The BlueConic Next Best Action (NBA) capability allows you to use Agentic AI models to decide in real time which Dialogue variant or action to show to each visitor. Instead of relying on manual A/B tests or static rotation rules, models adapt dynamically based on profile data and conversion outcomes. This leads to higher engagement, improved retention, and increased revenue.
Models can be trained directly inside AI Workbench or externally in your own data science environment, then uploaded into BlueConic. Once stored in BlueConic, models can be used to power Dialogue optimization strategies.
Dialogue Models
Dialogue Models require
profileandvariant_idsas inputs.Requires
variant_idas output but also allow additional outputs.The
variant_idsoutput contains all the valid variant IDs the model can choose from.The
profileinput contains a feature vector representing the profile, profile vectorization is determined by theprofilePropertyIdsandfeatureNamesparameters of the model.The
variant_idoutput should contain the variant ID that is chosen by the model.Any additional outputs will be stored in the
modelOutputproperty of theviewtimeline eventThe values of the profile properties in the
profilePropertyIdsparameter will be stored in themodelInputproperty of theviewtimeline event.
Before you begin
Ensure you have the right user permissions to edit and view AI Workbench.
Confirm access to Gen AI features in your BlueConic environment.
Add a Dialogue Model
Log into your BlueConic environment.
Select More > AI Workbench.
In the interface, select Models.
Click Add model.
Enter the metadata, labels and description.
From Model type, select Dialogue. Choose how the model will be used, such as a Dialogue variant.
In Model file, upload the ONNX file that contains your AI model.
Select one or more Profile properties that the model should learn from. Once selected, click ok.
In Feature names, define the input features used by the model to make predictions. For example; browsername=chrome
Click Save.
Select a Model for a Dialogue
When a profile visits the website, the selected Model will be executed to predict which variant will perform the best, and the winning variant will be shown.
Via the BlueConic navigation bar, select Dialogues.
Add, search or select a Dialogue.
Select Optimization within the Dialogues settings.
Navigate to the Variant strategy section.
Choose Model-based optimization and select or create a model.
Switch the Dialogue to ON and select Save.
Note: If a Dialogue has multiple variants, you can select “Model-based optimization” as a strategy before/after evaluation.
Next steps
Periodically retrain your models in AI Workbench or upload updated ONNX models from your data science environment to keep up with shifting visitor behavior.
FAQ
What is the role of models in the BlueConic Next Best Action (NBA)?
Models determine the best action or Dialogue variant to serve each visitor in real time. They use profile data (like churn scores, engagement signals, or lifecycle stage) plus conversion outcomes to continuously learn which actions drive the most value.
What is the difference between an Agentic AI model and standard propensity models?
A standard model gives you a static score that predicts a single outcome. An agentic AI model uses reinforcement learning. It doesn't just score propensity; it actively learns by balancing exploiting actions known to work with exploring new ones to find better outcomes, ensuring it adapts to changing subscriber behavior.
Where can I train models?
Models can be trained inside the AI Workbench - the BlueConic built-in data science environment for building and training models on our own infrastructure.
How does the model learn?
Models learn from click or conversion signal timeline events that fire when an action is completed (e.g., newsletter signup, app install, video view). Each event updates the model’s understanding of which actions are most effective.
The
viewtimeline event has been extended withmodelInputandmodelOutputproperties that can be used to train a new model.For the
conversionevent you can now also use thevalueproperty of theconversionevent.
What if my model is new or untrained?
If you create or upload a new model without training data, BlueConic shows a warning: “Model untrained – default rotation in use.” In this case, dialogues fall back to automatic optimization until the model has learned enough.
How do I handle performance and speed?
Models are stored in memory and in our performance tests the execution of a model takes less than a millisecond, ensuring no visitor delays.
Guardrails (eligibility and suppression rules) prevent actions from firing in the wrong context.
Can I retrain or update models?
Yes. Retrain models inside AI Workbench as needed. Re-upload updated ONNX models if training externally. BlueConic automatically begins using the latest trained version.