Target users based on likelihood to convert (AI Targeting)
Contextual Bandit and AI Targeting is an add-on available in both Web Experimentation and Feature Experimentation.
For more details, please contact your Customer Success Manager.
Kameleoon AI targeting leverages in-house machine learning algorithms to generate propensity scores. These scores enable you to adapt and target your offers and content in real-time based on your visitors' buying behavior and intent.
Step 1: Define KPIs and triggers to track conversions and activate Kameleoon’s AI
Kameleoon's predictive setup consists of two parts: defining what you want to predict, and when you want the prediction to occur. Both are essential, as they provide flexibility to suit your specific use case.
Defining KPIs
Kameleoon allows you to define multiple KPIs and activate machine learning algorithms that automatically calculate, in real time, the likelihood of a visitor converting for each KPI.
Once your goal is defined, the only action required to use AI predictive targeting is enabling the "Use this goal for machine learning" option from your goal's Advanced settings.
In the Advanced Settings pop-in, you can also select when the AI propensity score for your goal should be calculated. Kameleoon lets you set up triggers for each moment.
Then click Save > Save to validate.
Defining triggers
A trigger represents a specific event that may occur during a visit. You will find the trigger builder in the Advanced Tools section, where you can define triggers relevant to your product's user experience.
More information on creating a trigger
Kameleoon’s AI can then calculate conversion propensity scores for your goal at any of these triggers, allowing you to use predictive targeting conditions when setting up personalization targeting. You can find more information on this here, here, and here.
Kameleoon Predict will then automatically rank each visitor based on their Likelihood to convert, from 0 to 100. A Likelihood to convert of 0 means the visitor has the lowest chance to convert compared to other visitors. A Likelihood to convert of 100 means that they have the highest chance of converting compared to other visitors.
More information on activating predictive targeting
When you create a goal in Kameleoon, for example, getting your visitors to add a product to their cart, you can decide whether you want to enable predictive targeting.
Once predictive targeting is activated for a goal, you can use it in an experiment or personalization once the following conditions are met:
- Our AI has learned from 100,000 visits;
- And our AI has learned from one full week of data.
For example, if your website receives 100,000 visits in a single day, the AI will still require one full week of data to complete the learning phase. Alternatively, if your site gets 50,000 visits per week, the AI will need two weeks to learn.
Once the learning phase is complete, the model is injected into the system on day 7 (D+7) and begins producing predictions. These predictions are used to build the KCS (Kameleoon Conversion Score) and become usable on day 8 (D+8), as one day of predictions is required before activation.
At that point, you’ll have the ability to create a segment targeting visitors within the specific probability range you define, enabling you to optimize your experiments and personalizations with precision.
These criteria do not apply to the triggers you create. To achieve optimal performance, ensure they target a reasonable portion of your audience and that your goals have a decent conversion rate within them.
2. Create a segment with the Likelihood to convert range you want to target
Create a segment using the Likelihood to convert criteria available in the segment builder. In the example below, we want to target visitors with the highest chance of converting, meaning we select the 70-100 range.
We could have also selected a range from the dropdown menu:
- Very low (0-20%)
- Low (20-40%)
- Moderate (40-60%)
- High (40-60%)
- Very high (80-100%)
Select the goal on which you activated Kameleoon's machine learning algorithms (Step 1) and the trigger.
Then, validate the creation of your segment.
Step 3: Create a personalization, a web experiment or a feature experiment
We can then use the segment in a web experiment, a personalization campaign or a feature experiment.