How Does Machine Learning Work in Paid Search Marketing?
All modern ad platforms are now embedding machine learning into their algorithms. Running successful campaigns requires an understanding of the machine learning in each ad network.
This Ask a PPC Question, Who Chote Lal in New Delhi, It is important for account managers and those who report to:
“How does Google’s machine learning work in paid marketing?”
In this column you will learn:
- What is machine learning?
- How does machine learning affect paid search campaigns?
- How to improve machine learning on paid search.
Since the question was specifically about search, we’ll focus on the uses of search first.
What is machine learning?
Algorithms for information processing are taught through machine learning. The more data, the faster it will learn what to do with that information.
Different data points can carry different weights in the algorithm. It is important to understand how to evaluate data points.
Data points can be completely objective, subjective, or a combination of human interaction and pure algorithmic learning.
Knowing what you can control is critical to your success when you are an ad network machine learning partner.
Another critical factor is the learning period (and that the algorithm is given enough time to process the data points).
How does machine learning affect paid search campaigns?
Machine learning affects almost all paid searches. Any significant change can affect how the algorithm processes your campaign.
These changes include:
- Bids and budgets: Radical changes in budgets or changing bid strategies.
- masses: Changing goals or excluding goals.
- creative: Changing or adding the creative creates a new version of the ad that does not have access to the old ad’s stats.
- campaign status: Pausing campaigns resets the learning period.
It is important to note that manual campaigns are not affected by these changes, however, it does get more difficult to run purely manual campaigns.
Running a manual campaign means opting out of over 60 signals that ad networks use in Smart Bidding.
These signals are used to adjust bids according to the chosen bid strategy and budget.
Additionally, while the verdict is still out on whether Expanded Text Ads (ETAs) or Responsive Search Ads (RSAs) work better, RSA tends to have higher impression share.
Machine learning is not always an active option. Matching keywords and audience tags happens in the background, and is based on historical data.
The original audiences (in market, affinity, etc.) depend on the algorithm learning that people who complete one action are more likely to complete another / have other associated attributes.
When you ask the ads platform to find “similar” audiences for a list of uploaders/website visitors, you are using the primary audience to help the advertising platform understand which prospects you find valuable and which are not.
Matching keywords and close variants are affected by the probability of profitable results, as well as real-time user behavior.
Algorithms are now smart enough to know if a user is bilingual and will allow their other language to run ads.
How to improve machine learning for paid search
Optimization is a lot easier when one gets to know machine learning in paid search.
The most important mechanic is honoring learning periods and avoiding accidental resets.
If you need to scale a campaign, for example, be sure to set aside two weeks between each major budget increase.
If you want your campaign to slow (or stop), lower the budget rather than pause so you don’t reset the learning period.
Keywords and negative audiences can help an ad platform’s algorithms understand which thoughts and behaviors should align with budget (and which should be avoided).
This is the most powerful way to influence machine learning and should be a part of all paid search accounts.
Conversion values and conversion are two underutilized machine learning tools. It is the easiest way to communicate with the paid search algorithm and allows you to view user behavior without requiring the ad channel to rate the action.
Machine learning affects almost all elements of paid search, and understanding how to teach the algorithm is critical to PPC success.
- The Automated Pay-Per-Click Year: 3 Survival Tips
- PPC Today – Let go of what we’ve learned
- PPC 101: A Complete Guide to the Basics of PPC Marketing
Do you have a question about PPC? Submit via this form or tweet me at @navahf using the hashtag #AskPPC. See you next month!
Featured image: Paulo Bobita / SearchEngineJournal