Preparing your data sources

This reference article covers the critical feedback data assets needed to close the AI decisioning loop and enable your agent to continuously learn and improve.

Closing the AI decisioning loop

While all customer data is important for the agent (see Connect data sources), the most important data assets are those that tell the agent what happened after customer engagement decisions were sent.

These assets create the feedback loop that allows the agent to learn.

Critical feedback data assets

There are three critical assets for creating the feedback loop:

1. Conversions data
2. Engagement data
3. Activations data

Conversions data

The conversion asset describes what happened to the customer after orchestration. For example, supposing an agent is optimizing on Net Present Value (NPV) for customers receiving optimized campaigns, the conversion asset might include a daily update of changes to NPV.

Requirement	Reason
Each record contains a unique customer identifier that is consistent with all data assets	Decisioning Studio needs to track the individual customer journey from recommendation, through activation, to conversion.
Each record has an associated timestamp	Understanding the time between communication and sequence of customer actions is extremely important for agent training and metric calculation.
If using a non-binary (such as, converted versus unconverted) target metric, the target metric value is provided with each conversion event	Decisioning Studio uses the target metric value to generate training experiences to appropriately reward/penalize the agent based on the outcomes of the recommended actions.
If conversions can be uniquely attributed to communications (e.g., coupon redemption), fields needed to match conversions to activations are provided	If a conversion event can be tied to a particular communication, this allows for clean and precise attribution. Direct attribution provides the clearest signal to the agent, but if not possible (as is often the case), proximity-based attribution will be used.

Engagement data

The engagement asset describes customer interactions, including clicks, opens, and other impressions. Engagement data may be included in the conversion data or may be separate. It plays a similar role as conversions data—telling the agent what happened after customer engagement.

Requirement	Reason
Each record contains a unique customer identifier that is consistent with all data assets	Decisioning Studio needs to track engagement events for each individual customer.
Each record has an associated timestamp	Understanding the time between communication and sequence of customer actions is extremely important for agent training and metric calculation.
If clicks, opens, or other engagement data can be uniquely attributed to communications, fields needed to match engagement to activations are provided	As with conversion data, if engagement can be tied to a particular communication, this allows for clean and precise attribution. Direct attribution provides the clearest signal to the agent.

Activations data

The activations asset tells the agent which communications were sent. This is often necessary depending on how orchestration is configured. If the agent orchestrates via a direct integration with Braze, SFMC, or Klaviyo, then the agent may be able to pull activation data directly.

Requirement	Reason
Each record contains a unique customer identifier that is consistent with all data assets	Decisioning Studio needs to track the individual customer journey from recommendation, through activation, to conversion.
Each record has an associated timestamp	Understanding the time between communication and sequence of customer actions is extremely important for agent training and metric calculation.
Fields needed to match communication content to activation events are provided (such as event_id)	Correctly matching communication characteristics to sends is necessary for agent attribution and training.