Pick His Brain!
I’d like to introduce one of our members, Eric Ortego, for our next ‘Pick His Brain’ session and I want to thank him for the participation.
Eric is the real data-driven marketer specializing in a predictive analytics system that operates on large datasets. These datasets that are used to predict buying behavior based on an identity graph on over 220 million Americans that is updated daily using more than 55 billion data points.
He is also an expert Linux system and security administrator. If you have any questions related to data-driven marketing or Linux, please feel free to pick his brain.
Note: Because of his hectic schedule, he may take his time to answer your questions.
Here are the rules.
1) I’ll let the thread go on until he asks me to stop. Theoretically, this thread can continue until bitcoin value goes to zero.
2) Please, no snarky remarks. I will not tolerate any intentional negativity. We are here to learn from each other’s success and strategies.
3) Please do not PM him and bother him. If you have a private question, ask for his permission on this thread when appropriate.
Can you tell us how predictive analysis works?
The way we use predictive analytics is via deterministic data analysis and machine learning geared to watch for certain behaviors that would indicate a person is ready to buy certain goods and services.
We monitor certain verticals that are most profitable because buying the data regularly is expensive and difficult to maintain.
The way this works is we buy data from large providers. We take that data and parse it with the AI. The AI is programmed to learn and improve its ability to forecast buying behavior based on previous sales data.
It’s called closed-loop predictive analytics because we close the loop on data points and predict buying behavior. We monitor between ~2.6 and ~4.3 device IDs per person right now.
This means that we are able to track a person across multiple devices and multiple email / logins to get a better understanding of what that person is interested in at any given moment. All of this data is stored in an identity graph.
We use that to maintain the identity of all the people we monitor so we can tie it back to a hash that is used in the various major ad platforms to make custom audiences.
How can a marketer take advantage of it?
A marketer can take advantage of our data services in two distinct ways.
The first is we can analyze the traffic you get across your website now and actually match that anonymous traffic back to a person in the identity graph mentioned above.
This is identity resolution. Any website can benefit from this. Because we monitor almost all the Amercian buyers in general, we can identify them. This isn’t tied to any specific niche.
After we match the person we can then send them email or direct mail. You can also use the matches in custom audiences in all the major ad platforms. So you don’t get locked into one platform.
This also helps boost your current marketing budgets because if you identify a person using a Facebook or AdWords campaign you can then take that match and remarket to them as you please.
You don’t get tied to using your discoveries only within the platform from which they were derived originally. Run ads on Facebook and remarket to them via AdWords, or vice versa.
The next thing is in-market lead data. We get these datasets based on the verticals we monitor.
Most major niches are covered. All types of legal, financial services, insurance, dentists, chiros, precious metals, etc. You can take these datasets and use them in all the ad platforms to create custom audiences and better-refined lookalikes.
This is because you’re getting a list of people in a certain geographical area that are ready to buy now or in the near future.
After you run ads to the datasets and have successfully remarketed them to your website, we can identify them again and then email or direct mail.
How did you discover this field?
I got into data services because I was looking for a way to differentiate myself from other marketers.
As you mentioned, I started out my career as a Linux systems and security admin. Before that, I put myself thru college using SEO and affiliate marketing back in 98/99 to around 2003.
Then graduated with my CS degree and went on with that path for a long time. A few years ago I decided I wanted to try to start a business.
I looked into ways to utilize my skills and settled on SEO. Then all my SEO clients wanted Facebook only so I switched for a while to that.
Then Facebook was so volatile I had to expand to other methods again so I set up the ability to do full-service marketing online. From email to the ad platforms.
Again I found myself average. So I looked into ways to set myself apart and found a partner who was into predictive analytics and data. Artificial intelligence is the way marketing will be conducted moving forward.
People have already been using these services with the partner categories within Facebook, for instance. But those go away October 1st.
Now people will have to pay for the data. I feel like this is a better fit for me as a person.
I am more at home with systems and data. It’s nice that those skills are important for modern marketing methods.
What do you think the future looks like in regards to site personalization depending on demographic? ie. I go to a site and it tailors all content and offers to me personally. Any insights on that from your research and data?
As for customizing a site based on that particular person. That’s an interesting question.
Right now you can do some rudimentary form of this using standard information collected from the traffic source.
The future is just ahead to do a much more laser-focused version of this based on identity graph matches. In the coming months, we’ll be able to derive a profile for each person that includes all their basic likes and dislikes etc.
I’m not sure how you could adjust a site on the fly for a person, but after they visited your site and return you could then display a more tailored experience to them. Or even get datasets for a certain vertical and use those matches to pre-customize an experience.
I could see certain high ticket item websites doing this, in fact.
If someone wanted to get into predictive analytics, would are some possible paths to get in?
Are you looking for a technical role or how to use the technology as a marketer?
If it’s a technical role, you’d come up against folks in the industry with some type of Computer Science degree coupled with extensive math and programming experience.
Understanding of algorithms and how they are used in prediction is required. Being able to work with large databases is key. I helped a recent college grad get a job in predictive analytics this past year.
He had a computer science degree with an emphasis on machine learning. It was enough to get an entry-level position. I’m sure if you have some experience with programming already it wouldn’t be too difficult to secure a position. The field is growing.
If it’s as a marketer, I mentioned in some detail above how this is done.
Basically, we use our system to identify people based on traffic analysis and correlating that information back to a huge identity graph that we maintain.
We also use it for marketing by feeding the AI with current data and allowing it to predict buying behaviors.
Have you done any machine learning on SERP data? I’m working on a project to classify SERPs based on user intent, and I’d love to know if you’ve made any headway there.
Thank you for your question first of all!
Neither myself nor my partner has done this type of analysis.
We focus on monitoring certain verticals for buying behavior. The system is special built for this task and identifying traffic only.
You pose an interesting scenario though. I’d love to hear more about it, actually.
What are some great public/open source or paid data sets specifically related to marketing?
Here’s a start for you.
There are some marketing uses you could derive from these datasets. But the thing that sets what we do apart from the capabilities of these datasets is that the data is updated daily.
Do you use the Tensorflow library for ML projects?
I’m actually not at liberty to discuss development, but I can tell you that the dev team has used TensorFlow in the past.
Good question, very detailed. Are you working on any AI projects currently?
Don’t shortchange yourself. Besides, I’m not coding on this project.
I do code, but not for this system. There’s a dedicated team working on that.
When doing this predictive analytics system, so most of your work is monitoring only? Since you have this learning machine installed to watch it, am I right?
The system takes in daily datasets from several sources. The machine learning code analyzes the datasets for behaviors it is programmed to predict.
So when it sees certain behaviors being triggered by a person, they get categorized into some number of potential datasets based on niche.
Since people are generally in the market to buy multiple things at any given time.
After we detect a certain number of hits from a person per niche we can weight them according to when they should buy. Based on previous buyer data.
What is your preferred stack for websites?
Personally, I prefer LAMP.
I like WordPress on top of that. And Oxygen for building the sites.
My favorite type of Linux is Kali though. I’m currently in training for my OSCP now. I am getting it to stay current on skills for pentesting. My other life.
But I don’t want to let all of that go to waste so I stay up on things in the security realm. I also have my CISSP and GSEC.
I’ve used Linux as an admin for over 20 years though. I’ve used many flavors along the way. Got my CS degree on Unix and some Linux systems.
But back then Linux was still sort of newish. At least for a university.
I spent a few years as head over a high performance compute cluster at UT with around 800 nodes that was tapped into a Bio-Linux stack.
For their genetic research labs and bioinformatics studies.That was kinda cool.
Linux has been good to me over the years.
Do you have a top 2-3 security MUST-DOs you’d suggest for a website/server security hardening?
I’d say to protect a WP site you surely want to change your admin login name, obfuscate the login path and leave a dummy one in place to log IPs, and limit the frequency of failed login attempts. Check your logs regularly.
And I like at least some sort of perimeter security even if its Wordfence.
Of course there’s many more things to consider. Like updating your plugins and replacing ones that aren’t updated regularly.