Man vs Machine: is manual mediabuying rapidly dying? (15)
Hey guys,
one of the biggest trends of 2016 (and some years before) was AI. Machine learning, deep learning, neural networks etc
its also implemented a lot in todays mediabuying - automated buying via ML DSPs, automated ML based MVT etc etc
seems that in no time machines will completely outplay humans in our field of work.
what do you guys think?
do you develop your own solutions?
will man made media buying die completely in the coming 2-5-10 years?
I don't know but it sounds like it will. But it will be the individual knowing how to game the systems in order to find an edge over each system.\
Originally Posted by buck johnson
anyone else??
So far systems like Monetizer and YTZ are using a lot of data , split-testing many more landers and offers than a single affiliate or affiliate team can, and doing it more efficiently. Still when we split-test those systems again our own winning campaigns with the same traffic in the majority of cases our campaigns performed significantly better, even though we did only a couple split-tests in
There could be many reasons for this, but a big one certainly is that our winning campaigns have landing pages and offers that are not available on those systems. Those are the men made parts of their systems. And it's obvious that us focusing in on a winner with creatives and offers to test just there is going to work out better than YTZ or Monetizer who have to support 100+ countries and much more broad traffic at the same time.
Is it always going to be this way? Probably not, I think those systems are going to get better, they're going to take over more of what's done manually today.
But at least in the near future , 10 - 20 years , I think there will always be humans needed to do a big part of the job, a make or break it part. A lot of the marketing is about understanding emotions, abstract thinking, being creative, also the negotiations with networks and advertisers of course, the relationships .. I don't see how a computer is going to do that anytime soon 
Developing your own solutions can certainly be a good idea, if you're big enough and it's worth the time and money this will take. I think MrGreen wrote about Tech as his Edge a while ago, and how it's a better Edge than Creativity in AM. We do not work on any tech like this right now, but no doubt if we had a system that would be more automated we'd safe time and money and could focus more on what we do best 
Machine Learning & Deep Learning will take over most of the "Non-Creative" work very soon.
Think about it, with image recognition (Computer Vision) and machine learning you could find out what gives a banner its high CTR (Color, Image, Text, Border etc..). You can do this on a scale no human could ever do. Same goes for Copywriting, Landing Pages (Layout, HTML Structure, Copy etc...)
These guys are developing a "Machine Learning as a Service" company.
https://www.youtube.com/watch?v=DSlGQ49EoWc
There are endless opportunities to apply Machine Learning to Affiliate / Performance Marketing
On one hand algos can manage way more campaigns or processes , on the other one, machine needs a ton of data to make a qualified decision - they cannot "guess" what's gonna work in terms of image to use for a banner for example ... at least not yet. Considering this, an algorithm will need to spend more than an experienced human, in order to find a winner and in some cases, that much data or time might not be available at all.
I can see machines taking over some analytical parts of the process and even beating humans at it, but I seriously doubt any algo can replace a human media buyer anytime soon. We were and still are using auto-optimization algos or smarlinks, but anytime we tried to optimize everything manually, the results were better. Algos just saved time and allowed us to push more volume, but at lower ROI. Such is my experience so far.
Originally Posted by kepe95
But at least in the near future , 10 - 20 years , I think there will always be humans needed to do a big part of the job, a make or break it part. A lot of the marketing is about understanding emotions, abstract thinking, being creative, also the negotiations with networks and advertisers of course, the relationships .. I don't see how a computer is going to do that anytime soon
The first is facebook, of course. Because in other system (Google Taboola etc) page content will also have an impact (momentary emotion), and this task will be more difficult
The AI tech that's currently available is much more advanced then the stuff that's available or being built by people in the affiliate industry.
So until people are investing tens of millions, the tools will always be lagging behind the current state of the art stuff. Just don't confuse the weakness of tools in the affiliate space with weakness of AI to outperform humans in many things including tasks at are traditionally more creative and suited to humans.
I haven't had a chance to test or use this, but check out https://persado.com they have AI that writes content that out performs humans 100% of the time. Imho pretty impressive claims, I'm sure there's some small print to that. But still they show stats of an average lift of 49%
We are not at the point where people are getting replaced by machines but that day is coming and it's not 10-20 years away. I'd say 3-5 years but wouldn't be surprised if it was sooner.
no it wont, because in order for all these automation tools to work, they need data, and they need a lot of data
most people do not have 100,000 to invest and lose half of it, to buy the data.. they have a marketing budget only a few hundred bucks per month if that
so, no media buying will not die, because there always will be people servicing the average little folks and their tiny ass budgets ..
When a homo sapien buys something, they do everything they can to look for genuine, human signals from the company they are buying from.
Scripts may work for a short burst of time, but they quickly stop working once people sense a pattern.
Its the same thing with pickup lines...
And its the same thing with marketing...
So just like with women....
When you find a winner, _________________. (fill in the blank
)
Originally Posted by Mr Baffoe
I like what Persado is talking about, but honestly, I dont necessarily believe the claims either.
I doubt the companys claims to AI based on two factors:
1. Clients: If the company's 'action-oriented' message AI was so powerful, they would have expanded outside of Fortune 50 brands MUCH quicker with lite/etc versions of their software to saturate the market. In my mind, marketing companies that service Fortune 50 companies tend to have: super high-paid MBA-type corporate business development teams and multiple levels of very NON-innovative account managers. it seems like this has happened way to early on the company's trajectory, and they probably dont have anything super useful, so they're cashing in.
2. Investors: If they really did have the AI they talk about, their investors would be better than the new york oriented investment arms of companies like goldman sachs or bain capital. I would have expected to see a sequoia or something in that realm in the loop. None of the Sequoias are in the mix though, which makes me extremely doubtful.
I've actually been having quite a bit of discussions recently with a fellow STM member about developing our own AI. Persado pique'd my interest because our whole concept is based around 'self-optimizing AIs', basically, giving our AI a conversion point of which to optimize around.
Alexa, and all these other AI's dont have 'conversion points' of which to optimize around fully, because they are meant to help with everything, instead of being the master of one area.
I believe each vertical will have its own AI's, that optimize around different points.
A flight-booking AI...
A luxury car sales AI...
A book suggesting AI...
A weight-loss plan AI...
etc.
I think that AI will basically be the websites of VR.
Websites are really lame, they're 2d and impersonal.
An AI can be made to look/act/talk/sound like a human (personal...human connection....etc) and can be much more interesting.
People wont visit websites in VR, they'll summon AI's for specific purposes (book a flight, etc) and talk to that AI about what they want to do.
My guesses on this one, based on my research in the space, are two-fold:
1) Yes, much if not all of affiliate marketing will eventually be automated.
2) It'll happen a lot slower than most people think.
AI is hard. AI skills are very expensive. As Attila says, they also need a lot of data to work with.
And progress in AI largely consists of trying a lot of things and 99% of them not working. Much like AM.
There is, in theory, no real reason a reinforcement learning AI couldn't outperform a human right now as an affiliate, but the cost of testing and training that AI would be prohibitive. And the weasel words "in theory" disguise the fact that in the course of spending the billion or so dollars you'd need to spend, you might well discover something that the AI just couldn't do, and have wasted the lot.
To expand on that; one of the reasons why AI will take longer to conquer this field than you might think is that reinforcement learning, the most powerful AI technique right now for gamelike applications (which AM fits into) requires the ability to extremely rapidly iterate on a problem. You start from an AI that knows nothing at all, give it some tools (Photoshop, say) and a goal ("Get conversions on this offer"), and then set it off iterating.
We iterate a lot as affiliates, but we think of "a lot" as, say, 50 or a hundred campaigns.
In order to train a deep learning AI to, say, create winning banner ads, you'd need millions of iterations. And the issue here is that you can't simulate the iterations, because they're entirely dependent on human interaction. That makes the process prohibitively slow and expensive.
(People with more AI knowledge than me, feel free to correct me if I've gotten something wrong here. I know quite a bit about the space - and indeed my task-list for today includes "train neural network" - but that includes knowing there are people who know a lot more than me!)
There are still a lot of very interesting areas for research in AI solutions in the affiliate space - particularly if you already have a large dataset, so don't throw away your old campaign databases - but as I've argued at AWC events before, you shouldn't worry about pure AIs taking over the space so much as you should worry about humans ENHANCED by AIs. At least for the next decade or so.
thanks a lot for your replies, guys!
but AI can definitely narrow out profitable variables combinations in 1 msec better than 300 analytics ever could. and see way better patterns.
also - automatically adjust the bid price to a certain set of variables (say "M, 40+, CA, ios" etc) prior to buying inventory based on the previous data. and also make all of the remnant traffic way more expensive after that due to the fact that it will set higher bids on something it knows is converting.
creating creatives and writing copies is a tougher task for it for sure, but look at this:
https://www.literai.com
its a community of artificial fiction. its rather raw and made by enthusiasts, but still. not long before we can see computer writing a way better copy than you and me.
other thing is that it's mostly big brands who are dipping their toes into AI (due to the facts mentioned above which is $$$) which actually is more scary at the moment for useless "brand mediabuying" companies whos only valuable report parameter used to be "raw amount of eyeballs".
but that will inevitably make buying inventory more expensive for AMs in the near future.
I've been reading a bunch of books centered around AI recently (authors including Neal Stephenson, Vernor Vinge, William Gibson, etc) and an AMAZING, academic book by an Oxford professor called SuperIntelligence, which discusses the paths in which superintelligence will come about (simulated intelligence, genetic intelligence, artificial intelligence), the dangers of superintelligence, and the strategies for dealing with it. Very interesting book, and its lauded by both Bill Gates and Elon Musk, so yea.
Fairly simple stuff so far but here are a couple of articles on the subject:
https://medium.com/the-mission/looki...16c#.8ukhikkna
https://www.theguardian.com/lifeands...ader-it-worked
An obvious next step would be to add some kind of simple NLP (Natural Language Processing, not the other one) to the bots, maybe some sentiment analysis, and then start split-testing responses...
I've heard good things about "Superintelligence" too from people I respect. Must get around to reading it!
Home > General > Affiliate Marketing Forum