Machine Learning & Paid Search Campaigns

How Machine Learning can Transform your Paid Search CampaignSince the beginning of search, assigning value to the effectiveness of specific paid ad campaigns was fairly subjective. Until the rise of machine learning, most analytics software and marketers have been unable to decipher user intent and click paths that led to conversions.

Google AdWords base ad campaign effectiveness and potential keyword rank on their Quality Score algorithm. Yet, even AdWords was picking the winners and losers of its own search contests fairly subjectively.

Credit for sales and conversions was primarily based on the rules of its own attribution model, which was unable to fully interpret the different factors of a click path that led to a final conversion. Marketers were falling for the pitfall of the Google AdWords attribution model.

The Pitfall

Attribution models are simply the rules that govern how your analytics software will give credit for a conversion. Google AdWords typically followed a last-click attribution model, which simply meant that the last click that led to a conversion was given all of the credit. That’s fine if your marketing team wants to pat itself on the back for that winning, branded keyword that performs so well.

Except we must understand that user click paths, which lead to a conversion, can be highly complex, often involve clicking on multiple ad copies, typing in different searches, and even browsing a few different websites. A last-click attribution model would ignore all of this and simply give credit to the last click.

This is a fairly complicated concept to understand. For example, if your business sells women’s hats, users may conduct multiple searches for different styled hats, such as “women’s winter hats,” click on your ad copy and then retype the same search query, only this time mentioning your brand.

This arbitrary set of rules can make adjusting your ad campaigns accordingly to user intent quite impossible without all of the given data outputs. You might find that one keyword has a 9:1 return and another has a 4:1 return. While the former keyword may appear to be performing poorly, scrapping that keyword altogether from your campaign may have an unknown effect on the latter keyword.

Google’s Attribution Model


Google’s attribution model typically followed a last-click attribution model. Yet, this is a highly ineffective model for evaluating user intent and evaluating how user search paths impact a last-click result. Credit for a conversion would be based on the last-click which led to the conversion, whether it be from a paid search campaign or your Facebook page.


First-click attribution assigns 100% of a conversion credit to the first click of a sales or conversions tunnel. This model was effective for companies who relied too much on branded keywords and wanted deeper insight into their user intent.


Under a rules based attribution model, time-decay attribution assigns credit to the points closest in time to the conversion. This could mean if somebody encountered your ad a week earlier and then navigated back to your website a week later from your social media page to make a purchase, the social media page would receive the credit for the attribution.


Linear attribution models distribute credit evenly to all of the touchpoint outputs across a conversion path.

Position-Based Attribution

This attribution model grants the first and last-click of a conversion 40% of the credit each. The touchpoint outputs that occur in between are given 20% of the credit as a whole.

It’s pretty easy to spot how this rules-based system was virtually incapable of giving insights to marketers about how different touchpoint outputs across a conversion path factored into their conversions funnel.

Data-Driven Models

As machine learning becomes a more powerful force in search marketing, attribution models are shifting to a multi-touch attribution model (MTA). MTA models rely on an analytics maturity scale, which is more descriptive and able to grant consideration to different touchpoint outputs in real-time. This model thus analyzes a conversion path and grants credit to different touchpoints based on its rule set.

The data-driven model uses the advanced capabilities of machine learning to assign credit to different touchpoint outputs across a conversions funnel more proportionately and objectively. This technology allows marketers to better understand how different factors, such as a separate set of ad copy and keywords, could factor into making a conversion.

Data-driven models analyze the click-paths of users who converted and matches them with the paths of those which didn’t to grant consideration to the different touchpoint outputs which factored into a user conversion. This model takes many factors into account, such as the number of ad interactions, order of exposure, ad creative, and other factors which determine the keywords and clicks most effective at driving conversions.

Data Requirements

In order for Google to create a model for your conversion attribution, it requires a fair amount of data. If your website does not qualify than you will still follow traditional attribution based models. An AdWords account must have over 20,000 clicks and a conversion action must have at least 800 conversions within 30 days.

Depending on the conversion action data, you may find the data-driven model only available to certain parts of your website, most likely your most active and profitable parts. Once the necessary data requirements are reached, Google will begin preparing a model for 30 consecutive days. Unfortunately, if your conversion action drops to 400 conversions or under 10,000 clicks within the 30 day period, your website will be unable to access the data-driven model.

Empowering Search Optimization

  1. Segmenting Audiences

The future of search will be dictated by machine learning and advertisers will only profit more because of this. Using the data-driven model, marketers will be able to better understand the user intent of different people and be able to segment audiences and ad campaigns based on varying demographics. This will also give better description of the life cycle of new versus returning users.

  1. Adjust Bids in Real-Time

By understanding how keywords operate within a conversions funnel, marketers will be able to adjust bids for keywords accordingly based on their importance to the overall conversion action. Users with an AdWords account will be able to sync bids for real-time adjustment based on the importance of keywords to a conversions funnel thus maximizing their ROI for the next quarter.

  1. Optimizing Touchpoint Outputs

By understanding how different ad copy and keywords operate within a conversions funnel, marketers will be able to optimize different touchpoint outputs for maximum effectiveness. By understanding at what point users convert and where users take longer to convert, you can optimize touchpoints to increase the relative speed of your conversions funnel.

  1. RLSAs

By segmenting your audiences and understanding their user intent, you can remarket advertisements to returning users to capture more sales. This can be effective at gaining conversions from customers who may have abandoned your cart or typically make large basket purchases.

  1. Data Mining

By segmenting audiences and analyzing mid-level funnel actions, marketers can better analyze certain aspects of their ad campaigns, which trigger user engagement. Marketers are utilizing data mining to better understand user behavior to understand trigger behaviors. These insights will allow marketers to change their messaging and even the time of day they are able to reach and trigger user engagement.

  1. Cross-device Tracking

Cross-device tracking capabilities have made it easier to assign attribution in a multi-touch attribution model. This data gives better insights into the behaviors and search paths of users across different devices. Users on mobile devices are more likely to use less clicks before making a conversion as opposed to a desktop user. This data is valuable in optimizing ad copy across different devices and adjusting the different touchpoint inputs that may not operate as effectively across different devices.

The data-driven model will fully empower marketers to completely optimize ad campaigns like never before. It’s hard to comprehend for most search marketers that the last-click or even first-click attribution model was fairly ineffective after relying on it for so many years.

With the rise of machine learning as a core part of Google’s algorithm, the golden age of paid advertising may be fully realized as marketers are becoming better able to understand user behaviors and optimize their ad campaigns accordingly.