Milena Traikovich is a powerhouse in the world of demand generation, recognized for her surgical precision in campaign analytics and lead optimization. With her finger on the pulse of evolving PPC landscapes, she has spent years decoding the friction between human intuition and machine learning algorithms. Her work focuses on helping brands move beyond superficial metrics to drive high-quality leads through rigorous performance optimization and strategic foresight. Today, she sheds light on a counterintuitive reality: the more we try to “help” Smart Bidding during massive shopping events like Black Friday, the more we might be sabotaging our own profit margins.
In this deep dive, we explore why traditional holiday bidding overrides are increasingly backfiring and how consumer shopping habits are flattening the once-sharp peaks of Cyber Week. We discuss the transition from manual “overrides” to strategic “guardrails,” highlighting the crucial role of AI assistants in auditing account health and maintaining data hygiene. Milena also breaks down the performance disparities between major shopping days and provides a framework for keeping a “human-in-the-loop” when managing complex, automated portfolios.
Data shows that manual seasonality adjustments during major holidays often lead to a 34% spike in costs. Why is this specific intervention becoming a liability rather than an asset?
When we look at the data from over 5,800 accounts, it becomes clear that telling Google to expect a conversion lift is essentially telling the system to bid aggressively at any cost. In 2024, advertisers who used these manual adjustments saw their cost-per-click rise by 34%, which is exactly double the 17% increase seen by those who let the algorithm run naturally. It feels like a gut punch when you see your ROAS fall by 15.7% while your competitors—those who stood back—actually watched their ROAS rise by 5.7%. The fundamental issue is that these adjustments act as a blunt force instrument in a world that requires a scalpel. You are essentially paying a premium for volume that the system was already prepared to find, leading to a massive loss in efficiency for a very marginal gain in conversion value.
We’ve seen the “Black Friday peak” start to look a bit different in recent years. How is the flattening of the promotional season affecting our ability to forecast conversion rates accurately?
The era of the single-day shopping frenzy is effectively over, as shoppers are now spreading their purchases across a much longer promotional window. If we look at the trajectory from 2022 to 2024, the conversion rate lift during Black Friday and Cyber Monday has been steadily shrinking. In 2022, we saw a 17.5% bump over the baseline, but by last year, that lift had dwindled to just 7.5%. When an advertiser sets a seasonality adjustment based on what they think a “typical” Black Friday looks like, they are almost always overestimating the current reality. Overbidding for a 20% lift that only turns out to be 7.5% means you are overpaying for every single click during the most expensive week of the year.
If Google’s Smart Bidding is already “aware” of major holidays, what is actually happening under the hood when a human adds a manual seasonality adjustment on top?
It is a classic case of double-counting the signal, and it creates a feedback loop that drives bids into the stratosphere. Smart Bidding is already trained on years of historical data across millions of advertisers, so it knows exactly when the holiday rush starts. When you layer an adjustment on top, you’re essentially shouting at the system to anticipate a surge it has already accounted for. This leads to the algorithm treating your input as an additional lift on top of the organic holiday trend, forcing it to bid twice as hard as necessary. Instead of a helpful nudge, it becomes a redundant command that prioritizes spending your budget over maintaining your profit margins.
Since manual adjustments can be so volatile, what is a more effective way for a business to signal its willingness to trade efficiency for more volume?
If you want to be aggressive, you should change your targets, not your forecast. Lowering a Target ROAS or increasing a Target CPA is a much more direct and transparent way to tell the system that you are willing to pay more for a conversion. It’s the difference between telling a driver to “expect more traffic” and telling them “it’s okay to take the toll road to get there faster.” By managing the target itself, you remain in control of the efficiency threshold rather than letting a seasonality adjustment guess at a conversion rate that may never materialize. This approach keeps the algorithm grounded in real-time auction signals while respecting the new economic boundaries you’ve set for the holiday period.
You’ve advocated for using AI assistants to set “guardrails” instead of overrides. How does this shift the daily workload for a PPC manager during the holiday rush?
The job shifts from being a fortune teller to being a specialized auditor, and that’s where an AI assistant connected to your data, like the Optmyzr MCP, really earns its keep. Instead of spending a week manually exporting spreadsheets, you can spend an afternoon asking a language model to identify every scheduled change across a massive portfolio of retail accounts. You can prompt the assistant to list every seasonality adjustment or budget change made in the last 21 days, ensuring that no “just in case” settings are left to sabotage your performance. This allows you to focus on the business context—deciding if the changes are actually ready for the peak—rather than getting bogged down in the retrieval of the data itself.
One of the most overlooked risks in automation is the quality of the conversion signal. How can “dirty” data lead to the same overbidding problems we see with seasonality adjustments?
If your conversion tracking is double-counting or triple-counting purchases, you are essentially lying to the algorithm, and it will punish you for it. We often find accounts where multiple primary conversion actions are firing for the same sale, tricking Smart Bidding into thinking it’s twice as successful as it actually is. This inflated value causes the system to overbid in exactly the same way a manual adjustment does, but it’s even more dangerous because it’s invisible unless you’re looking for it. We use AI agents to compare Google Ads purchases against GA4 data for the last 30 days to rank accounts by the size of the gap. If the discrepancy is significant, you have to fix that signal before you let the machines take the wheel for a major sale.
Looking at the performance of the 2025 shopping season, what did the data reveal about how we should be allocating budgets between Black Friday and Cyber Monday?
The performance gap between the two days is wider than most people realize, and last year’s data from over 5,000 accounts proved that Black Friday is still the undisputed king. We saw that Black Friday generated $144.8 million in conversion value compared to $107.9 million on Cyber Monday—a 34% difference. Even more importantly, Black Friday maintained a much higher ROAS at 835% versus 674% for Cyber Monday. If you are holding back a significant portion of your budget for Monday, you are likely arriving late to the party and paying more for less efficient results. The smartest move is to identify the campaigns that hit their budget caps while still producing high ROAS on Friday and ensure they have the fuel they need when the demand is at its peak.
As accounts move toward consolidated structures like Performance Max, how can advertisers maintain control over product categories with vastly different profit margins?
Consolidation is great for data density, but it can be a disaster for business strategy if you’re discounting different categories at different rates. If one product group is marked down 40% while another is only 10% off, they shouldn’t be fighting for the same budget in a single PMax campaign. Segmenting these products into different groups allows you to set specific ROAS targets and budgets that reflect their actual margins and inventory levels. We use AI to monitor which product groups are gaining spend versus which are losing efficiency in real-time. This level of segmentation gives you a “kill switch” for specific categories that might sell out by Saturday afternoon, preventing you from wasting spend on out-of-stock items.
You often mention the “Human AI Sandwich” as a best practice. How does this framework prevent the types of automated errors that can ruin a campaign’s performance?
The “Human AI Sandwich” is about ensuring that an experienced professional is present at both the start and the end of every automated process. A human defines the problem and the policy—deciding, for example, that spend can never run more than 20% ahead of pace without an alert. The AI then does the heavy lifting in the middle, processing millions of data points to find the anomalies or suggest budget shifts. Finally, a human evaluates that output before anything actually goes live in the account. This prevents the “planet average” mistakes that occur when AI is left to make inferences on its own, like misinterpreting a drop in volume as a success just because ROAS stayed stable.
What is your forecast for the future of holiday campaign management as automation becomes even more sophisticated?
I believe we are moving toward a “policy-based” management style where our primary job will be defining the boundaries within which the machines operate. We won’t be setting individual bids or even daily budgets in the traditional sense; instead, we will be writing sophisticated rules and alerts that fire the moment an anomaly is detected. By 2027 and 2028, the competitive edge will go to the advertisers who have the cleanest data signals and the most robust “human-in-the-loop” approval workflows. The real winners won’t be those with the most complex manual strategies, but those who have built the best guardrails to let the AI run at full speed without flying off the tracks.
