Amazing idea. I guess user interfaces for websites could follow a similar route some day. You specify the parameter range for the appearance of a website and the system finds the appropriate settings for a given demographic/geographic segment/time of day.
Hey guys, I wrote something called Genetify that is designed to be a general-purpose, unobtrusive way to optimize any kind of content on a web page for any arbitrary goal. Check it out at http://demo.genetify.com .
I really do want to make this into an open-source project for everyone. There a lot of cool optimizations you could do with a web page once a goal has been defined. If you are interested, could you post something at least indicating your support on John's blog? I think it will be the de facto place to discuss the open-sourcing of Genetify.
Actually, I have been thinking a lot on this idea. It is called adaptive web design. Cluttered web pages can be made a lot less cluttered by adaptively removing elements which are rarely used. Plus the system can also automatically test variations in designs, colours, etc depending on the user profile.
My idea was to make a javascript library which when included would automatically analyze the page structure and make the pages adaptive without much of an effort from the developer. The only thing here is that heavy javascript processing would be required to modify pages on the fly. Or, this can also be done on the backend but that would require templatizing existing websites (which could further slow down adoption of this technique).
I think the coder of this service replied with a similar example. Apparfently he started from customizable js websites and did ads later. I hope he moves the website thing to an open source project.
I wrote the software behind SnapAds. Before that I wrote an optimizer service for web pages that operates on HTML, CSS or Javascript. Check it out at http://demo.genetify.com .
Let me know if there is any interest -- I'd like to open-source the project.
landing pages for user signups,
homepages for media websites (nytimes and other big blogs),
email campaigns,
product promotions and offers or even defining pricing if you do a saas company
as a point of clarification, the system greg's built goes much further than simple a/b testing. there are quite a few complications with implementing a genetic algorithm in the real world of constantly shifting preferences.
once done right, it can efficiently optimize millions of permutations with millions of ad impressions (<2000 clicks). more permutations means more chance for improvement, and this just isn't possible with a simple a/b testing solution.
Slightly off-topic...Just over three years ago I came up with names for my future products that all started with Snap. Only SnapLeague made it anywhere. But I still have other domains. I wish I hadn't named it like that though because Snap as a prefix and suffix is getting overdone now. Makes me wonder if I should rename an upcoming product to something other than what I had planned.
They don't actually. Ad agencies have people who do this stuff for a living (generating what they feel will be the optimal ads). Agencies have been held back by these people in a way, because they are naturally very resistive to someone telling them that a computer might be able to do their job better than they can, even if the end result may not be as aesthetically pleasing to them.
In terms of automating the variations, that can be done ahead of time with photoshop scripting/etc, and then leaving it up to the adserver's optimization algorithms to serve the best-performing creative.
you're talking a couple generations behind what we're doing. there is a huge difference between plugging 10 variations into an ad server and using a genetic algorithm to quickly and efficiently find a local (or global) maxima in a search space of over 1 billion possibilities with constantly shifting preferences.
smartads is also a very different product, very focused on offer optimization, with a rigid system that doesn't expand well beyond the specific verticals they've created templates for. most players in this space have fairly limited optimization abilities, ie a/b testing, which does not scale up.
We're actually working on both at the moment, and it's certainly not just a side project. Over time, the two companies will eventually each have an entirely dedicated staff.
I wish Google had an API to retrieve Adsense click data for specific ads, you could easily implement the same 'genetic optimization' process and drive your revenue to their max.
What's really needed is an open source version of this so that websites can integrate custom fitness functions.
Obviuously for something as easy as "clicks" this is trivial, but the real money will be made when you can integrate it into payment systems. The problem is that it's so easy to write your own that I doubt an open source one would get much traction.
