Google Analytics will leverage its machine learning to suggest new and improved audiences for your App + Web marketing campaigns.
A pair of new metrics—on the likelihood of purchase and the likelihood of churn—could do wonders for both campaign targeting and analysis. If your web property generates the quantity of purchase activity needed to power them, the new Predictive audiences have impressive potential.
A Purchase Probability metric predicts how likely it is that a user who has visited your site or app will make a purchase in the next seven days.
Simply called ‘Likely 7-day purchasers’ this audience comprises the folks who’ve swung by your shop and shown signs of being ready to buy. Google suggests that ‘a well-crafted remarketing campaign can provide that last nudge they need to complete the process’.
Deeper, Bespoke Purchasing Probability
For businesses of a certain scale, the new Purchase Probability metric may functionally replace more simplistic indications, such as whether a user has put products into their cart.
While products-in-cart is almost certainly considered, Analytics also looks at ‘deep patterns of behaviour’ that may be completely unique to each property. The power of machine learning means that the factors Google deems important for predicting purchase probability on one site or app may be completely different for all other sites and apps.
If the system detects that a users is displaying waning interest in your business, they’ll be deemed ‘likely to churn’. This data point informs a particularly interesting audience set: ‘Likely 7-day churning purchaser’.
Google says this represents a market that's ripe for re-engagement. They suggest you reach out with ‘reminders of the value you offer in terms of product variety, quality, and price’ and promote competitive benefits like convenient shipping, return options, and special offers.
Eligibility: Settings & Site Activity
To set the scene for the new suggested audiences, you must make sure your Analytics account has the right options switched on and you’re getting enough site activity for the model to be reliable.
Websites must have benchmarking enabled in their data-sharing options, and need to collect purchase events data. Soon, being set-up to automatically measure in-app purchase events may replace (or maybe compliment) website purchase events data.
Traffic-wise, predictive audiences will only be generated for sites that achieve a certain scale of visitation. To be eligible, sites must have seen at least 1,000 users who trigger the relevant predictive condition, plus at least another 1,000 users who do not trigger it. And, that level of activity must be sustained: Both purchase and churn probabilities are trained on 28 days of data.
Will it Work?
The machine learning behind the new predicative audiences is all about modelling, objective numbers and subjective contexts. It’s not sentimental as it is based purely on quantifiable behaviours on your web property. The same metrics can also be flipped to better evaluate which campaigns achieve which outcomes.
None of this is guaranteed to boost your bottom line, but they are new tools that seem likely to streamline the processes for effective audience selection and campaign effectiveness evaluations. At the very least, that should save time and resources… if your already selling enough to qualify.
Copy Transmission is a Melbourne-based agency :: Better Brands. Loud & Clear.