In a world where nearly 90% of advertisers use generative AI for video, simply adopting the technology is no longer enough. We sat down with digital marketing expert Anastasia Braitsik, a leader in leveraging AI for performance marketing, to uncover the strategies that separate winning campaigns from the rest. This conversation explores how to architect AI-driven video campaigns for success, from building modular creative libraries and orchestrating user intent to implementing value-based bidding and proving video’s true impact beyond last-click attribution. We’ll also touch on the critical importance of designing for a sound-off environment and the evolving role of the modern PPC manager.
Many advertisers still create one “perfect” video for campaigns. Why has this TV-style approach become a liability with AI-driven tools like Performance Max? Could you walk us through the specific components of a modular asset library—from hooks to CTAs—that you would recommend for testing?
That TV-style workflow is a relic of a bygone era, and clinging to it is like trying to race a horse against a modern car; it’s just not built for the new environment. The core issue is that a single “perfect” 30-second spot completely handcuffs the AI. Platforms like Performance Max are designed to be dynamic, to test and learn and assemble the right message for the right person at the right moment. When you only give it one finished video, you’ve robbed it of its primary function. You’re essentially telling a master chef they can only use three specific ingredients for every single dish they make.
To truly empower the algorithm, you need to provide a library of building blocks. I always recommend starting with the hook—create three to five different six-second openings. You should have a mix of styles: one that’s purely visual and product-focused, another that’s heavy on text overlays for sound-off environments, and maybe even a user-generated content (UGC) style clip that feels more authentic. Then for the body, you can break out your value propositions—one asset highlighting speed, another focusing on price, and a third on quality. Finally, for the call to action, provide varied end cards. Sometimes a soft “Learn More” is best; other times, a direct “Buy Now” is what will drive the conversion. This modular approach allows the AI to discover that a user browsing Shorts at midnight converts best on that UGC hook, while someone watching a tech review on their desktop responds to the polished product demo.
When campaign targeting is left open, AI can default to low-quality placements. How can PPC managers use negative keywords and first-party data to “orchestrate intent” more effectively? Please share an anecdote or metric that demonstrates how this improves results over a hands-off approach.
It’s a classic case of the AI taking the path of least resistance. If you just let it run wild without guardrails, it will absolutely find the cheapest impressions and easiest clicks, which often means your beautiful, high-intent ad ends up on a kid’s gaming channel or gets an accidental click from a mobile app. The feeling is incredibly frustrating because you see activity, but it’s all noise. This is where the concept of “intent orchestration” becomes so powerful. In this new landscape, telling the AI who not to target with a robust list of negative keywords is often far more impactful than trying to pinpoint exactly who to target.
The most effective lever we have, however, is first-party data. Instead of letting the algorithm guess, we can seed it with a high-value customer list from our CRM. When you designate that list as a primary signal, you’re not just giving the AI a vague direction; you’re giving it a clear blueprint of what your best customer looks like. This fundamentally shifts its objective from finding any user to finding users who resemble your top customers. I’ve seen campaigns where simply layering in a high-value customer list as a seed audience cut the cost per qualified lead in half within weeks, because the algorithm stopped wasting budget chasing low-intent clicks and started focusing on users who mirrored the behaviors of actual purchasers.
Optimizing for raw conversions often trains the algorithm to chase low-value clicks. Can you explain the step-by-step process of implementing value-based bidding using offline data from a CRM? What impact have you seen this have on customer acquisition costs when scaling video spend?
This is probably the single biggest—and most costly—mistake I see teams make. They set their video campaign to “Maximize conversions,” and the conversion event is something flimsy like a newsletter signup or a simple form fill. The AI does exactly what you told it to do: it goes out and finds a ton of people who will perform that one, low-value action. You get a flurry of activity, your conversion numbers look great on the surface, but the sales team is pulling their hair out because the leads are all junk. You’ve successfully trained a brilliant system to find tire-kickers.
The solution is to feed it better data through offline conversion imports. The process is straightforward but requires connecting your systems. First, a user clicks your video ad and submits a lead form on your site—that’s the initial online event. But the process doesn’t stop there. That lead then goes into your CRM, where it’s scored by your sales team or an automated process—is it a qualified lead or is it junk? Step three is the crucial part: you send that “qualified” status back to Google as the true conversion event. Now, the AI isn’t optimizing for form fills; it’s optimizing for leads that your business has already vetted and valued. The impact this has when scaling video spend is transformative. It allows you to increase your budget with confidence because you’re no longer just driving up volume; you’re driving up value. It’s the key to keeping customer acquisition costs stable, or even decreasing them, as you pour more money into video.
Last-click attribution models often undervalue video ads, leading to misguided budget cuts. How can teams set up a simple lift measurement or holdout test to prove video’s true impact? What directional metrics should they monitor to confirm that increased video spend is driving overall growth?
The last-click model is video advertising’s worst enemy. It’s a painful conversation to have when a finance team looks at a report and says, “This YouTube campaign has a terrible ROAS, let’s cut it,” because they can’t see the whole picture. They don’t see the user who watched the ad on their phone during their commute, became aware of the brand, and then searched for it directly on their laptop a few days later. In that scenario, last-click gives 100% of the credit to brand search, and zero to the video ad that created the demand in the first place. When the video budget gets slashed, it’s no surprise that brand search volume often mysteriously dries up a few weeks later.
Instead of getting bogged down in complex attribution debates, a simple lift measurement or holdout test can tell a much clearer story. Google’s own lift measurement tools make this relatively easy by splitting your target audience into two groups: one that sees your video ads and one that is held back and doesn’t. You can then measure the difference in behavior between the two groups to prove the true, incremental impact of your campaigns. For a more directional, ongoing approach, you can run a simple test: increase your video spend by a significant amount, say 20%, and monitor your blended, overall customer acquisition cost. If your total revenue grows while your blended CPA remains stable, that’s a very strong signal that the video investment is efficiently driving overall business growth, regardless of what the last-click report says.
With so much video content viewed on mute, a clear visual message is critical. Within the first three seconds, what three questions must the ad answer visually? Describe how AI-based tools can pre-test creative to ensure brand assets are prominent enough to drive performance.
Absolutely, we have to assume sound is a bonus, not a given. Many people are scrolling through feeds in public places, on transit, or late at night, and the sound is off. If your message depends entirely on a voiceover, it’s a message that’s never being heard. The creative has to do the heavy lifting from the very first frame. Within the first three seconds, a viewer looking at your ad on mute should be able to instantly answer three fundamental questions just from the visuals. The first is, “What is it?”—the product or the brand needs to be immediately visible and understandable. The second is, “Who is it for?”—the creative needs to signal the target demographic so the right people feel seen. And the third is, “What do I do?”—there has to be a clear, visible call to action or next step.
This isn’t just guesswork anymore. We can now use AI-based creative analysis tools to pre-test our ads before they ever go live. You can upload your video, and these tools will use object recognition to analyze it frame by frame. They can tell you if your brand logo is clearly detectable within the first 25% of the video, or if the product is obscured. If the AI can’t even “see” your brand, it’s a huge red flag that the ad won’t be classified properly by the ad platform’s own systems and that your brand lift performance will be abysmal. This kind of pre-flight check is becoming essential for ensuring your creative is built to perform in a silent, fast-scrolling world.
What is your forecast for the role of the PPC manager as an “architect” of AI systems? What new skills will be most critical for success in this evolving landscape?
My forecast is that the “architect” role isn’t just the future; it’s the present for the most successful teams. The days of the PPC manager as a “pilot,” constantly in the cockpit pulling levers and making manual bid adjustments, are over. That approach is simply too slow and inefficient to compete with systems processing millions of signals every second. The new role is far more strategic. The most critical skills will be data stewardship and creative strategy. Instead of tweaking keywords, the best managers will be experts at curating and feeding the algorithm high-quality signals, like connecting a CRM to provide value-based conversion data. They’ll also need a deep understanding of creative, not necessarily to produce it themselves, but to direct the creation of modular asset libraries that give the AI the flexibility it needs to work its magic. Success will be defined not by who can click buttons the fastest, but by who can design the most intelligent and well-supplied environment for the AI to operate within.
