Smart Bidding is a familiar application of artificial intelligence that Google uses to facilitate paid search advertising. Smart Bidding has automated digital advertising, enabling businesses to reach their target audience effectively and optimize their campaigns. In this article, we will delve into the inner workings of Google's AI algorithm and explore how businesses can set it up for success to unlock the full potential of paid search optimization.
Understanding Google's AI Algorithm
The idyllic version: Google's AI algorithm utilizes machine learning and artificial intelligence to optimize paid search campaigns. At its core, the algorithm analyzes vast amounts of data, including historical campaign performance, user behavior, search patterns, and contextual factors. By processing this information, the algorithm identifies patterns, trends, and insights to make real-time adjustments and optimize ad placements, bids, and targeting strategies.
The realistic version: Everything above describing Google’s Smart Bidding is all true, but it’s also a very optimistic version of what could happen. About like if I described a chainsaw as “an automated way to harvest earth’s bounty using decades of engineering know how.” Sure. That’s also true, but it doesn’t keep the saw from binding in the tree, the chain from coming off if it’s not tightened properly, or the tree from falling in an unexpected direction. To harness the full potential of Google's AI algorithm it’s useful to understand what often goes awry with machine learning algorithms, what signs to look for to see if it’s happening in your account, and how to avoid the issues. To be sure, Smart Bidding is a black box, which is to say that Google doesn’t reveal the formula, or exactly what data is being used, but I’ve built hundreds of models myself and will share with you some key principles for setting up Smart Bidding for success.
Conversion Values are Key: The Basics
What Smart Bidding optimizes for is what data scientists would call your dependent variable. The good news is that it’s the most important part of any model and you have full control over it in Google Ads. The reason it’s the most important part of your optimization is simple. Everything done within any optimization is an attempt to obtain more of the goal you set. Therefore, if the goal set is only a 50% accurate representation of what’s valuable to your business, then the optimization can never be more valuable than that, and given that it’s an estimate, it will be less.
Google provides you with several options to set goals which they call Bid Strategies. For most businesses they are from worst to best: Target Impression Share, Max Clicks, Max Conversions, and Max Conversion Values. There are variations within those such as Target ROAS that work off conversion values. There’s an exception to every rule, but most of the time, you want to be using conversion values, which can be set up to be the best representation of value of your business. For example, if you’re an online coding bootcamp, you could set enrollments in your immersive program to be worth the most, enrollments in prep courses to be worth the second most, leads 3rd, etc. Compare that to using target impression share and only knowing the person saw an impression of your ad. That can be like building a McDonalds next to a highway where it’ll be seen, but with no exit ramp so it gets no sales.
Conversion Values: Advanced Concepts
Sticking with the online coding bootcamp example, let’s assume you’ve established several conversion actions, looked at the probabilities of going from a lead, to an application, to an enrollment, as well as examining the dollar values of different kinds of enrollments. You’ve then used all that to set appropriate conversion values. What else should you consider? Three important considerations are: tracking fidelity, data outliers, and non-marketing driven variation.
1. Tracking Fidelity
Google provides several mechanisms to track your conversions. It’s a huge mistake to just assume that they’re working. Your conversion values are only as good as the rate at which they are accurately recorded and tracked back to your advertisements. If Google can’t track the advertisement as being associated with a conversion, then it won’t optimize for more of those. Unfortunately, with the changing digital landscape, privacy laws, digital ad blockers, VPNs, cookie permissions, etc. a lot of the tools Google has provided can become dysfunctional even if you’ve closely followed the instructions. It’s important to verify the number of conversions being recorded and whether they’re being recorded under different circumstances.
2. Data Outliers
A lot of people have heard of data outliers and the outliers in many examples are data errors. However, even accurate data can have outliers that can diminish Google’s ability to optimize to your goals. Imagine that you’re an online university and you normally sell to individual students. But one day a corporate customer clicked on your ad and ended up enrolling 500 of their employees for a course. That’s a great sale and you’d be crazy not to want some more of those. That said, a single lucky sale doesn’t represent a reliable relationship to use in optimization.
Realistically Google may know that the user was likely a woman, using an apple device, and a safari browser, who clicked your link about “Spanish Courses” at 11am. No matter what the specifics happen to be, it’d be unwise to send all your ad dollars chasing the profile of a single click and that’s what will happen if you weight that sale as being 500 times more important than the others. It would be more productive to cap the max value in your consumer campaign and start a new campaign focused on corporate customers with key words likely to consistently be used by them.
3. Non-Marketing Driven Variation
Marketing can drive a lot of behavior, but other things are outside its ability to influence. It’s advisable to purposely exclude sources of variation not largely influenced by marketing from your conversion values. An example could be the final enrollment status of students in some states. There are some states that have enrollment caps for K-12 online students, but also have a higher demand than the number of available seats. In those cases, they might have a lottery to see which students get the seats. Clearly the lottery outcome isn’t impacted by marketing. In a case like this, setting the conversion value of a final enrollment, higher than an application, can hurt optimization, because it’s tasking the optimization to obtain more enrollments which it has no power to do. It only sends the optimization on a wild goose chase.
The lottery example is clear cut, but the problem can be more nuanced. There are often cases where marketing may have a minor influence, but the uninfluenced sources of variation are so much larger that it’s still not worth including it in your conversion values.
Artificial Intelligence Dreaming
It’s a mistake to think of AI as intelligent. AI is a cool buzzword, but thinking of it as “intelligence” can lead to some poor business decisions. A less glamorous, but more realistic way to think about Smart Bidding, is as a factory conveyor belt with robotic arms building something, all behind a big black curtain. There is a lot of “intelligence” built into the algorithm running the robot arms at the factory, but the robot arm isn’t intelligent per se and can’t adapt to anything beyond what it’s programmed to do.
The reason this is an important distinction is that I’ve repeatedly heard people daydream about what Google probably knows, and what the AI might be doing. And most critically, assuming that a potential problem can be ignored because they hope the AI will take care of it. IMO a more appropriate phrase to remember is that “Where there’s mystery, there’s margin.” Google’s main business goal is to make money for Google. It’s not safe nor prudent to not check that things are working the way they are supposed to work.
Google Smart bidding has automated a lot of tasks and improved the productivity of paid search advertising for a lot of people. To get the most out of it you should utilize conversion values to accurately reflect business value while taking into consideration tracking concerns, data outliers, and what marketing can reasonably be expected to influence. Smart Bidding can be like a race car. The faster you drive it the more you need to keep your eye on the road.
Take advantage of our experience. Shoot me an email and let’s have a chat about how to improve your media performance.