Help Machine Learning – Don’t Edit Campaigns

Help Machine Learning- Don’t Edit Campaigns

You have a campaign in Google or Facebook ads that’s doing really well over several months. Maybe this was for a product that is no longer sold.

You have a new product to sell. And it seems obvious to try and make some edits in the text,images,video of the ad and change the landing pages for the new product.


Here is why:

Machine learning takes time: the machine learning process had worked over months on the campaign and had been gathering data. For Google Display ads you would get recommendations about which images are not performing well so that you can try and change them. Similarly, you’ll get recommendations for headlines and text if they are not doing well. It’s fine to edit those. Because the machine is telling you. (Sorry if you feel like a slave to the machine at this point). This is called the learning phase and although Facebook and Google mention 7 to 10 days, it’s really important not to fiddle with an ad when you set it up. Because every time you make edits – it re-enters the learning phase. While the learning phase may be a couple of weeks- campaign that has been running for a month or more has more data for the campaign settings. This includes campaign objective, geography, ad schedule,device preference etc.

Unlearning is hard: Unlearning is hard for humans. It is hard for machines as well. What happens when you change an ongoing campaign is that the machine has to unlearn and that takes a lot of time.

So what should you do?: Copy the successful campaign that you want to use for a different product. Now make your edits. If the old successful campaign is not required (assuming its a discontinued product). Pause it. The copied campaign has nothing to unlearn but you have all the content including text,images,videos running. If it is a new product its important to keep the campaign objective as traffic and not conversions. Why? because conversions take time as consumers go through their buying journey. Staying with traffic is a good idea as that gives you a chance to convert and you build data on the target market behavior.

Acknowledgement: we acknowledge the discussion with a helpful Google Ad specialist that clarified the suggestions in this post. While all the learning materials around the Google and Facebook exams do cover this point- the digital marketer is most frustrated when a campaign while approved, does not seem to be serving. We hope this post helps digital marketers.

About StratoServe

Leave a Reply

%d bloggers like this: