Skip to main content
AI Predictive Targeting is an add-on available in both Web Experimentation and Feature Experimentation. Contact your Customer Success Manager for details.
When you target all visitors the same way, you treat someone who is very likely to convert the same as someone who is unlikely to, which makes it harder to decide when to intervene, what message to show, and how strong an incentive (if any) to offer. AI Predictive Targeting solves this by training a machine learning model that estimates each visitor’s likelihood of converting for a specific goal. Once training completes, you can use the Likelihood to convert condition in segments to target visitors by their predicted conversion probability and tailor your campaign accordingly.

​
How the model works

Kameleoon’s machine learning model estimates each visitor’s expected conversion probability from three categories of data points. The examples in each category are not exhaustive: the model takes into account most of the data points Kameleoon collects. For the full inventory, see List of collected data.

​
Previous sessions

Data points from the visitor’s past sessions, including:
  • Number of sessions
  • Average time between sessions
  • Number of visited pages
  • Number of clicks

​
Static data points

Data points that describe the visitor’s context, including:
  • Browser
  • Operating system
  • Landing page
  • Referrer
  • Time since last session

​
Dynamic data points

Data points that Kameleoon derives from the visitor’s behavior during the current session, including:
  • Number of pages visited
  • Number of clicks
  • Time spent on the website

​
How often the model and scores update

The model is not static. Kameleoon retrains it every 10 minutes as it collects new data points and analyzes more visits and conversions. Kameleoon also rebuilds the engine.js application file every hour to deploy the latest model, unless one of your own changes (for example, updating a campaign) has already triggered a rebuild. Each visitor’s score updates in real time: Kameleoon recalculates it at every measured trigger, such as a step in the customer journey.

​
Use with Feature Experimentation SDKs

When you retrieve visitor data through a Feature Experimentation SDK, set kcs to true in your getRemoteVisitorData call. Without this flag, Kameleoon does not return the Kameleoon Conversion Score (KCS), and the Likelihood to convert condition has no score to evaluate. For the full parameter reference, see getRemoteVisitorData in the SDK documentation. The kcs parameter is available in all Kameleoon SDKs.

​
Enrich the model with your own data

You can improve prediction accuracy by supplying your own business data. When you create a custom data point, enable Use this custom data as input for AI Predictive Targeting to make it available to the model.
The model only accepts Number and Boolean custom data types. Avoid adding sensitive data such as dates, timestamps, or unique visitor identifiers (for example, email addresses or account numbers), as these values are unsuitable for predictive models. Also avoid custom data that is highly targeted, for example, data specific to individual visitors or very small segments.
For setup instructions, see Create custom data.

​
Set up AI Predictive Targeting

1

Enable AI Predictive Targeting on a goal

Choose a conversion goal and activate AI Predictive Targeting to start the model’s learning phase. Select a trigger that represents a meaningful moment in your customer journey, such as a product page view, cart addition, or checkout abandonment. Enable AI Predictive Targeting on a goal
2

Wait for the learning phase to complete

The model requires 7 days of data and 100,000 visits, and both conditions must be met. Once training completes, the goal badge changes from AI Learning to AI Active, and predictions become available the following day. To learn what each badge means, see AI model status badges.
3

Create a segment with Likelihood to convert

Use the Likelihood to convert condition to create a segment that targets visitors by their predicted conversion probability. Kameleoon groups visitors into five score buckets: Very low, Low, Moderate, High, and Very high. Use AI Predictive Targeting in a segmentYou can also apply the Likelihood to convert condition directly to an experiment through quick targeting, without creating a reusable segment.
4

Apply the segment to a campaign

Add the segment to a web experiment, personalization, or feature experiment to deliver the right experience to each visitor based on their propensity to convert.