Putting SCIENCE to work for porn
 

Are there any webmasters out there that wish to collaborate in using statistical models to fine tune advertising campaigns and increase conversions?

Are there existing scripts that already do this?

mmmm.....collaborate

I use a modified version of PHPAdsNew that integrates traffic stats with click tracking and with my sponsor sales stats... really it's going to depend on what you're trying to do and what kind of campaigns you're talking about.

Don't hate......... collaborate :):thumbsup

Good point. That may be a starting point. But one thing that came to mind is that a system wide/multi site approach would be a great test since they have different kinds of traffic.

It's relatively easy to modify PHPAdsNew to cope with a whole network of sites, as long as you design the database that you write everything to is well thought out then you can do the reporting in any number of different ways. The scripting is relatively easy (I just use PHP and MySql), it's the design of your database and which variables you want to track that is tricky.

I basically only do organic SE traffic and PPC though so it makes things a little easier, if you had a trillion different types of traffic coming into your system and bouncing around things would be a little trickier. But I am a great believer in scientific advertising, John Caples is my idol.
:thumbsup

Maybe PHPAdsNew can be modified to do behavioral modeling and also campaign testing that is dynamic (ie., ad structure changes to accomodate clicking behavior)

There are some powerful scripts out there, but I think the best ones are privately created and owned by people using them for their own sites.

Testing ad variations in real time isn't a huge problem. My solution (which is a bit of a nasty hack) that I use on my testing sites is to run an analysis script as a cron job, which basically tracks the clicks and sales for that hour/ day etc and then swaps the ad through a rotation. If you're using dynamically generated ads then you could swap the ad components around (eg. make changes to headlines, copy etc and then try them on different templates) and then at the end of the day you can print out a nice page of results.

Once again the scripting itself isn't that difficult, but if you have a lot of these tests running at once then you're looking at high server load. I find it best to test when I'm first starting to promote a new niche or sponsor and then leave it with the same ad for a while. I send all my traffic for a given niche into the same mousetrap which makes things easier.

That set up looks good.

What I'd like to know if there's a script that can do this:

Create ad with tags for the following variants text, graphics, content

Generate report

Mutate the variant based on the report

Repeat until we find a sustainable CTR and conversion ratio

yo yo...I got all the scripts :glugglug

but, if you got some cash, maybe we can combine a few Harvard professors with some India programmers!

I use the same type of statistical modeling analysis that Hef uses. If the curves point up you're good to go and that's all you need to know.

http://seattletimes.nwsource.com/art...002/hefner.JPG

:1orglaugh :1orglaugh :1orglaugh

:thumbsup :thumbsup :thumbsup

That would be simple, someone on scriptlance could probably do it for you for $100. You are much better getting a custom script done (based around an existing open source product so you don't reinvent the wheel, and so you have the full source for modifications later), rather than buying someone's precanned binary offering.

Hugh's the original pimp. :thumbsup

Hot!

Heheheh that's ironic given what's going on :)

hmmm...lucky guess?

Looks similar to the Tom Tom club art of the early 80s

If I knew C+ I would take this on. Sounds like a fun as well as profitable project.

When you think about it, the range is so great that it is kind of hard to go about it that way. Just think about all the possiblities of sell phrases or design.

I think this is a question more of your dataset than your program. Anyone can program a self-mutating script to find the max CTR, etc. It is more a question of your inputs.

BTW I have been royally flamed in the newsgroups for top posting...I don't really care, but some people find it rude ;-)

The old CEO Andy from Sextracker wrote some articles a good
while back on what he called "traffic vectors". For each site he
draws a map of where the traffic is coming from and what it's
been exposed to already. This map is then used to choose
sponsors and techniques to appeal to the given sites traffic. I
believe this technology is still a part of MoneyTree however they
may have moved to a simplier system. MoneyTree tracks/tracked
which banners surfers have seen and clicked on then displays
adverts that appeal to that peticular surfer based on their
experiance on other sites running the MoneyTree program.

-Ben

That's sweet. I'll try MoneyTree if they use this model.

Is there a link to these articles?

I'd like to read them too if they're still around.

http://trix.sexswap.com

dig around in there

Thanks for the link, hudson!

Oh, np $5 submissions...care to buy me a ticket to India :glugglug

Since this is posted on a public messageboard, I guess its "open source" :)

Thanks for the ideas, fellas!:thumbsup

Although we are in the pharmaceutical business, not pr0n per-se, we do extensive tracking and analysis. Our back-end uses many of the methodologies and techniques found in this open source project.

http://www.phpopentracker.de/

Looks great, definitely a good place to start:
The analysis of clickpaths is essential when you want to measure the usability of your website. phpOpenTracker's API functions 'shortest_paths' and 'top_paths' help to answer questions like: Is the navigation intuitive enough? Do visitors find the shortest, most direct way from A to B? How are ads percepted? On what document do they leave the website -- and where to? The answers to these questions help you to improve the user experience and the way your website is percepted by your customers

Maybe plug this into an algorithm. The problem with assigning values to subjective areas like "look and feel" of a site is that there's no hard and fast objective values.

You are left with the WHAT and HOW of a click... but rarely the WHY.

Better to get CLOSE than not even be in the ball park. I still say this is still worth attempting.

yeah lets post for more association and collaboration to gain more sites

Hef's on top of it all ...

that's what I love about gfy...how quickly it goes from "statistical modeling analysis" to hef and bunny curves