Businesses want their investment in advertising to pay off quickly and reliably. Predictive marketing can help with it. In this article we are going to explain how a predictive model works and what advantages it has based on the example of Go Predicts.
Predictive marketing makes it possible to evaluate advertising campaigns not based on just initial behavior of users, but on the forecasts for deep and significant business metrics — LTV, revenue or the number of orders. In other words, it allows to focus not on the actual user results, but on predictive ones.
How does predictive model work in mobile advertising
Users complete initial actions — adding products to the cart, making purchases. Data about these actions is sent via the advertiser’s tracking system to Go Predicts and the model is learning based on the audience behavior.
Go Predicts use this data to analyze behavior of a new user within the first 24–72 hours and predict the metric requested by the advertiser.
Forecasts are sent back to the advertiser’s tracker and the advertiser can use them on the ad platforms.
This algorithm allows to attract only those users whose predictive metric value meets the business objectives.
How does predictive model work in web advertising
The basic principle of web advertising is almost the same as in mobile advertising: data from analytics systems is processed in a predictive model and returned to advertising channels to attract new users.
However there is one important distinction. For mobile ads it is usually AppsFlyer or Adjust and for web — Google Analytics.
Also in web ads, in some cases, you need to use a third-party connector for data transfer. For example, when the user did not interact with the site, but simply viewed the ad. This connector can be created manually or you can use services for automatization and optimization of ad campaigns, where such integration has already been added.
Using predictive analytics on web has other peculiarities:
- Not all data from sources can be integrated into Google Analytics — if this data is lost the model becomes inaccurate.
- Google will stop using third-party cookies to track user activity on the Internet by the end of 2023. Without collecting personal data, it’s not possible to see about 60% of the traffic and that will complicate model learning.
What predictive analytics give from marketing perspective
01/ Users are attracted based on their future payback
With the help of predictive analytics, you can estimate what the user’s payback will be, how much profit users will bring and how soon. Based on this information, you can attract users with a high pLTV and calculate the amount of money that is necessary for the advertisers to attract such users.
Let’s look at an example.To attract one user, the advertiser paid CPI $10, and for the second one he paid $5. He deliberately pays more money for the first user, knowing that his LTV will be higher and he will bring him more profit.
02/ Optimization objective is moved to the metric important for business
The most common objective for a business is revenue. To understand whether a user will bring money, in classic marketing you need to wait for at least a month.
In predictive marketing, by analyzing the primary behavior of each specific user it is possible to estimate the value of the long-term metric that is important for business.
Let’s look at two examples of how to attract profitable customers.
Without predictive analysis:
If a business needs users who buy a lot, the media buyer sets the conversion action as «add to cart». He asks the ad network to bring in users that are similar to those who have added items to the cart. The hypothesis is that those who add to the cart will bring more profit. However, for some users this hypothesis will work, and for others it will not.
With predictive analysis:
Working with a predictive model, the advertiser asks the ad network to attract users similar to those with a high target metric. The advertising network starts to bring not 1000 users who will add the product to the cart and of which 20 will buy it, but it brings 1000 users who will be similar to those who have a high LTV. It means they will all buy the product.
03/ It is possible to buy more traffic
In performance campaigns, a benchmark is being set — the cost of a conversion action. For example, for the «add to cart» conversion event, the advertiser is willing to pay the advertising service $2 per user. But if you put up one benchmark, then you can buy at the auction only what you put up.
With the help of predictive analysis, you can find out exactly how much money the user will bring. Then you can set individual benchmarks for campaigns that are optimized for the predictive metric values from different audience segments. So you can pay for a user depending on how much money he brings based on the value of the metric.
Let’s look at an example. Let’s create an event generated by users whose pLTV is 1000–2000, 2000–15000, 15000–20000. We will pay $2 for the first segment, $15 for the second segment, and $75 for the third segment. Thus, we buy out not just one segment of the auction, but we break it down into benchmarks necessary for business and buy out the entire auction.
This is important for the businesses that want to purchase a big amount of traffic profitably.
04/ Retargeting campaigns return only segments with the highest potential
In regular retargeting campaigns, customer ROI and unit economics are blurred. The income from such a campaign will be less, since some of the users return, and only some of them order.
With predictive analytics, the income is higher, because you can immediately find out how much money the user will bring, whether he will bring it at all and whether it makes sense to return him — maybe he will return himself. This allows you to optimize costs and increase the profitability of campaigns.
05/ It is possible to accurately evaluate new user acquisition channels
When you launch ads on a new acquisition channel, you don’t have to spend months testing it to understand its CRR — the ratio of advertising spend to that ad revenue.
With predictive analytics, in a couple of days it is clear what the profitability of users in this channel is. If the results are not satisfactory, you can turn off the campaigns and not spend money on this source.
This principle can be applied to new user cohorts, new creatives, new campaigns. Predictive analysis helps to quickly evaluate their effectiveness and make optimization decisions.
Where predictive analytics will work effectively
Predictive marketing is suitable for almost all companies because the technology is universal. It is most suitable for businesses with frequent in-app purchases — there is a sufficient amount of audience on which the model can learn.
Predictive analysis might not be the best for:
- Small businesses, because there is not enough data for the system to learn.
- Businesses that don’t have in-app purchases or those who monetize through ads. For example, a free game that earns money by showing ads. The user does not have an opportunity to perform actions that will bring revenue to the app, therefore there will be nothing to predict.
- Subscription based businesses. There are few purchases that can be predicted. Most likely, the user will subscribe on the first day, and it will automatically renew.
- Businesses with a short payback cycle. With a short payback cycle, the advertiser spends money to attract a user, and already at the first order, the user brings this amount back. In this case, it makes no sense to use predictive analysis — it is already known when the user will pay off.
Maybe such launches will suit your business. You can leave a request at Go Predicts and we will tell you about the results that predictive campaigns can achieve in your particular case.
