corner image corner image
corner image corner image
corner image corner image
corner image corner image
corner image corner image
corner image corner image
corner image corner image

Some thoughts about image and text ads

Lately, I’ve been thinking about whether mixing text and image ads is really profitable. The case that most advertising companies (like Google adsense, adbrite and lots of others) make is that allowing image ads in the same ad unit as text ads is usually better because they compete in price with text ads. But, my understanding is that image ads will only show if they beat the highest text ad price. I tried to analyze and see whether the final profit will be larger in case of mixed ads or not and I found conditions for image ads to make more money if allowed to compete against text ads in the same ad unit.

Symbols

P_i Probability to click an image ad

P_t Probability to click a text ad

N Number of text units inside the ad

C_{ti} Cost per click for image ad unit number i

C_t=\frac{1}{N}\sum_{i=1\ldots N}C_{ti} Average cost per click for text ads

C_i Cost per click for image ads

M_i The money made through image ads

M_t The money made through text ads

I_i The number of impressions of image ads

I_t The number of impressions of text ads

D_i The dollars profit of image ads

D_t The dollars profit of text ads

Z_i The number click throughs of image ads

Z_t The number click throughs of text ads

So,

M_t=NP_tC_t

M_i=P_iC_i

If an image ad wins the auction, this implies C_i > MAX[C_{ti}] > C_t. However, for the image ad to be more profitable the average money that it will generate must be more than the average money that the text ad will generate, i.e. M_i>M_t which is P_iC_i>NP_tC_t.

However, I do believe (although I don’t have enough statistics yeet) that P_i\approx P_t and might be slightly higher, but not much bigger. That leaves us with C_i>NC_t, which means that the condition for the image ads to actually generate larger profit is

C_i>\sum_iC_{ti}

i.e. the payout of the image ad has to be more than the sum of all the payouts of the text ad units. This is based on the assumption that the probabilities of text and image clicks are approximately the same. Since the ratio P_i/P_t is not actually known, the worst case scenario is that P_i=P_t since in average image ads are more appealing to people and usually will get seen more. In that case, the condition earlier can be used as a sufficient condition for an image ad winning a campaign over a set of text ads and be of higher profit.

But from our point of view, and given that we know that advertisers do not apply this criteria how can we determine whether or not to let image ads run in the auction? To answer that question we note that C_i>C_t as the image ad won the auction. This is because C_i>MAX[C_{ti}]>C_t. For the inequality P_iC_i>NP_tC_t to always be true, we need to make sure that P_i>NP_t as well. If the latter is true, then the inequality holds at all times, and an auction winning image ad will be more profitable than the N text units.

This boils down to knowing the ratio P_i /P_t which is required to be P_i/P_t>N. We can actually know this ratio by running an only text ad, and an only image ad that alternate in the same ad spot and of course having two different channels or ad zones. Using the stats of your ad server (adsense/adbrite or any other), then you can determine M_t=D_t/I_t and can approximate C_t by D_t/Z_t, and N is already known from the ad block that you chose, which is the number of text units in that block. Therefore P_t=M_t/(NC_t)=Z_t/(NI_t). Similarly you can find P_i. If P_i/P_t>N then you know for sure now that it is more profitable for you to allow image ads to bid against text ads in the same auction for that ad unit spot.

Disclaimer: The information above is stated as is with no guarantees of any type. Use it on your own risk.

Technorati Tags: , , , , , , , ,

Tags: , , , , , , , ,

corner image corner image

corner image corner image

Leave a Reply

corner image corner image
9,496 spam comments
blocked by
Akismet