When working with a template in AI Lab → Templates, beyond defining properties and registering twins, you can further enrich each template with structured metadata that the mobile app and dashboards use to dynamically render KPI tiles, model activation controls, and AI recommendation cards.

These configurations are available as dedicated tabs inside any template: KPIs, Models, and Recommendations. Because they are defined at the template level, every twin registered against that template automatically inherits them. You configure once and it applies across all your assets.

Before continuing with the following article, make sure to check out Tutorial: Anomaly Detection Solution, as the examples mentioned below are a continuation of the main one detailed there.

KPIs

KPIs (Key Performance Indicators) let you define measurable targets against any template property. Each KPI compares the live value of a datapoint to a constant target using an operator, and the result is used by the mobile app to render health indicators per twin.
To add a KPI, navigate to the KPIs tab inside your template and click Edit → + Add KPI.

Each KPI has the following fields:

• Property: the template property whose value will be evaluated. Must be a datapoint type.
• Label: the display name shown in dashboards and the mobile app.
• Operator: the comparison logic: gt, gte, lt, lte, eq, or between.
• Target: the numeric threshold to compare against. For between, provide a [low, high] range.
• Unit: the unit displayed alongside the value.

Example: Bearing Anomaly Detection

In the my_example_factory_template, the pipeline writes a continuous anomaly score to the bearing_anomaly datapoint for each motor. We define a KPI to track whether that score stays within healthy bounds:

The Isolation Forest model outputs a score between 0 and 1. A value at or below 0.3 indicates normal vibration patterns; anything above signals a potential bearing fault. Using lte with a target of 0.3 means the KPI passes when the motor is healthy and fails when an anomaly is detected.

Since bearing_anomaly is a template property, this single KPI definition automatically evaluates against mt1_bearing_anomaly for Motor 1 and mt2_bearing_anomaly for Motor 2, with no per-twin configuration required.

Models

The Models tab lets you attach ML models to a template by referencing their model key. An optional Activation Property links the model to a template datapoint that acts as an on/off switch. When the value of that datapoint changes, it triggers or deactivates the model. If no activation property is set, the model runs continuously.

To attach a model, navigate to the Models tab and click Edit → + Attach Model.

Each entry has the following fields:

• Model Key: the key of the ML model stored in AI Lab → Models.
• Activation Property: a datapoint property from the template that operators can toggle to enable or disable the model. Leave as -- continuous -- if the model should always run.

Example: Bearing Anomaly Detection

The Isolation Forest model trained in the bearing_anomaly_pipeline is stored under the key bearing_anomaly_pipeline. We attach it to the template and link it to the bearing_model_active property, which was added to my_example_factory_template as a datapoint. This allows operators to enable or disable anomaly detection per motor directly from the mobile app by writing to that datapoint.

When a twin is evaluated, the platform resolves bearing_model_active to the actual datapoint registered for that specific motor, so toggling the model on Motor 1 has no effect on Motor 2.

Recommendations

Recommendations group related template properties together to give the mobile app a structured view of what the AI model is suggesting, what is currently measured, and what the target setpoint is. Each recommendation ties a measured value, a setpoint, and an AI-generated output together under a single named entry, optionally linked to the model that produces it.

To add a recommendation, navigate to the Recommendations tab and click Edit → + Add Recommendation.

Each entry has the following fields:

• Name: a human-readable label for this recommendation group, shown in the mobile app.
• Model Key (optional): the model responsible for generating the recommendation value.
• Measured: the template property representing the current real-world sensor reading.
• Setpoint: the template property representing the target or threshold value.
• Recommendation: the template property where the AI model writes its suggested output.
• Decimals: the number of decimal places used when displaying values.

Example: Bearing Anomaly Detection

For our bearing use case, the recommendation surfaces the live vibration reading alongside the configured threshold and the anomaly score produced by the model:

This tells the mobile app that for each motor, the model bearing_anomaly_pipeline is comparing the live Bearing1 vibration reading against the Threshold value defined per twin and writing its anomaly assessment to bearing_anomaly. The app can then render a recommendation card showing all three values side by side for each registered motor, without any hardcoding per asset.