Considerations for using AI in paid search

AI has become all but ubiquitous as a marketing tool.

Forrester Research has predicted that businesses who use AI to drive marketing will gain $1.2 trillion per annum from those who don’t. According to McKinsey, 35% of Amazon’s revenue and 75% of what customers watch on Netflix is drawn by AI-driven product recommendations.

Email marketers are using automation for one-to-one marketing: dynamic content is generated based upon real-time, behavior-based metrics. In addition to directly improving ROI, AI can also reduce campaign production time (and expense), freeing email marketers to focus on higher-level tactics and strategy.

In the programmatic space, automated bidding has been proving its worth for quite some time: the combinations and permutations that result from multiple ad exchanges, in which multiple ad sizes and formats can appear, means that an attempt to manually oversee every single auction would be problematic, to say the least.

When it comes to Paid Search (SEM), conventional wisdom holds that all we need to do is flip an account’s automated bidding “ON” switch, then sit back and watch the return on investment roll in. However, while many SEM automated bidding programs are much improved, they still come with potential drawbacks.

• Most algorithms only use performance data from within Paid Search to make their keyword level bid changes, usually up to several times a day.
• Many SEM keywords not only convert themselves, but assist in other conversions (e.g., brand, organic, direct, etc.).
• Because these assist values are assigned over time, it means they cannot inform in-day bidding decisions.

What this means is that an automated bidding strategy can “work” within Paid Search, yet because the bidding algorithm didn’t give high-assisting keywords the appropriate consideration, overall performance can and does suffer, often because of decreased organic and direct demand.

One workaround is to assign high-assisting keywords higher performance goals against which the algorithm will optimize: so for example, allow high-assisting keywords a cost per conversion goal of $10 when the account average is $7. This can work provided that the traffic volume in the separate keyword groupings is high enough to provide the bidding algorithm with sufficiently sized data sets.

For smaller- and medium-sized accounts however, the choice is often between large keyword portfolios optimized to one goal, with the risk that cross-channel performance may suffer, and highly granular goals that may result in some keywords performing sub-optimally due to limited data sets.

Ultimately, the solution to automated SEM bidding is to continue testing within higher-level strategies, making sure to always measure cross-channel impact prior to declaring a winning strategy.

For more on how we can help your company optimize its SEM program, contact DigitalCX_Search@epsilon.com

**Brad Giddens also contributed to this post.