Hey STM folks,
Let me make this simple – if you have any questions, let me know in the comments.
To start with, I have a computer science/math background and this is my first venture into PPC.
I’m working on a spytool that auto-suggests, with some level of confidence, winning campaigns. I’m starting by training a decision tree classifier. The problem is: I need more features!
So far, my features include:
How long the campaign has been running
The image of the campaign (I’ll be experimenting with this)
What times the campaign is spotted
As you can see, 3 features probably won’t cut it so I need your help: what are signs that look at to see if a campaign is profitable or not. If we can boil down the potential features that make a campaign profitable, we might be able to automate the process of discovering profitable campaigns.
Sidenote: for the curious, I will be experimenting with converting the image to a feature using perceptual hashing or detecting objects using a state of the art algorithm like SIFT to pick out features in the image for comparison.
there’s already this out there though …..
There is? Where?
Did you have any specific traffic source in mind?
and yet another aff stuck in the "copy" mindset
Profitable campaigns are the ones that are running for the longest period. It’s that simple.
I don’t think I can name names of products but theres half a dozen I can name
Sounds like a recipe to find burnt out banner blind campaigns. Ones where you know part of the equation, you’ll never know what specific targeting the person is "really" doing to make it work. Constantly playing catch up and always losing.
Learn to be creative and innovative. Write copy. And know your traffic source inside in out.
The easy way rarely if ever works.
Maybe this is the drunk me talking.
But my 2 cents…
My strength is creativity. I take pride in coming up with "awesome" new campaigns, angles, landers, the works. It’s what makes me money, and stimulates me. However, I know for a FACT people make a lot of money copying my campaigns. You can make money ripping other peoples campaigns, everyone one who has replied in this thread has done so.
If you want to make the big bucks then you have to find your own advantage. Whether it’s a traffic deal, offer deal, scaling set up, tracking set up…something that the general aff crowd doesn’t have.
Hmm don’t think I’ve ripped you campaigns… just saying
It’s not just about ripping campaigns. I’m trying to develop tools to discover novel patterns that make campaigns successful, too.
As an example, it’s possible to take images from campaigns that are successful and perform clustering on them to try to find new ways to group them. Maybe we have a group of images that cluster together and we find it’s because they have cats in them. In addition, we find that this group is extremely profitable. Same goes for colors – maybe cats and the color purple perform extremely well for some reason. There are many possibilities. And there are algorithms developed specifically to discover as many novel pattern as possible.
While it is possible to do this manually, imagine trying to make sense of a database of many thousands if not millions of ad images. It’s impossible.
Big data and data science is huge! Don’t be afraid to take advantage of it
Note: I am drunk posting this, too. lol
I think your missing the point(at least my point)…time is the most valuable asset we have as humans….you want to spend yours focusing on a tool that may or may not work right, and provide you with meh returns in the long run…
why not take that time your gonna devote to this project and go launch camps instead…..if you tap into the creative side of your brain….you might come to find out that the creative side trumps so called "big data" any day of the week….because at the end of your "big data" pile your gonna be left going in circles, testing images and concepts that everyone else has already beaten to a pulp…some of the best camps/winning ads have come not from copying what others are doing, but testing wack ass outliers and locking down deals and leverage points that my competition doesn’t have…
I wasn’t claiming everyone on this thread has ripped my campaigns and made money. I claimed that everyone on this thread has ripped campaigns and made money.
Please tell me if I’m wrong
there is no real "formula" in the strict sense of the word to a winning campaign, so a program is going to inherently fail here. Spy tools are only even useful when viewed through the eyes of someone who already has an idea of what they are looking for and knows what to do with it once they see it. With the very large amount of effort this project would require to pull off properly, you could have learned the in’s and outs of a traffic source and know how to repeatedly make winning campaigns. Most of my biggest campaigns, I didn’t even know they were going to take off the way they did. I thought they had a good chance, I was putting my best effort forward based on what I knew, but in a scraper, they don’t look anything special. I see my ads in the scraper all the time, and very few people copy them, because they look the exact same as every other campaign running. Unless the copier had an in depth knowledge of the traffic source, understood my bidding strategy and targeting an day parting, and several other factors i’ll leave out, then having that data would be useless and the campaign would not be profitable.
There is no magic "cats + purple" in affiliate marketing. I know that was just a broad example, but the people telling you to split test your CTA with 50 different colors are the same people trying to sell you get rich quick ebooks. There’s no real magic combo like that. Those kind of changes typically only make a marginal difference at best. The changes that make the difference are the factors mr green listed in his previous post, about how to set yourself apart from the other affiliates as far as what tools you have access to that the rest don’t (by tools I’m referring to traffic deal, exclusivity on offer, etc).
I come from a computer science background and with all the tools I’ve created for myself, it doesn’t necessarily make the most money. You can "game" system all you want, but when it comes down to it, being creative wins.
Since you have programming experience like me, what I would suggest is learning how to get something profitable and then build a tool that helps you automate those tasks.
I built out my own text ad maker using a Clickbank hopfeed and it did help me generate $xxx,xxx in revenue, but I ended up rewriting the ads anyway. All it did was tell me there was traffic, but not what was profitable.
Focus on deploying campaigns and if necessary analysing your own data in this way -e.g. image processing of all your FB dating ads that you’ve tested to see if there is any relationship between image content and CTR from a perceptual/mathematical sense. But don’t get carried away with this as a side product or with spying in general. Train your marketing sense first before polishing your spy goggles.
There are plenty of established spy tools that will cost less than the time value you will spend on this.
I’ve spoken to some extremely skilled computer scientists and AI experts about this sort of program in the past, and there’s definitely some potential there – but as Zeno says, your current plan sounds like it’s biting off a bit more than it can readily chew.
If you’re looking to develop an autonomous machine learning tool for AM, I’d recommend starting on a much smaller problem – and as Zeno says, analyze your own data, as it’s the only dataset you can really trust. Perhaps look at identifying the best headline or image to begin with – there are some really good natural language processing and computer vision algorithms out there that you could start testing. They’re both big, hairy, complex problems in their own right, and if you solve either of those to a reasonable degree of performance, you’ll make enough money to buy a medium-sized island.
However, just learning to do AM and making a truckload of money that way will, I suspect, be significantly easier.
people telling you to split test your CTA with 50 different colors are the same people trying to sell you get rich quick ebooks. There’s no real magic combo like that.
I’ve got to politely disagree with the statement above – while I agree that testing 50 colours is a tad on the extreme side, I’ve had campaigns go from a 20% ROI to 120% by simply changing colours. I’m not saying that it has that effect on every campaign but a fair few of them.
another great example was comparing 2 PPC where the only difference was having/not having the "www" in the display URL – MASSIVE DIFFERENCE!
Still, definitely not a magic combo and I only both with that degree of testing on profitable campaigns but EVERYTHING is worth testing!