Magnite announced its acquisition of Streamr.ai, a company that describes itself as the "easiest way to generate video ads and launch them on CTV in less than 2 minutes." 

According to the press release, Magnite hopes that Streamr.ai technology will help ramp up small and medium-sized business spending: 

Magnite is not the only player in this game, as other demand platforms are salivating at the prospect of onboarding more SMBs to CTV. And is that any surprise, given that Google and Meta have earned billions creating tools that make it easy for SMBs to advertise on their platforms?

But CTV is an entirely separate ball game, and one of the biggest blockers that could prevent an SMB from advertising on CTV is the cost and complexity associated with developing video ads. While developing display creatives requires little more than a nice camera and a basic knowledge of graphic design, creating a video creative worthy of a 4K screen demands more investment and skill.

Ever since OpenAI released its text-to-video AI model, Sora, in February 2024, with other models quickly following, some companies saw an opportunity to use text-to-video AI technology to whip up creatives on behalf of the advertisers that otherwise didn't have the time or money to develop video creatives. 

And now, we have real-world examples of companies cranking out video AI creative. First, it was the big companies like Coca-Cola or Toys R Us dipping their toes in.

And Kalshi reportedly spent just $2,000 to create this unhinged ad that ran during the NBA Finals:

And there are more examples of individual creators producing concepts for brands like Nike or Adidas:

The only thing holding AI advertising back is backlash from consumers. Take one look at the YouTube comment section in the previously mentioned Coca-Cola or Toys R Us ads, and you'll find a sea of backlash against the use of AI for advertising:

Coca-Cola:

• Coming from a billion-dollar company like Coca-Cola, this is an AI embarrassment, and likely a deliberate marketing strategy.

• Nothing like celebrating the spirit of Christmas with the most soulless commercial possible.

• No animators were employed in the making of this advertisement.

• Words cannot describe the level of hatred I have for this ad

Toys R Us

• This is like if I had a nightmare about Toys R Us.

• This commercial killed the magic

• Soulless venture capitalists further ruin Toys R Us by producing soulless ad

• Gosh this is horrendous. I felt nothing watching it (other than a bit sick)

Given that creatives were the first career paths threatened by the generative AI boom, it is no wonder that the first examples of AI-generated video advertisements would receive the most criticism. The harsh reality is that we will most likely view comments like these in the same way as the destruction of textile machinery by the Luddites in the 19th century. 

Generative AI creatives will become indistinguishable from reality, and the cost to produce them will only decrease, which will mean consumers won't even know what to criticize as they eventually flood CTV. 

The prior examples were generated by creative professionals, but with companies like Streamr.ai and others developing more automated solutions, it’s only a matter of time until AI creatives from advertisers of all sizes start hitting the living room.

Big players like Amazon have released new tools to generate video creatives, and independent players like Creatify can also produce video ads based on Amazon product listings. And it's not just full-on AI-generated creative. Spaceback has created tools for advertisers that can crank out CTV ads using existing social assets.

But what does this coming onslaught of generated creatives mean from an ad tech perspective? The one thing I can be certain of is that it will mean more unique creatives flowing through the video advertising ecosystem.

I anticipate car commercials that dynamically swap environments based on a user's location or even advertisements that dynamically insert the products you've added to your cart. Scenarios like this could lead to an onslaught of creatives that publishers must deal with. 

Last Monday night, I got hit with this Gillette ad during the Buccaneers vs. Texans game. It is customized to the matchup, with Troy Aikman referencing the Buccaneers, along with the Buccaneers logo and branding in the ad itself. Now, I'm not saying this is an AI-generated Troy Aikman, but it got me thinking about the potential of using content metadata to target uniquely generated creative. 

Perhaps Mr. Aikman filmed 32 separate segments for each NFL team (although I've only found one other for the Cowboys vs. Eagles game, excluding Troy Aikman), but this is a manual and costly process. I anticipate advertisers will start creating variations like this automatically through generative AI or other tools. 

If generated creative variations become more commonplace, there will be an even greater desire for accurate programmatic content metadata and user identity. Advertisers and ad tech providers have always craved these data points for targeting and reporting purposes, but now they can use these inputs to generate relevant creative on the fly. For example, if a publisher passes which game an ad opportunity is for in a bid request, then Gillette can slot in the appropriate NFL team-based ad more seamlessly. 

There could be a world where advertisers customize CTV ads down to the user level, allowing them to showcase products the user has shown interest in or tailor an ad to target their interests and desires more effectively. But the number of unique creatives generated from a practice like this could put an immense strain on publishers who must transcode every variation of an ad. However, the possibility exists and may become an engineering challenge to solve in the future. 

Publishers and ad platforms already have to deal with potentially thousands of creatives a day. What happens when this increases to tens and hundreds of thousands? 

CTV publishers can already enforce creative standards by analyzing the creative metadata and files. Pubs can quickly interrogate things like width, height, bitrate, advertiser, category, and more, so dealing with a deluge of new creatives is more of a matter of scaling existing systems. 

However, what happens when anybody with a credit card and a dream can access premium streaming inventory? Interrogating the content of the advertisement itself may present unique challenges that publishers will eventually have to deal with. Objectionable, uncanny, or low-quality advertisements may present a jarring experience to users. 

Computer vision technology can detect inappropriate frames of a video (nudity, hate symbols, curse words, etc.), but determining quality may ultimately come down to human taste. If a publisher opens up access to small businesses without guardrails, they may be in for a rude awakening of the AI slop that starts rolling through their programmatic pipes. 

Publishers may want to deal with AI creatives differently than "professionally produced" advertisements, much in the same way advertisers make a distinction between professional content and user-generated content. We may see publishers requiring AI creative to flow through separate deals that require further review for auto-generated creative to make sure they adhere to creative guidelines.

Additionally, a high amount of unique creatives introduces challenges for server-side ad insertion use cases. In some SSAI setups, the first time a DSP bids with a new creative, that opportunity is "burned" because the creative must be registered in a creative approval queue and transcoded before it is ready to serve. This quirk may be fine in VOD use cases, but in live sports environments, for example, a DSP could bid with the same creative thousands of times all at once, and if that creative is approved, it could be thousands of burned opportunities before the creative is ready to serve. 

SSPs sometimes offer creative pre-ingestion tools to allow DSPs to pre-register creatives before they start bidding with them to avoid these situations. In a world of generative AI, creative pre-ingestion becomes much more crucial. 

If generative AI is the key to making SMB spending take off in CTV, one last thing I'm interested in is the economic impact on the video advertising ecosystem. An increase in demand should push prices higher, but SMBs don't have the same deep pockets as large advertisers or agencies. The reality is that these smaller advertisers are likely looking to come in at a much lower price point, which relegates them to remnant or low-quality inventory. 

I anticipate that we will first start seeing SMB AI ads on FAST channels where supply is plentiful. That's not to say that the big streamers won't have their fair share of ads created using generative AI, but it will be in the form of big advertisers streamlining their production process through AI, most likely with much higher quality results.