Thought I'd share some of our thoughts on the future, as we working on bring more intelligence into the tools that you use daily as affiliates and marketers.
We initially started of looking to build an AI Brain for just Prosper202, but as we started working we realized how much better it would be if we could build a centralized AI brain that can be bolted onto any of the marketing tools and services you use.
Data is ingested from any and all tools you use, so the AI Brain can learn and get smarter much faster.
One thing we are pushing hard for is a system that works on your own Private SaaS or Cloud.
The advantage of this is that, you end up building our own personal AI brain based only on your data. All your hard earned testing and data collection is not going to be pooled together to share with competitors. There are still a few issues we need to figure out, since there are cases where pooled data may make the most sense.
Finally one thing that we have changed our mind on is to role of AI and Machine Intelligence in your marketing stack.
We started off looking to build something that could replace you 100%, but now we feel a more effective system is providing you with something like JARVIS and Iron Man's suit. Both work to make you better at what you do, but at the end of the day your skills and brains play an important part of your success too.
There's been some interesting research (https://www.hfes.org/web/DetailNews.aspx?ID=362) on how as things get more automated, people become less skilled, so when things go wrong the human doesn't have the skills to resolve the problem.
I've got a longer post on some of what we are working on here:
http://prosper.tracking202.com/blog/...ated-marketing
Love to hear your thoughts and input.
I have a custom tracker + automated optimizer / scaler, which has been in development now for over 3 years.
And a major problem that is preventing the machines to learn correctly is noise:
the inability to replicate campaigns in same conditions.
For example.. A campaign is made with 10 creatives, 2 lps, 100 placements. System runs it and determines the winning creative, lp and lets say 10 placements.
The system then replicates the campaing using the winning creative, lp and placements. The new campaign does not replicate the sucess, it sinks and poisons the algorithms completly.
Another problem: System uses defined CPA, let say 10$ for an offer that gives a payout of 13$ for example, to give a healthy 25%ish margin.
Optimizer runs.. kills every killable thing (placement, os version, creative.. whatever) that is over the CPA. But consider this.. It kills a parameter that does 90% of the conversions yet is just slightly above the CPA lets say 10.5$. What do you do ? Adjust the CPA ? It solves nothing.. there will always be bordeline cases. So now in order for algo to work properly it has to have a fuzzy logic implemented, that takes into account also volume. Fuzzy logic = shitload of potential problems (eg campaigns auto-optimizing at 0 profit).
The main issue however is the selection of data window. Advertising has a creative decay, everything is affected by it. So which data interval is chosen so it has some resemblance of statistical significance that is valid for the recent data as well as the longer periods. The signal for the machine needs to be statistically significant, yet by making it statistically significant the data becomes toxic, since its has a performance decay in the recent period. So now we need to deal with decay curves, which are specific for each creative/campaign/lander.
Am a bit sceptical if your or other algorithms on the market address the issue of the data window, noise or fuzzy cpas properly.
perhaps the solution is not to rely on automation to make decisions but rather to remove all the manual work in getting setup and ready to analyze. I see AI in the near term solving one simple problem, The computer should never ask me to input something it can know from previous inputs, and should give me one click recommendations based on ML and AI best guesses. why should I filter my data the same way today as I did yesterday. It should get me to the point of just making a simple decision.. just my two cents on what excites me about chatbots and AI.