The days relying on PPC analysts are over. Or at least that’s what many in the industry believe. When looking at the current digital marketing landscape, I think the advent of AI/Machine Learning-based bidding actually complicates, rather than simplifies, AdWords.
Along with the increased nuance of attribution options, AI requires agencies and marketers to give up significant specific control, and it also takes a much more qualified marketer to use these strategies to their full effect. If a bad campaign is created, using bad data and bad creative, it doesn’t matter how good the AI is at optimizing when to show an ad, it will still fail. In this article I want to examine two pieces of the overall PPC landscape that I am a fan of (Smart Bidding/AI and Attribution Modeling), and believe will make all the difference when it comes to what good digital marketing actually means.
Attribution Modeling Review
First let’s break down attribution and why it matters. The default attribution model for AdWords is Last Click, though I typically push for Time Decay or Position-Based Modeling unless there is a specific option. This distinction may seem arbitrary to most analysts, as you can compare, contrast and rerun your campaign results to analyze and report the numbers as you see fit. However, the use of these models and the advent of Google’s Data Driven modeling make a monumental difference in how they are set if you are going to take advantage of Google’s AI technology.
Identifying all the goals in a customer journey and identifying which attribution model should be applied to each is necessary from the beginning. If any of these steps isn’t set in an effective manner, each data point analyzed will push you further away from an effective campaign instead of closer.
AI/Machine Learning Case Studies
As an example, my agency works with a volunteer organization that was in an early beta of Maximize Conversions Smart Bidding. Most of their campaigns performed exceptionally well under these conditions, however one campaign was targeting a very niche subdomain audience without enough data. So rather than improving the conversions we had defined and manually gathered, the AI simply stopped bidding in auctions. It took several weeks to recover to the previous level of conversions.
This same campaign also illustrates the other end of this spectrum. After this initial crash we defined lesser, engagement based goals. To these, we assigned value for our own reporting (though Max Conversion bidding doesn’t take this into account, rather bid types such as Target ROAS), targeted manually until we had some of our real conversions and enough of the lesser conversions that an automated bid strategy would have enough data, then tested again.
Here we see an initial surge in clicks and conversions, a drop back to previous levels, followed by significantly increased average days. This strategy is not for every account or campaign, however, assuming you have a similar niche and a budget to target lesser goals in the short term, you can see significant results. Additionally worth noting, the increase is both in engagement conversions and signups in roughly equal ratio.
Typically, it is best to reach beyond a last click attribution model when using machine learning in order to give the system more data to work with (Google recommends “that advertisers have at least 30 conversions in the past 30 days before using Target CPA… 50 conversions in the past 30 days for Target ROAS”. The last example used Time Decay Attribution, however this isn’t always the best as the next case explains.
Over the course of a year our agency had used manual techniques to promote magazine subscriptions tracked via outbound links to off site checkout. Before making the leap to a Automated Bid Strategy we analyzed how and when users actually converted. We found that the majority of users that converted did so on the first time, on the page, rather than during an ongoing research process. Using this analysis, we activated a maximize conversion bidding with the effect of increasing weekly conversions from 15 to 35 within one months time while only increasing average CPC by ten cents.
This only scratches the surface of either the benefits of using AI assisted campaigns, and of the necessity of good PPC analysts to make them effective. It doesn’t touch on the creativity required, the ads written or designed, or pattern recognition skills needed to see when the AI is learning from the wrong data, but I hope it does spark a few ideas that improve how we all use these tools and remember that each of the settings matters.
Sean Kerr is the COO & CoFounder of Cause Inspired Media. Working with around 250 nonprofits across the country ranging from community foundations to international initiatives gives him a unique perspective on the nonprofit fundraising landscape. This perspective is tempered by realistic expectations of digital advertising and a strong team of support. You can reach Sean via LinkedIn.