PPC Machine Learning

Machine learning is beginning to transform how PPC campaigns are managed.

Three years ago, we managed the majority of our campaigns using different types of manual bidding; sometimes extending that when there was enough conversion data, to use “enhanced CPC” bidding.  But today, we design all of our clients’ campaigns with automated bidding as the planned outcome, once the campaign has run long enough to generate a conversion history (usually 2 to 6 weeks).  This requires 100% conversion tracking implemented, and zero conversion leaks” on landing pages and websites because that facilitates getting the campaigns onto automated bidding.

Automated bidding can often deliver a marked increase in campaign performance – ROAS.  We have seen as much as a 300-400% improvement in campaign performance once the transition onto automated bidding is put into place.

How does automated bidding work?

By adopting automated bidding in your campaign, we are creating a “feedback loop” on conversion information (including keywords) back into the campaign. The automated bidding algorithm is a machine learning algorithm that essentially turns your campaign from an open-loop system, into a closed-loop system.

There are different types of automated bidding algorithms that Google and Microsoft provide.  Three of our favorites are Target CPA, Maximize Conversion Volume, and Maximize Return on Ad Spend.

The Target CPA algorithm requires at least 90 conversions over a month, to “gain traction”.  It looks back at the past 30 to 90 days of conversion data, extracting various information from those conversions – things like the Geographic location, the person’s browsing history, the time of day, etc.  Google then essentially solves a matrix for a multi-variable equation to determine whether to bid and if so how aggressively to bid.

Optimizing a Target CPA to the right CPA level, the one parameter you must seed the algorithm with can be a little bit tricky, and is a moving target.  Maximize conversion volume is a simpler algorithm and can often be used before a campaign has generated enough conversion data to deploy Target CPA, but usually with less impressive results.  Target Return on Ad Spend is another Google Automated algorithm, which attempts to maximize the value of conversions for a given ad spend.  With Return on Ad Spend, you must quantify the value of each conversion, which can vary with product and service types.

Using Google to quote, “place Chess for you”, by setting up your campaign to track 100% of conversions, then getting your campaigns onto automated bidding as quickly as possible after launch, is almost always a winning strategy.  We have seen dramatic, even breathtaking leaps in campaign performance while pursuing the approach.

Despite that, some clients are resistant to implementing 100% conversion tracking, especially telephone call tracking, and eliminating conversion leaks such as openly displayed email addresses on the website, or adopting “pool numbers” to display.