Similar Audiences has been available for a while in Google, but it’s had some struggles to gain the traction produced by the original tool if its kind, Facebook’s Lookalike Audiences.
That said, the combination of intent-driven keywords and a big new audience pool is too good not to test. We recently ran a search campaign targeting very broad non-brand terms to try to entice more customers into the top of the conversion funnel. In this post, we’ll break down Similar Audiences, how we tested, and what we learned. Let’s jump in.
What It Is
Google says it best: “Similar Audiences looks at data about your existing remarketing audiences and finds new and qualified consumers who have shared interests with that audience. It’s a powerful – but simple – way to reach a much larger audience and drive clicks and conversions among new prospects.”
The Test (April 3, 2017 – May 8, 2017)
All search marketers know that upper-funnel keywords can be a tough nut to crack, especially when being held to the same performance goals as the mid and lower funnel
- Upper-funnel terms are broad in nature and competitive, and search intent is not well known.
The advertiser we are testing this for is in the women’s style vertical. For this client’s campaign:
- We created a search campaign targeting extremely broad non-brand terms: fashion, clothes, women’s clothing, fashion trends, shopping, style, dresses, etc.
- On top of this, we layered on audience lists of users who were similar to our most valuable customer lists (target & bid). These lists included people who have signed up, people who have completed a profile, and people who have made an order.
- Additionally, we added demographic restrictions ensuring we were only serving ads to users on said lists who were also the gender and age range of our target user.
Summary and Takeaways
- Better results for “light” conversions (Sign Ups, Profile Completions)
- Latency could be a factor in converting similar audiences all the way to an Order conversion as Similar Audience users are just becoming familiar with the brand/offering vs. remarketing audiences who have previously been exposed to the brand
- Higher CTR than traditional RLSA
- Remarketing users have been to our site before and know what we are offering; they may be more apt to ignore our ad
- Higher CPCs than traditional RLSA
- Prospecting new vs. returning users comes at a higher cost
- When all conversion actions are merged together for high-level analysis, we see that Similar Audience targeting had higher CVR and a slightly lower CPA. There is hope for us yet!
- Overall lower volume than remarking audiences (impressions, clicks, and conversions)
- Similar audiences make up a smaller piece of the overall campaign
- RLSA lists are very large for this advertiser
While we aren’t 100% satisfied with the initial performance being driven by our upper-funnel similar audience campaign for search, we do see some promising signs that we can optimize further to bring this program to the CPA range we are capturing in our traditional RLSA targeting. We are also excited about the endless expansion opportunities similar audiences brings to the table: broader keywords, broader/larger lists (similar to site visitors, similar to blog readers, etc.) and the ability to layer with other targeting methods (Dynamic Search Ads layered with Similar Audience lists).
If you are ready to test this out for yourself, consider some of the tips below:
- Examine search queries and add negative keywords often
- Promote converting queries to your keyword portfolio
- Adjust bids frequently and consider testing on Conversion Optimizer
- Analyze device performance and adjust accordingly
- Analyze demographic data (age ranges and gender) and adjust bids/make exclusions frequently
- Start with broad match modified or phrase match types and loosen restrictions for high-performing terms
- g. BMM working well? Try pure broad match…..Phrase match working well? Try BMM.
- Give it time. It took about 3 weeks to gather enough data to make needle-moving optimizations confidently
- Try several high-value lists and weed out low performers as data comes in (see below screenshot – some lists just work better than others)
If you have any questions, please leave a comment!