Role of machine learning in the digital advertising sphere

Role of machine learning in the digital advertising sphere

Key figures: How much data is produced every day?

CONTEXT:

Artificial intelligence is playing globally a significant role in people’s daily life. As a digital user, you encounter AL in your day-to-day routine whether on Google Maps trying to find your way, using chat Bots on Facebook Messenger looking to find the perfect shoe model or using “Siri” or “OK Google” to seek some support. In this exceptionally fast-moving condition where people tend to request instant information on-the-go, AL found its practicality.

Digital advertising didn’t miss the boat being one of the first industries that embraced AL pretty fast. Due to the increasing computing power, the abundance of connected data as well as human & financial limitations, using machine learning in advertising today never seemed more convenient allowing advertisers to handle humongous amount of data while considering endless data signals (see below):

OBJECTIVE: Understand how Machine learning algorithms, part of AL technologies, are used to maximize profits and reach a cost-effective advertising model. APPROACH &

TAKEAWAYS: Automated bidding strategies are probably one of the most powerful features when it comes to optimizing your digital campaigns whether on Google, Facebook, Twitter or any other advanced Ad management platforms against a given target (CPL, Conversions, CPA, etc.). At Google, probably among the most developed ones, they are represented with a set of conversion-based bid strategies named : Target CPA, Target ROAS, Maximize Conversions and Enhanced CPC. These strategies use historical data as well as all the signals mentioned above during the auction-time, to define an optimal CPC for your eligible impressions.

Smart bidding, when used properly, allows you to:

  • Keep up with a continuously evolving consumer behavior (thanks to all the signals used to optimize)
  • Not to dilute your efforts but rather stay focused and accurate
  • Better understand your most qualified audiences
  • Understand the full picture across devices & channels, locations & languages, Browsers, etc.
  • Reach cost-effectiveness in a real-time logic

Smart bidding needs historical data to work properly, Google recommends having at least 15 conversions in the past 30 days for this feature to start working properly.

We suggest you follow the steps below:

  • Step 1: Start with enhanced CPC on your Google campaigns while monitoring your average position and conversions number
  • Step 2: Once the required number of conversions is reached (at least 15 conversions in the past 30 days), switch to a target CPA or ROAS (to define using your historical data)
  • Step 3: Closely monitor your CPA during this period since an increase will probably be seen, this increase should start depressing over time (after 2 weeks at least) to get the closest to your target.
  • Step 4: Once enough data is recorded, you can make a statistical significance test to decide if you would want to keep the automated bidding running, or not.

WAYS FORWARD: DDA (Data-driven Attribution) is a Google attribution model that also uses machine learning to give credit to your site’s conversions based on Ads, keywords and other account’ elements that are most likely to drive positive results. Using historical converting and non-converting paths as well as predictive models, DDA can produce a more accurate picture of each element’s participation in the conversion.Keep in mind:

  • DDA requires at least 600 conversions and at least 15,000 clicks (in the past 30 days) to perform best.
  • DDA feeds into Smart-bidding (best-practices)
  • Better switch DDA on then Smart bidding for more accurate results
  • DDA as of today (02.12.17) works only on Google Search and Shopping (GDN and YouTube stay on a last-click Model)

Source: *http://www.northeastern.edu/levelblog/2016/05/13/how-much-data-produced-every-day/Learn more about Smart-bidding: https://support.google.com/adwords/answer/7065882?hl=en