Blueshift Predictive Scores empower you to focus more on creating engaging user experiences and less on rote spreadsheet analysis. This is done by applying and automating the best of the breed AI models to identify the most likely users to reach your custom marketing goals, all without any coding or scripting on your part. This is an advanced Blueshift capability, so contact your Blueshift Success Manager if you wish to turn this on for your account.
Marketing campaigns aimed at driving specific KPIs often involve targeting users who are likely to perform a specific action. For example, users who are likely to upgrade to a paid subscription, purchase a product, submit a form or download specific content. These desired actions are typically preceded by a flurry of activity indicating level of interest by each user in reaching that desired goal. Harnessing the latest in AI techniques, Blueshift can model sum total of user behavior, attributes and catalog interactions to predict likelihood of each individual user to perform specific actions and bucket them into precise segments. Using these precise segments, you can send very targeted and personalized messaging to nudge them towards specific parts of the site or apps that are expected to yield the highest follow through or conversion.
Here are a few sample predictive scores for your reference. You can can use the predictive studio to create these based on the individual account data and desired business outcomes.
| Predictive Score Type | Description | Use cases |
| Conversion | Conversion predictive scores model the likelihood of each individual user to reach a specific state in a funnel. | Travel, E-Commerce, Auction Marketplaces sites usually have a series of steps leading to a booking/checkout and purchase funnel. Finance, RealEstate and Job boards typically have series of steps leading to submit a lead/form and applying for a job. Media, Education and Entertainment sites typically have a free-mium model for users to move from free to a paid subscription model. |
| Engagement | Engagement predictive scores model the likelihood of each individual user to come back to the site/apps within a time window, typically tracked as part of the DAUs/MAUs (Daily Active Users/Monthly Active Users) | Users repeatedly coming back to the site/apps/store is a strong indicator of user engagement with a brand. Modeling that likelihood allows you to take proactive steps to improve engagement and reduce inactive user base. |
| Churn | Churn predictive scores model the likelihood of each individual user to cancel or downgrade a subscription. | Subscription based commerce models. |
| Retention | Retention predictive scores model the likelihood of each individual user to repeat a specific action/goal. While this correlates highly with engagement, it's more precise. It aims to compute repeat purchase or transaction likelihood towards a specific goal event, rather than just a repeat visit. | Repeat transactions within a time window are strong indicators of strong retention and high LTV (Lifetime Value). |
| Data type | Description | Sample data |
| User Attributes | User attributes like location, device type, category or brand or catalog attribute preferences | Favorite author, City/State |
| Events | Aggregates on past behaviors | lifetime_orders, lifetime_visits, last_7d_visits, most_recent_visit_at,... |
| Goals | Custom goals | 'submitted lead', 'purchase', 'booked a room', 'upgraded to paid subscription',... |
- Customizable Inputs and goals - You can select what events, user attributes and goals go into the models.
- Automated feature engineering - Blueshift's fully automated feature engineering takes away the grunt work of cleaning, de-duping and normalizing raw data to compute a rich set of features that feed into learning models. We also automatically remove noise features that do not contribute to the accuracy of the model.
- Automated Inference of training data - Blueshift's automated inference of training data from past user behaviors takes away the manual task of curating and labeling training data to feed the learning models. We do this by advanced sampling techniques over past user behaviors to select sufficient positive and negative examples for the desired outcome.
- Automated model building and validation - Blueshift's automated model building and validation uses state of the art learning models to build an ensemble of models that generalize well without overfitting.
- Easy to use outputs - The output score for each user is converted into a percentile rank between 0 and 100, allowing you to pick subsets of the range that best suits your campaign needs.
- Rich visualization - The model visualization, showing how the goal completion rates vary by each percentile, is built into the dashboard along with the relative importance of each feature.