Need help with algorithm .... KEI for PPC

Status
Not open for further replies.

erect

New member
Jun 3, 2007
3,795
154
0
Esoterica
twitter.com
I'm openly soliciting help from all the affiliate marketers here on how to get the best keywords for a PPC campaign.

KEI is an analysis of a keywords potential that is used by wordtracker, but it's based on SEO ... not PPC. I'm trying work something out so I can attach a number to a keyword's potential & rank a list of words best to worst. Here's what I'm working with

$keyword = Keyword
$cpc = estimated CPC (adwords tool)
$volume = estimated volume (adwords tool)
$total_competition = total competition for that keyword found here
$ea = erect's algorithm ... the number that represents a keyword's potential for a PPC campaign (the bigger the better)

PHP:
if ($total_competition > 5) { $total_competition = ( $total_competition * 2 ) ; }
$ea = ( ($volume / $cpc) * (1 / $total_competition) ) ;

It looks simplistic and should do a decent job, but I'm interested in something better ... which is why we're having this conversation.

Basically the above 1st takes the $total_competition and if the number is > 5, double it. I've got this in place because I know 10 ads or so show up on each search results page. Ignoring broad matching, QS & bid price ... having less than 5 people competing for ad space is valuable because even if you bid $0.05 it's possible you'll show up when using an exact match.

From there the function computes the volume / ctr fraction, which translates roughly into how much volume you get for your click price.

Then I take the result and multiply it by 1 / $number_of_competition ... the bigger the competition the smaller the fraction therefore reducing the final $ea total

To make this more challenging for me, I have no data to test with and I don't run PPC campaigns ... I'm totally an organic guy. I'd appreciate WF chiming with your experience in this so that I can make this formula golden (like a brick, not a shower).

Ooh yea, don't say "volume is the only factor I look at". I know you're a badass who pwns the QS and has an unlimited budget. Good for you, pretend you need to convert on those cheap long-tails for a sec and help a brother out.
 


I think the multiplier is going to change based on the niche. Try your algorithm out with a few sample competitive markets like ringtones, diet, or credit reports; then again for some less competitive stuff from CJ, Amazon, or Shareasale, and then post some results if you can without giving away the store. Guys that have run many campaigns and network owners will be be able to recognize relative degree of difficulty and let you know if you are on track.
 
Bump for the late night crowd

I think the multiplier is going to change based on the niche.

That concerns me as I don't want to directly tie keywords to a niche. It very much limits my program and it becomes a maintenance headache .... it's meant to be pretty hands free

Try your algorithm out with a few sample competitive markets like ringtones, diet, or credit reports; then again for some less competitive stuff from CJ, Amazon, or Shareasale,

Well that's kind of the problem, I don't run PPC stuff so there's nothing to test ... and I really don't have the time to learn it in a competitive marketplace. This is why I'm asking you guys.

and then post some results if you can without giving away the store. Guys that have run many campaigns and network owners will be be able to recognize relative degree of difficulty and let you know if you are on track.

I want to make this very clear, I have no problem giving stuff away as long as the end result is a better product and I won't lose a penny.

So what you're asking is to run a couple of tests (not in PPC, rather with the algorithm above) and see what the experts here have to say ... ok will do tomorrow as I wrote the part that scrapes sponsored listings this afternoon.

Another problem is everybody's drunk with a hooker in Vegas right about now ... or is that the solution?
 
Where's my fucking algebra teacher when I need him ....

ok, after some crunching with a sample report, here's what I've come up with

( ( $volume ^ 2 ) / $CPC ) * ( 1 / $competition )

Only a slight change ... I do the ^2 (that's volume squared) so that total # of visitors plays a much larger role in the algo. Before it had good words ranking high using EA, but many of the top ranking words only had 1 search per month ... well that's not really a good one as you and I really don't need an extra quarter that bad (unless it's herb).

The forumla seem to be looking great with the reports I'm making today.

THIS THREAD IS NOT DEAD!!!!


I'm still looking for advice, please Zany or anyone else ... drop 2c on the table and lets discuss.
 
Allright, this took about 2 hours to work out so I hope it helps:

Let's say that you don't have a product and are relying on the information out there to assess the quality of a keyword in the dark. I have just realized that this is probably not the case that you're asking about (since you probably have a product already, and are simply choosing a few good KWs for it). If that's the case then seriously skip to the next large heading. This upcoming writeup is for choosing a good keyword *blind*, for finding a niche.

So, to start, let's call our score a Zs (:p) which = profit from a keyword.
Zs = Profit = clicks * profit per click
= clicks * (Revenue per click - CPC)
= clicks * ( (1 + desired ROI) * CPC - CPC)
= clicks * (CPC * desired ROI)

Explanation: so, we don't have an offer yet and don't know the payout. We have to rely on what others are paying and presume that they earn a reasonable % on top of the CPC. Profit is simply the CPC times the desired ROI to keep us advertising (ex. 0.3)

Now, what is "clicks"? Clicks is the volume x Ad CTR. But assuming that we don't even know if the keyword is commercially viable, we can reduce the pure search volume to "searchers w/ commercial intent". We assess that by the number of existing advertisers. No advertisers means that most of these searches are probably don't have commercial intent. A lot of advertisers means that this keyword is very lucrative and a lot of the searchers have commercial intent.

So, we'll assume that if there are >=20 advertisers then all searches are relevant and commercial. If there are 0 advertisers, then we'll assume that only 1/20 of them are relevant and monetizable. We make the assumption that smart advertisers have already flocked to good keywords and left out the not-so-good ones.

clicks = volume x Reduction Factor x Ad CTR
= volume x (Min( Max( 1, Advertisers ), 20 ) / 20 ) x Ad CTR

Assume Ad CTR = 2%, since Google will shut you down otherwise.
An easier way to do this is to skip the whole reduction, and assume that you'll get a 2% CTR as long as you're on the first page. To make that assumption you need to run this algorithm only if there are only 5-7 other advertisers, otherwise assume that you won't get on the front page, and simply assign a Zs of $0.

Ok, so now we have the clicks portion, now we need an intelligent way to calculate the expected ROI on any given keyword. Let's assume that the more advertisers on a KW, the lower the ROI. Let's further assume that the highest ROI is 60% when you're alone, and it goes down like this:

  • 0 advertisers - ROI 0.6
  • 1 advertiser - ROI 0.5
  • 2 advertisers - ROI 0.4
  • 3 advertisers - ROI 0.3 etc.
so ROI can be modeled by:

ROI = Max( 0, 0.6 - Advertisers * 0.1 )

Let's put that together with
Clicks = volume x (Min( Max( 1, Advertisers ), 20 ) / 20 ) x Ad CTR
CPC = <from Google>

and we get
Zs = volume x (Min( Max( 1, Advertisers ), 20 ) / 20 ) x 0.02% x CPC x Max( 0, 0.6 - Advertisers * 0.1 )


What parameters can we tweak?
- the assumption that CTR is 0.02 (made because Google tries to keep it at that)
- the assumption that 0 advertisers means there are 1/20 relevant searches. We can even dump that section altogether and just multiply by CTR given you're guaranteed to get on the front page.
- ROI starts at 60% and goes down by 10% for each advertiser. ROI floor = 0%

Calculating Zs when you picked a niche

Ok, so let's say that you picked a niche and picked a product.You now know the payout and the approx conversion rate once a visitor lands on the merchant's page.

This changes our formula in that we know that all the volume is relevant (since you picked the niche and made sure its not a dud keyword we're looking at, ex. "me you").
We know the payout, so we can use our offer instead of working the payout backwards from someone else's CPC.

The formula is still the same:

Zs = clicks * profit per click

clicks = volume x Ad CTR

profit per click
= ( Payout * Merchant CR ) - CPC
this is for direct linking, if using an LP use the below:

profit per click = ( Payout * LP CTR * Merchant CR ) - CPC

....

Zs = volume x 0.02 x ( payout * merchant CR - CPC )


Now, if there are 50 other advertisers, this formula won't hold. That's why you can simply set it to zero when there are more than about 6 other advertisers - don't even compete for that KW.



Sheesh, that's a lot of writing and I hope it makes sense. Its fun to try this out in Excel.
 
Status
Not open for further replies.