Before plunging into this blog, I recommend you read the preceding chapter of our Predictive Remarketing series here. To give a prelude, we started off with our remarketing problem. The problem universal to all digital marketers is that you spend heaps of dollars, employ best digital marketing agencies to make creatives and still end up with not so satisfying ROIs. To tackle this problem, we introduced Tatvic’s Predictive Analytics model PredictN. It is an answer to every marketer’s remarketing woe:
“If I could remarket to only the audience who carry the highest probability of transacting on my site or submitting their details with help of Predictions; I could save on my Ad costs and get a higher conversion rate”
What’s PredictN?
Marketers understand the impact of Predictive Modelling on their critical business KPIs. But no good marketer runs any campaign blindfolded. We do need to understand how, a model classifies one visitor to convert on your website and the other to not convert based on Machine Learning. What type of patterns or customer profiles is enabling these predictions? How do you test the accuracy of this model’s audience segment? You are asking the right questions.
So let me decode the science behind these predictions by highlighting the thought process, data sources, model variables, process of implementation and performance monitoring.
Model Basis
PredictN Model is a supervised classification model. It is powered by Deep Learning algorithms with multiple layers of feature extraction and runs on Cloud services. It classifies a unique visitor id/ advertising id/ user id to successfully convert on your site or not. The predictors are a host of behavioral, conversion and custom business-specific attributes of website visitors.
Model Data Sources
PredictN Model feeds in analytics data from –
- Web Analytics Tools (Google Analytics, MixPanel, Omniture, etc)
- App Analytics Tools (Firebase, Localytics, Apsalar, etc)
- Data Collection & Query Tools (BigQuery, Segment, AWS, etc)
- Offline CRM/BI Tools
Model Engineering
PredictN Model inputs the past one or two months of your visitor data from the multiple data sources based on your business requirements. Say, we take Google Analytics 360 as the data source. We bucket your standard and custom dimensions data (e.g. session duration, visits, device category, geo, transactions, days since the last transaction, revenue, etc) and run series of feature engineering on it using PCA, error analysis, outliers removal, bucketing, under/oversampling and the like. Most of the feature engineering is done either from the data’s pattern or from a business point of view.
Model Process
Model decodes the periodic pattern of visitors who have earlier converted on your site and weighs the predictors which contribute significantly towards conversions. On this, it predicts who is your next set of visitors who will convert in the future period.
Model Output Connectors
Once PredictN Model results with a high AUC, we import the high probable converting audience list back to GA 360. And we export this segment to your remarketing platforms, say AdWords/DoubleClick/MailChimp. We recommend to run dynamic display campaigns to achieve better personalization and high CTR ratio. You may allocate 10% of your remarketing budget on a weekly basis for PredictN campaign. But, you should exclude this segment from all other campaigns to avoid cannibalization.
Model Performance Monitoring
You can track PredictN’s model performance on GA 360 itself or your remarketing platform dashboards on a weekly basis. You can compare campaign KPIs – conversion rate, revenue per transaction, Adcost per transaction, RoI/RoAS. No surprises here! PredictN Model will outrank most of your campaigns with a higher RoI and lesser adCost. In my previous blog chapter, I have shared one of the impressive campaign performance. Go ahead and have a look!
So what are you waiting for? Remarketing with a 1.5X RoI sounds impressive to me as a marketer. If you feel the same, please drop us a mail at marketing@tatvic.com. We are offering a limited number of customized demos every week. And post your comments on how predictions help in your business.
Related Webinar: Here’s a must-watch webinar on predictive remarketing using PredictN model.
The following two tabs change content below.
Bismayy Mohapatra
Bismayy is a Product Manager at Tatvic. He has developed Badger, an intelligent insights generation engine powered by Machine Learning. He is working with team to develop PredictN, an automated prediction SaaS platform for businesses to boost their marketing RoI. He applies R, Python, Cloud Prediction models to help clients make sense of their digital analytics data. He loves playing and following football.
Latest posts by Bismayy Mohapatra (see all)
- How Tatvic’s PredictN Solution Can Help You Increase Your Remarketing ROI - July 19, 2017
- The Science Behind App Uninstalls Prediction – Part 1/2 - June 15, 2017
