|Audiences: Similar to.|
| 3:34 pm on Nov 17, 2013 (gmt 0)|
It looks like Google has automatically created a "Similar to" audience for each of my audiences. How does that work?
| 9:44 pm on Nov 17, 2013 (gmt 0)|
Your site tags visitors, who you can remarket to...
When G finds someone who hasn't been to your site, but in many other regards, looks similar to your remarketing pools of people, they think those folks might be likely to also shop at your store.
So it's not REmarketing, they haven't been to your site. It's an audience of people similar to those you are remarketing to.
If 7up is the unCola, Sim Aud is the unRemarketing.
If you sell high end goods for example, can you see how this would make a ton of sense?
Of let's say you sell something that people who speak foreign languages need more often than others, like passport services, can you see how you might do better with this Sim Aud slice?
Can you see how the concept of card abandoners breaks down when translated to Sim Aud? Look at your broadest visitor lists, and certainly those who bought, but depending on what you sell, and how you drive traffic, your buyer rem lists may demographically look like your visitor rem lists... so experiment.
You might also use Sim Auds as bid modifiers, vs straight targeting mechanisms. Set up the Sim Aud for an existing Campaign, so you can see the data for this pool of people, but leave the bid unspecified. Let the data pile up, and if a Sim Aud slice outperforms the larger targeted pool, assign a larger bid via the Sim Aud.
Careful though, Sim Aud are people who don't know you, they haven't been to your site before - whereas your repeat buyers have. Be a little slow to draw conclusions when comparing Sim Aud slices to Gen Pop slices, your Gen Pops are often biased data (mix of new and repeat visitors).
Think of all the things they know about people, all of the sites they visit, their info if they sign into G services, their email contents...
| 5:38 pm on Nov 24, 2013 (gmt 0)|
Tonearm, are you saying Google both created Similar Audiences for you *and* implemented it (turned it on)?
| 8:50 pm on Nov 24, 2013 (gmt 0)|
In the Shared Audience library, they'll create Audience lists automatically, but that's all.
You have to build a campaign to use those lists.
| 11:21 pm on Nov 24, 2013 (gmt 0)|
Many thanks RhinoFish, I'm experimenting now.
shorebreak, they created the audiences automatically but did not activate them.
Anyone else tried targeting these "similar to" audiences?
| 3:29 pm on Nov 27, 2013 (gmt 0)|
Is it typical to see massively lower volume for "Similar to" audience targeting? Yesterday I got 50 clicks for regular audiences and 1 click for "Similar to" audiences.
| 5:11 pm on Nov 27, 2013 (gmt 0)|
"Regular" know your site already, while "Similar to" don't.
But your Sim Aud lists are very likely many times larger than their "Regular" counterparts.
So Sim Aud is usually a bigger Audience, but with a lower CTR.
Many factors affect your specific performance though.
| 5:21 pm on Nov 27, 2013 (gmt 0)|
I'm trying to figure out where exactly Google gets the cookie information against which it builds Similar Audiences. Most of the advertisers using it are transactional, direct response advertisers who clearly are looking for people who are *similar to their website visitors*, so if Google is looking for people who are similar to ecommerce intent-laden visitors, is it not logical to assume that Google is pooling cookied users from all those AdWords Remarketing advertisers remarketing lists, then slicing & dicing it for use by other advertisers?
| 9:27 pm on Dec 2, 2013 (gmt 0)|
I assume it's a weighted, multi-dimensional, correlation matrix that would be impossible to reverse engineer (too many variables).
Only one of which would be a deep data bucket itself, the one you described.
| 11:47 pm on Dec 2, 2013 (gmt 0)|
@RhinoFish - in the retargeting industry, the only data retargeting firms have been able to make work is 1st party data. Case in point: I've yet to see a single case of search query data being used to retarget against *and* hith the ROI metrics most direct response advertisers have to hit to make an ad buy ROI-positive.
I agree that G's probably taking into account many other dimensions of data (which as you say would be impossible to reverse engineer), but that notwithstanding, I'm inclined to think that 1st party data is 75%+ of the intent 'signal' G itself uses, which would then lead back to the question:
Is it not *likely* that Google's primary intent signal for Similar Audiences is the 1st party intent signals it gets from all the advertisers who do remarketing through AdWords?
| 6:04 pm on Dec 3, 2013 (gmt 0)|
Any tips on increasing similar audience performance? I'm seeing nothing from it. Is it something I should try bidding highly for or would conversions likely not justify that?
| 6:16 pm on Dec 3, 2013 (gmt 0)|
Shorebreak, i disagree with everything you said above in your 1st paragraph. Which is weird, we usually agree on things. :-)
Yep, certainly, 1st party data is core to their signal scheme. But you seem to be thinking of imminent buying intent and some sort of scramble to intercede by advertisers (that far down the funnel, display opps diminish), I'm basing that on your "ecommerce intent-laden" comment above. I think it's much broader than that, and I think that for several reasons - the way G describes the product, the reach is so large it can't be late funnel, and the conversion rate we see with it. I think it's more like demographic and interest based targeting combined sprinkled with steroids, than imminent buy targeting. But these are just guesses on my part. The way we run things, DR / ROAS, we often don't care too deeply about why, but focus on the how. The complexities of the inner workings, which we expect nobody will ever reveal to us, are fascinating, but we always end up experimenting and adjusting, not dissecting and rev engineering. We do care a lot about how it works (relative to most), but we look for just enough insight to give us clues to tactics we should test. Perhaps we're more empirical statisticians here, than behavioral scientists or voodoo gurus. Here, we're boring effective pounders, not brilliant whiz kids.
Let's say I could tell you for certain (I can't) that the answer to your question is 83.4%... then what?