How To Use Your Campaign Data To Reverse Engineer Yahoo Gemini Ad Delivery Algorithms (Part 1, eCPM vs Clicks)

February 2, 2015

I haven’t had much time to blog lately so I had one of my A4D media buying team members write up something cool that we’re doing to figure out how Yahoo Gemini’s algorithm works.

Consistent ad delivery is the bane of every advertiser that uses Yahoo Gemini. Here’s how you can use data from your Yahoo campaigns to better understand their algorithm.


To an owner of online advertising real estate like Yahoo, all that matters is an ad’s eCPM (effective cost-per-thousand impressions). To a publisher, it’s pretty much “earnings-per-thousand impressions”, because it’s the dollar amount of which you end up paying them to advertise. On Yahoo, if you’re bidding on CPM, that’s also your eCPM. If you’re bidding on a CPC, then eCPM is calculated as “CTR x CPC x 1000”. The higher your eCPM, the more Yahoo should theoretically be delivering your ad. Theoretically, that is.

Pulling The Data

To take a closer look at what Yahoo’s actually doing with ad delivery, we’re going to need to look at a graph of each of your campaigns’ eCPM and total clicks over their lifetimes. Export your campaign data in Yahoo by going to the reporting tab, setting the date range to as far back as you can up to today. The export to .CSV button is to the right of that, it’s a little square with an arrow coming out of it.

Yahoo Gemini Report Select

In case you’ve never tried to look at Yahoo’s reporting, this is the bar you want.

Using The Data

In your exported CSV file, you’ll see a column for clicks and for CPM (same as eCPM). Create a scatterplot of CPM vs Clicks with this data, and generate a line of best fit. I’d add instructions, but Google is much better at explaining excel than I am. Make sure to show the line’s equation. Here’s what we got:

Yahoo Gemini Campaign Scatter Plot

Shoutout to Janell and Lorie for actually putting together this data.

Data Analysis

First, let’s suspend all judgment on how many campaigns we have cluttering around the origin. Online advertising is hard sometimes, and that’s okay.

Second, let’s take a look at the line’s equation.

Y = 3490.9X – 897.5

So what does this mean?

1). For every dollar you increase your eCPM, you can expect to get 3490.9 more clicks on average.

2). Your eCPM must be at least $.257 (897.5/3490.9) to expect to get ANY clicks at all. With an estimated CPC of $.30, this means you need at least a CTR of .08%.

3). According to this linear model, if you manage to get an eCPM below $.257, Yahoo will actually make you give them clicks. (hah, math jokes.)

Third, let’s just take a look at the graph itself.

The variance is enormous. This goes hand-in-hand with Yahoo being an unpredictable ad platform. Even with pruning this data for outliers (taking out ads with an eCPM of $3 and only 50 clicks, or vice-versa) we still have plotted points flying way above or below the line.

Yahoo’s platform probably delivers exponentially. This is just from taking a look at the points and how they’re scattered (the amount of clicks scales upward extremely quickly as eCPM increases). If you’ve ever experienced a campaign explode overnight, that would be an example. Here’s the line of best fit exponentially:

Yahoo Gemini Scatterplot

So this means if I bid $8 on a CPM I get infinite traffic, right?


In conclusion, an ad’s delivery still comes down to its CTR. This is truth for pretty much any advertising platform ever. Sure you could artificially inflate an ad’s CPC/CPM bid, but as people who’ve already tried that, let us just say that you’ll either a). be unprofitable or b). see that Yahoo’s still aware that your ad sucks and doesn’t want to embarrass its users with it. Moral of the story? If you’re having a hard time getting Yahoo to deliver, take the time to develop yourself into a better copywriter. It’ll raise your overall average eCPM, and that’ll definitely get you more delivery, guaranteed.

P.S. If you take the time to do this analysis (and you should), share your results in the comments. Particularly interested in gathering more data on minimum eCPM required for delivery in the linear model ($.257 in this example).

This content brought to you by the A4D Team. Are you looking for users or want to generate users for others????

CLICK HERE to start working with A4D today!



Want More? The more people listening the more I’ll write.
Subscribe to get business insights in your inbox
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.