Cost-per-conversion math wrong in AdWords?

[Craig87]

I recently noticed that the cost per conversion numbers shown in AdWords (1-per-click) were wrong. They were under-reported by as much at 15%. Simple long division (Cost divided by one-per-click conversions) was not what was showing up in adwords.

I called and IMd support and they have no idea what would cause this. Of course, being Google, they would never EVER admit a mistake so they just say they will get back to me.


Basically, this leaves two options 1) Google screwed up math or 2) They have up and decided to count something else as a conversion (I thought it might be phone calls, but that doesn't ad up either.)

Has anyone seen anything like this?
[imgur.com...]

[Craig87]

I should note that I have view-through conversion tracking de-duping enabled. Even then, including view through conversions, I am still off.

[RhinoFish]

does look wrong to me. any chance you haven't let enough time go by and you happened to catch it near updating? when you check that same time frame now, does it still show the same?

[Craig87]

Nah, this is an issue that goes back 2 months. It doesn't change day by day.

I just got this highly cryptic response from Google support:

"Our specialist team investigated the issue and found that there are 541 invalid clicks totally in the month of April 2012. The clicks that resulted in conversions were filtered out as invalid clicks. Since conversions do not get filtered out, this results in zero cost for the conversions. Hence, there is a discrepancy in cost per conversion."

Well, that does not really make sense either, or at least I cannot follow it. It would certainly explain the discrepancy if I got a credit for 541 clicks, but I did not.

[LucidSW]

What they are saying is this:

You got 10 clicks and one resulted in a sale. Therefore your conversion rate shows as 10%. CPS would show $10 on total cost of $10.

However, they later credited you for an invalid click (new actual clicks = 9 and cost of $9). But for some reason, those numbers don't get updated in the Adwords stats. You are left with the impression that you got 10 clicks and one sale that cost $10.

I don't pay attention to the calculated values shown in Adwords so I've never noticed this. But I have on at least two occasions noticed a sale without any clicks or costs. Obviously a sale on what was considered an invalid click which the system caught and the click and cost not counted. Why it was considered invalid I have no idea.

[Craig87]

Which would make sense if I got credited for the invalid click. However, in your example, my cost has remained at $10 and I received no credits (of consequence, anyway).

In order to balance March, for example, I would need a credit (or variance) of over $1,500 dollars.

[Craig87]

So I thought I would update this thread in case other people have problems with adwords giving your bad data regarding your cost-per-conversion numbers.

Basically, AdWords does not count the cost generated by clicks where they cannot track the cookies.

Support said:
"I re-consulted the specialist team with regards to your query and they have confirmed that they drop the clicks which cannot convert as including them in the total conversions cost would not be justified. This includes clicks that happened before Conversion Tracking was enabled in the AW account and clicks on non-cookie enabled browsers (such as WAP phones and some browsers with cookies disabled)."



First, this problem is entirely undocumented. I asked why it is not in the knowledge base they refused to say.

Second this is a horrible answer for any advertisers that care about overall cost per conversion. Basically, Google will charge you for clicks it cannot track (fine, it happens) but they they choose not to include that cost in your results. This causes the AdWords interface to show you a LOWER cost per conversion than you are actually spending.

This is a profound lie from Google.
As more people use incognito and other cookie blocking technologies the cost-per-conversion will become increasingly deceptive.


Another helpful feature from your friends who are not evil.

[RhinoFish]

undocumented, i agree, not good.

they should make it a configurable option.

but your argument is off in my view (so is theirs). the truth would likely be somewhere in between (some untracked sales will happen), but depending on many factors (like what if WAP was a high converting medium for you) the actual numbers could outside of the range defined by the reported-including-all-clicks and reported-excluding-some-clicks.

to call it a lie, over done in my view - though the lack of documentation (if accurate) makes it an omission, which is closer to lie than truth.

if it can't be configured, they should display both - raw and corrected.