Maximizing AI in PPC Bidding: When Automation Works and When It Doesn’t

Automation and AI-driven bidding have revolutionized PPC campaigns on platforms like Google Ads and Meta. These advanced tools promise enhanced efficiency and better results, but they are not a one-size-fits-all solution. For AI to truly maximize your advertising efforts, specific conditions must be met. Without the right data, even the most sophisticated AI can falter, sometimes leading to worse outcomes than manual bidding. In this post, we’ll explore when AI-driven bidding excels, when manual control is preferable, and how to strike the perfect balance to optimize your PPC campaigns.

‍

AI-driven (automated) campaigns work really well when you have three things:

1. A high volume of conversion data
2. High variance in traffic quality
3. Tracking that accurately reflects real business outcomes (RBOs)
   

1. High Volume of Conversion Data:

The more conversion data you have, the better the AI will be at consistently maximizing results.

With smaller data sets (fewer than 30 weekly conversions per campaign), it’s preferable to have less variance in the data and ensure these conversions come from your most valuable customer profiles. This improves statistical significance and helps the AI “exit the learning phase” (increase confidence and accuracy) more quickly.

For example, training an AI to predict the probability of a flipped coin landing on tails—where there are only two possible outcomes—is easier than training it to predict the probability of drawing the ace of spades from a 52-card deck.

In a marketing scenario, if you’re only getting 10 weekly conversions, you’d want those conversions to come from your 10 best customers who share similar attributes. It’s more effective to train an AI on conversions from 10 customers who are all avid runners buying high-end running shoes than from 10 completely different people purchasing various unrelated items. With only 10 conversions to model from, the cost to acquire the next avid runner will be lower than if the AI were attempting to find the next random person interested in different products.

Additionally, you’d prefer your 10 conversions to come from your ideal customer profile, buying products that are core to your larger marketing strategy. For example, the YETI brand primarily sells premium insulated coolers and coffee mugs, but they also sell merchandise. YETI’s Google Ad campaigns would be at a disadvantage if their first 10 conversions were from purchases of a $20 YETI-branded t-shirt—this wouldn’t help the AI get better at profitably selling premium coolers and coffee mugs.

With larger data sets (30+ weekly conversions per campaign), more variance in your data is preferable. In this training environment, the AI is more likely to understand the signal clusters that produce the best results, which will help drive consistency.

Implications: We recommend a consolidated campaign structure. This approach offers multiple benefits, including budget flexibility and more effective AI training, which increases accuracy and confidence.

• In low-conversion (or new) accounts, we favor tighter account structures with more control, such as using Exact Match keywords and restricting ads during certain times of day or days of the week. This helps control the learning environment to find the best data.

• In high-conversion accounts (30+ weekly conversions per campaign), broader targeting—liquidity—is preferred.

‍

2. High Variance in Traffic Quality

Not all traffic is created equal. Two people searching for the same keyword at the same time might have very different customer profiles, leading to significantly different expected conversion rates and values. AI excels at understanding these individual nuances and can consider external factors when determining real-time expected conversion rates.

For example, an Instagram ad served to a user standing in line at the grocery store might have a much lower expected conversion rate than if the same user were at home later at night, in a setting where they’ve historically converted through Instagram ads.

The AI can increase bids for users with higher expected value or lower bids (or refuse to bid) for lower-value auctions. This typically results in a higher blended CPC/CPM but better overall results in the long term. However, in cases where traffic cohorts have little variance, AI might not be more effective than a manual approach.

Implications:

• This is often the case with Branded Search. For many brands, customers searching for branded keywords have similar expected conversion rates.

• In a non-brand example, we found that low variance exists for phone repair keywords. Users with a broken iPhone searching “phone repair near me” have a similar conversion rate when seeking directions to a local repair shop. This makes sense, as users with a broken iPhone are highly motivated to get their phone repaired quickly (they mostly fall within a few degrees of variance). Therefore, a manual approach to bidding and optimization can likely help keep CPC costs low and generate similar results over time.

It should be noted that, in the above example, the conversion goal is a user calling a repair shop, scheduling an appointment, or entering the store’s address into Google Maps. We do not have the ability to track the actual revenue earned by the business, which is part of the reason why we don’t see variance in traffic quality and why the manual approach is more effective. If, instead, we were able to optimize for “closed sales” or “revenue,” we would likely benefit from AI optimizing budgets and bids in real time.

This brings us to our final point:

‍

3. Tracking That Accurately Reflects Real Business Outcomes (RBOs)

AI-driven campaigns work best when your conversion actions are directly tied to business results. For instance, an e-commerce brand would prefer to optimize exclusively for “purchases” rather than “add-to-carts.” Even if 20% of add-to-carts consistently translate to purchases, this percentage is unlikely to hold at scale with an AI-driven campaign. The AI might excel at finding individuals (or bots) who complete the add-to-cart action, but these would convert into purchases significantly less than 20% of the time.

In a recent edition of the "Digital Download Newsletter," Sam Tomlinson articulates this point by stating, "Any campaign that bids on something other than the optimal target is deploying capital suboptimally - even if performance is the same, the risk profile of bidding on (for example) form submissions instead of MQLs/SQLs raises the cost of that campaign (risk isn’t free!)."

Advertisers should strive to directly optimize for real business outcomes (RBOs), especially when leveraging automation. If you're not doing this, but your competitors are, you'll be at a significant disadvantage.

However, for many reasons, brands sometimes lack the infrastructure to optimize directly for RBOs. In this case, leading indicators like add-to-carts or form submissions are helpful. The absence of direct conversion clarity should not prevent a brand from leveraging AI campaigns; however, they must proceed with caution. An AI-driven campaign with an imperfect conversion-tracking system is still often better than a manual campaign.

Implications: Most clients driving traffic to a retail location have an imperfect system. In these cases, we might rely on leading indicators like “Get Direction” clicks, calls, and scheduled appointments. For some brands, we have Store Visits, which is better than a “Get Direction” click, but this action still does not guarantee revenue. Moreover, when users become customers, we often lack clarity on which product or service they buy, or the associated average order value (AOV).

Therefore, we often optimize for the lowest blended cost per acquisition (CPA) targets and use analog feedback mechanisms to adjust our targets over time. While imperfect, this feedback loop is still helpful.

‍

In summary, AI-driven bidding can significantly enhance the performance of your PPC campaigns on platforms like Google Ads and Meta, but it’s not without its caveats. Ensuring you have a high volume of conversion data, high variance in traffic quality, and tracking that accurately reflects real business results is crucial for automation to succeed. While AI can offer remarkable efficiency and insights, there are scenarios where manual bidding and tighter controls may be more effective. By understanding these nuances and leveraging the strengths of both automated and manual approaches, you can achieve optimal results and maximize your ROI in digital advertising.