1. Question: How many ads do I need to run to determine if a program is legitimate?

The conversion rate in affiliate marketing is extremely low. Lets say I decide that a 1 in 50 conversion rate is acceptable.

So I am testing for a 1 in 50 conversion. If I get one click, there is a 1 in 50 chance that I get a sale (proving the program is legit). Conversely, there is a 49 in 50 chance I don't.

There is a 49 in 50 chance the second ad fails, a 49/50 chance the third test fails, ...

If I get 10 hits, there is a 49^10/50^10 chance of getting a sale. This equals 60.35%.

If I get 100 hits, there is a 49^100/50^100 = 13.26% chance of failure.

This means that if I chose 1 in 100 hits as my program test, one in ten of the companies I throw out will be legit.

The following table shows the number of hits and the percent of accuracy:

Pos - Percent
1 --- 98.00%
25 -- 60.35%
50 -- 36.42%
75 -- 21.98%
100 - 13.26%
125 - 8.00%
150 - 4.83%
175 - 2.91%
200 - 1.76%
225 - 1.06%

This chart shows the curve:

To get down to 1% accuracy, I would need 229 hits. Let's say I have a 1% click through rate. I would have to display about 22,900 ads to get a decent test of a program.

To test a program, I have to give 23k free views to a bogus company.

Let's say I am getting a \$15.00 EPC ads. To do a decent test cost \$30.00 in ad space. This does not take any consideration for time or anquish.

I figure the real cost of testing a bogus merchant to be along the line of \$100 in time and lost opportunity.

Affiliate marketing fails without performance metrics from a trusted third party.

Protophoto - Short Stories

2. Yintercept,

I have to say that this may be the best post I've seen here yet.

I've been thinking about this myself recently when I was explaining to someone who claimed that there must be tracking issues because they didn't convert in 700 clicks.

By my reckoning (and stats book), in the case I was looking at, you'd need more than 10,000 clicks to get a statistically significant sample and make a reasonably accurate statement about whether a merchant has a tracking problem. Otherwise, all you need is to be unlucky.

Your analysis (clearly you're a mathematician) demonstrates the same much more thoroughly.

I recognize that there's significant risk to trying out any new merchant for this reason and others and I'm working to reduce that risk (Displaying return days was one step and clearly there's more.)

The problem is that merchants' conversion information is, by contract, business confidential information. So we just can't give it out.

Perhaps we just need to push hard to get merchants to disclose this data. Perhaps there's some way to mask it so you get what you need and merchants aren't sharing confidential info with their competitors, as that's the big sticking point.

As we search for the right balance, let me ask you the question. With the above challenge in mind, what would be useful to you?

--scott

--
Scott Jangro
sjangro@befree.com

3. I would think the best merchants would want a metric posted; Basic performance metrics is probably something that you could get on a voluntary basis.

If it costs an affiliate \$100.00 in time and lost opportunities for each failed test, you can see why they tend to be angry and cynical.

You are also right, a lot of what you see is people who don't understand how hard it is to make a buck in this industry. 700 impressions is not a legitimate test. The 700 impressions probably includes 50 impressions from the affiliate while making and testing the page. The hit count probably includes 5 test clicks from the affiliate.

I wish the industry did a better job explaining just how many ads they will have to display and the expect rate that they will get paid in this market. It would save a lot of grief.

Protophoto - Short Stories

4. Thanks yintercept and Scott for addressing this issue. Scott was referring to my inquiry as to why, from the exact sames pages with similar product displays -equal pricing and same shoppers, would ComputerGeeks not show one sale on 800 CLICKS (not impressions) when TigerDirect reported 1 sale per every 60 clicks.

Scott was kind enough to run a test purchase, since I sure wished to trust my hard earned savvy shoppers to an alternate source for computer closeouts. The test purchase went thru even though the merchant was alerted to the test. I trust Scott and only challenged that the report showed zero orders or amt. but did show items shipped and shipped \$ amt.

I stated to Scott that any merchant from his own provided links and landing pages who can't convert 1 sale per every 500 clicks does not deserve to be in the BeFree network. I represent the cream of the crop of affiliate traffic to ComputerGeeks. I even got them to correct their return days to 30 so that is no reason for poor conversion off static product -banner or DHTML display links. The only diversion tactic present is 800# so I inquired about reporting interface.

I stand by the ABW proposal to devulge all diversionary tricks by merchants in their profiles and Scott has taken the first steps to see this thru. Many merchants object because their A-G ratings would imply they do NOT wish to pay out commission and use affiliates just for advertising purposes.
___________________________________
We, the affiliates and merchants of ABestWeb™, submit for the respective Ad Networks review, this "Honesty & Fair Practice" statement for the industry standards meeting on November 7th, 2002.

Definitions

1. ParasiteWare™ is technology (eg. bho's, browser plug-ins, toolbars or pop ups/sliders) used to retain or acquire new customers that knowingly or unknowingly undermines or removes another affiliate's ability to compete by changing , intercepting or redirecting a link from the originating link. Parasiteware™ may be installed intentionally or unknowingly by the end user altering their normal web browser functions and/or installing a 3rd party application that works through the user's altered browser.

In order to utilize the the services offered by [insert network name] all affiliates and merchants, as defined below, must agree to use honest and ethical business practices as defined by existing USA and International business and criminal laws. Any action -misrepresentation or software application that circumvents or violates existing or future laws will be just cause for immediate termination from the [insert network name] affiliate network with loss of all network benefits and acrued commissions.

*MERCHANT: A licensed business who collects money for products or services and takes responsibility for the delivery of the same to their customer. Merchants agree to pay all commissions derived by honest affiliate sales activity in a timely manner and devulge any activities that negate referral commissions. Those entities who agree to pay commissions on the [insert network name] network trackable actions, who are not involved in delivering a billable product or service, will be clearly marked as "special class merchants" during the affiliate signup process.

AFFILIATE: An easily identifiable hosted domain who seeks commissions for driving targeted traffic to [insert network name] merchants using trackable affiliate coded links assigned to them. In order to qualify for a commission an affiliate has to originate the trackable click from their own authorized domain or from an e-mail marketing piece identifying their authorized domain.

___________________________

Part 2: Merchant must divulge in their profile if they use any of the following diversion tactics:

a] Whether the merchant uses a 800# call center that either acts as an affiliate or works on a commission basis without recording original referral ID# for commission credits.

b] Third party gift certificate or gift registry services, who themselves are earning affiliate commissions by swapping their ID# with original referrers.

c] This merchant selectively approves and works with "incent" and reward affiliates.

d] The merchant's number of return cookie days and whether they are disabled after the first sale. ( no need to single out CJ for this request)

e] This merchant has passed all network test purchase transactions done on a random basis at least once a month.

f] Some sections of this merchant's website contain non-commissionable products or links to partner sites that do not track commissionable sales.

g] This is a special class merchant who offers commissions based upon recruiting customers for outside parties or offers loyality shopping incentives.

WebMaster Mike

"Vision without Action is a daydream .. Action without vision can be a nightmare"
Combine the two and you have the makings of a profitable reality!

5. >>I would think the best merchants would want a metric posted; Basic performance metrics is probably something that you could get on a voluntary basis.<<

While there's the ones who don't want to post because their metrics suck, there are others who don't want to post because they're concerned about their business info. Some companies are funny like that.

>>If it costs an affiliate \$100.00 in time and lost opportunities for each failed test, you can see why they tend to be angry and cynical.<<

No question.

>> You are also right, a lot of what you see is people who don't understand how hard it is to make a buck in this industry. 700 impressions is not a legitimate test. The 700 impressions probably includes 50 impressions from the affiliate while making and testing the page. The hit count probably includes 5 test clicks from the affiliate.<<

Did I say impressions? I meant 700 clicks. It's easier to show how 700 impressions can only equal a few clicks as you just did and that a few clicks does not a valid test make.

It's very difficult to understand how a program that converts 1% still may not generate a sale after 700 clicks. As you clearly know, stats and averages don't work well on small data sets and in this world, 700 is still small.

--scott

--
Scott Jangro
sjangro@befree.com

6. 800 clicks at an EPC of \$10 is an \$80 loss...ouch.

Protophoto - Short Stories

7. I knew you'd own up to that Mike. [img]/infopop/emoticons/icon_smile.gif[/img]

>>Thanks yintercept and Scott for addressing this issue. Scott was referring to my inquiry as to why, from the exact sames pages with similar product displays -equal pricing and same shoppers, would ComputerGeeks not show one sale on 800 CLICKS (not impressions) when TigerDirect reported 1 sale per every 60 clicks.<<

The thing is, you sent thousands of clicks to Tiger to get that 1.666...% conversion rate. With that conversion rate, there could have easily been a stretch of 800 clicks where there was no sale.

Mike, your list is excellent. Twist it around into best practices ("having" return days, "allowing" repeat orders, "not" displaying an 800 number, etc.) and decouple it from the scumware issue and you're onto something that merchants might actually want to participate in.

--scott

--
Scott Jangro
sjangro@befree.com

8. I think the chance of having a gap of 800 plus clicks without a sale with a client that has a 1% conversion rate in pure random data is 99^800 = 0.000322222. Of course, web hits are not a completely random function, there could be something about ecomcity that causes a disproportionate number of fruitless clicks on this particular ad, but 800 unit gaps should be rare.

Protophoto - Short Stories

9. I'm certainly not going to argue with your math. What I do know is that on the web, things are very unpredictable in small numbers.

And as you suggest, search engine traffic to a website is anything but random.

--
Scott Jangro
sjangro@befree.com

10. I'd have been willing to lay good odds that someone who chooses a user name like Yintercept would have better-than-average mathematical talent!

The math shows a big reason why being able to talk with other affiliates in contexts such as this or other boards is valuable. Consulting with other affiliates provides access to a larger data sample (informally, at least), which helps to sort out true problems from "normal" data spikes.

A book I have found helpful for keeping a balanced perspective about my affiliate stats is "Innumeracy: Mathematical Illiteracy and Its Consequences", by John Allen Paulos. I recommend it highly.

Elisabeth Archambault

11. Math is definetley not my forte ... But my common sense tends to make up for it!

Point to Newbies ... Use your common sense.

Dave

12. I think I'll change my name to "diff-eq". [img]/infopop/emoticons/icon_wink.gif[/img]

--
Scott Jangro
sjangro@befree.com

13. quote:

Point to Newbies ... Use your common sense.

Dave

I agree. But I also think it is important to ground one's common sense in reason, and to do some good back of the envelope calculations before launching in on a venture.

For example, lets say the click through rate on a ad is 1%, and the conversion rate of the merchant is 1%. You will need to display 100 ads to get a click, and 100 clicks to get a sale. Your expectations would be 10000 clicks per sale. So if a sale nets \$5.00, you would get \$5.00 for displaying 10,000 pages.

These are low end numbers, but make a very good starting point. Your goal is to make the different numbers better. Common sense says, don't run that ad. [img]/infopop/emoticons/icon_wink.gif[/img] find a better use of your time.

The problem of course it that we don't have the information we need to make sound business decisions.

Protophoto - Short Stories

14. Great post, Y. :thumbs up:

quote:
700 impressions is not a legitimate test. The 700 impressions probably includes 50 impressions from the affiliate while making and testing the page. The hit count probably includes 5 test clicks from the affiliate.

Also, add in another 50(?) clicks for spiders, etc. Depending on how much and by who you get spidered.

I'm really getting tired of I sent merchant X 100 clicks and I got no sales, they suck!

--
...only in the affiliates interest...