If you could choose a super power, what would it be? Mine would be to predict the future. If you’re an online publisher or blogger, just imagine the possibilities of being able to know what’s coming next. Actually, that’s one super power we can actually give you.
Stop bowing. We’re no gods. We just want to talk with you about predictive analytics. If there’s one thing I remember from our short and unsuccessful attempt at being a boy scout, it’s that there’s nothing more important than good preparation.
Whether you’re developing a new online property from scratch or creating new content for an existing website, being prepared — both for costs and for potential earnings — gives you the kind of insight that can skyrocket your operation forward and turn it into a truly passive money making machine.
Predictive analytics is all the rage in marketing at the moment but it’s actually been around for a while and powers most of the software you’re already using. Predictive analytics guides for search marketers , display advertisers and other marketing professionals are relatively easy to find and packed with useful information. We won’t get deep into model creation here though.
The data is used to build statistical models and algorithms that can analyze the data to find trends and, based on these trends, predict with a certain degree of certainty future events.
By creating predictive models, businesses can manage their exposure to risk not just now, but for months and years into the future.
What is a predictive analytics model?
A predictive analytics model is a model that predicts the chance of a certain outcome based on existing data. Simply put, predictive modeling uses statistics, such as an ad’s audience profile or CTR history, to predict a future outcome .
For example, as an advertiser, you can use a predictive model to analyze data from an existing ad campaign, such as the audience’s age, gender and location, to determine how well it might perform when targeted to a similar audience.
These variable factors are called predictors , as they’re the factors that influence the results of the campaign.
If you already have existing data from an advertising campaign (such as a placement report or keyword report from Google AdWords, or an audience report from Facebook Ads) you can use this data to predict how effectively the campaign would port to another platform, for example.
If this seems a bit over-complicated or you’re skeptical that it would work then you should know that larger publishers are already doing it, meaning that you’re already paying more for traffic and monetizing less by not using predictive models.
Don’t believe me? Pretty much every piece of software you’re using to operate your website and marketing operation uses a predictive model at some capacity but here’s a use case you’ll appreciate: Mashable, one of the largest tech blogs around, considers itself a data-driven publisher. They have developed a proprietary platform called Velocity that is able to predict what going viral.
Velocity is able to crawl 1 million URLS per day to predict web traffic trends. Using Velocity, Mashable saw an increase of 55% in traffic in the first 2 years using the software.
Velocity is an app for Android and you can download it here .
How can publishers benefit from predictive analytics?
Simply put, how can they not?! Publishers or bloggers who monetizes their content with advertising have been creating content based on (educated) guesswork.
If you run a website that is oriented towards organic search traffic they you’re very likely performing keyword research to find keyword traffic volume and create content that will capitalize on that keyword’s traffic. However, most publishers aren’t creating content based on predicted traffic volumes.
If you run atraffic arbitrageoperation, which means you’re buying most of your content, then you’re always trying to find the cheapest traffic sources and the best paying ad networks. Just imagine if you could successfully create a model to predict that. You’d be the Obi Wan Kenobi of traffic arbitrage.
So what can publishers actually do with predictive models?
- Create websites in growing content niches that will be more profitable in the future than they are today
- Predict higher ad CPCs for monetizing your websites and content
- Predict lower ad CPCs for display, search, and social ad buying
- Predict the performance of an ad campaign as it reaches a wider audience
- Predict the cost of creating new content, be it text or video
Predictive modeling is definitely on the advanced end of increasing ad revenue. However, it’s a strategy that can produce real results and help you capitalize on opportunities that might not be apparent without predictive insight.
Below, we’ll explain how you can build predictive models to generate higher earnings from your existing content and ad units, capitalize on variations in user engagement, and keep readers on your website for longer.
The 3 main predictive analytics models for publishers
There are three main types of predictive models that are relevant for marketers and publishers:
- Propensity models
- Cluster models
- Collaborative filtering
Let’s start with propensity models, which use past data such as a person’s demographic profile or location to predict their behavior.
Propensity models use propensity score matching to estimate the effect of a certain action. In simple terminology, this means that a propensity model will look at existing variables (such as an audience’s demographic profile) to predict actions.
When used effectively with a large enough data set, a propensity model can accurately predict the amount a new customer will spend on your product, the frequency at which a new user will click on your website’s ads, or the percentage chance of a person converting into a lead.
For example, you can use propensity modeling to analyze your existing advertising data to see which country or region produces the highest clickthrough rate, then model this data to view the results of a 20%, 50% or 100% increase in traffic from this region.
You can also use propensity modeling to predict the results of optimizing using more than one variable. For example, a propensity modeling algorithm may discover that a specific region and device type produces a significantly higher ad clickthrough rate than the norm.
Using this data, you can model the effect that an increase in this type of traffic will have on your advertising revenue, letting you make informed decisions about which audience types are worth targeting, and which aren’t a major priority.
Cluster models algorithmically sort users into segments based on their behavior. This sorting is automatic and uses a huge range of variables — often 20 or more — to differentiate one cluster of users from another.
Cluster modeling can be used to predict factors like the amount of time an audience segment is likely to spend on your website, the average number of pageviews or ad impressions a specific segment will produce, or how long it takes for a segment to return to your website after a visit.
Since a lot of the data used to sort users into clusters is related to monetization, you can use the results of cluster analysis to determine which segments are most deserving of your marketing or ad buying efforts.
For example, if one segment spends more time watching videos on your website than any other, and clicks in-video overlay ads at a much higher rate than any other segment, you can focus on doing more to keep this high-value segment on your website consuming video content.
Free-to-play game developers use this type of predictive analysis very heavily to reach the small percentage of their users that are most likely to become loyal customers over the long term.
Cluster modeling is particularly valuable if you monetize your traffic through the sale of a product or service. Instead of optimizing for your entire audience, you can predict the results of focusing on the audience segment that’s most likely to purchase your product.
Have you ever bought a product on Amazon, only to receive recommendations for products that are “frequently purchased together” before you checked out?
That’s collaborative filtering in action. Collaborative filtering models analyze data using Bayesian networks, latent semantic analysis, and other techniques to find common factors between similar items, such as products in an e-commerce store or videos on a content-driven website . Basically, the model predicts that if you like The Godfather , for example, you will also like Scarface .
Using collaborative filtering, you can discover useful data that keeps visitors on your website for longer, encourages shoppers to add multiple items to their shopping cart, and vastly increases the number of ad impressions each visitors to your website generates. Content recommendation platforms like Outbrain are based on a collaborative filtering algorithm, usually in combination with a content-based filter. So do conversion pixels like the used by Google and Facebook.
You can also see collaborative filtering used for this purpose whenever you visit YouTube — just look for the recommendations in the “Up Next” column.
Like other types of predictive modeling, a collaborative filtering model gets more accurate as its data set grows. As users view, rate and comment on content, the model can better understand user preferences and content consumption habits.
When it comes to content, there are plenty of affordable solutions you can use that will introduce collaborative filtering to your audience. But, if you want to use this model to better predict clickthrough rates and increase your ad revenue you’re either going to have to create the model yourself or use AdNgin, which incorporates predictive analytics into the ad layout testing platform (case study).
Using predictive models
Predictive modeling can help you increase traffic, engagement and revenue by understanding how users interact with your website. It can even help you identify new opportunities for growth and optimization, whether through new content or outreach to a valuable audience.
As a publisher, you lose nothing and gain a great deal by being prepared. By using predictive analytics effectively, you can learn about the results of an decision before you make it, giving you greater insight into how you can grow your digital business.
I’m Head of Marketing Operations at AdNgin. Before coming to AdNgin, I was a marketing professional focused on SAAS business models. When I’m not working, which is rare, I sail and hang out with my son, Jonathan, and wife, Meital.