Behavioral Intelligence (L3)

The Intelligence Layer transforms the raw interaction history captured by the Behavioral Graph into actionable recommendations. Before your agent decides what to do, it can ask Fusemomo: "Based on everything we know, what actions are most likely to work?"

NOTE

The Intelligence Layer requires a Builder or Enterprise plan. Free-tier accounts have full access to L1 and L2.

The Recommendation Response

{
  "recommendation_id": "rec_01A2B3C",
  "entity_id": "ent_550e8400...",
  "intent": "resolve_incident",
  "data_sufficient": true,
  "confidence_score": 0.87,
  "primary": {
    "api": "pagerduty",
    "action_type": "escalate_to_oncall",
    "raw_success_rate": 0.87,
    "success_count": 13,
    "total_count": 15,
    "composite_score": 0.87,
    "is_primary": true
  },
  "opportunity_set": [
    {
      "api": "pagerduty",
      "action_type": "trigger_alert",
      "composite_score": 0.94,
      "success_count": 21,
      "total_count": 23,
      "is_primary": true
    },
    ...
  ]
}

Key Fields

Field	Description
data_sufficient	false if there is not enough historical data to make a confident recommendation
confidence_score	Composite score of the primary recommendation (0.0–1.0)
primary	The single best recommended action
opportunity_set	Full ranked list of all qualifying action types, from highest to lowest score

Intent Parameter

Passing an intent scope constrains the recommendation to interactions where that intent was recorded. This prevents cross-contamination between different types of agent workflows.

For example, an entity may have a different optimal action for support_escalation vs outreach using intent keeps them separate.

Closing the Feedback Loop

The recommendation system improves through an explicit feedback mechanism. After you act on a recommendation, call the Feedback API with the recommendation_id:

{
  "was_followed": true,
  "outcome_interaction_id": "int_3A21F5..."
}

This tells the system:

• Whether the agent followed the advice
• What the actual outcome was (via linking to the interaction)

Consistently closing the loop is what enables the recommendation quality to improve over time.

When data_sufficient is false

A recommendation can be returned even when data_sufficient: false. This means the system found some data but it may not be statistically representative. Treat these as directional hints rather than high-confidence recommendations.

Common causes:

• The entity is new (few recorded interactions)
• The lookback window contains very few relevant interactions
• No interactions have been recorded for the specified intent

Configuring the Recommendation

Parameter	Default	Description
lookback_days	90	How far back to look in interaction history
min_success_count	1	Minimum number of successes for an action type to be eligible
intent	(none)	Scope recommendations to a specific intent
agent_id	(none)	Optionally scope to recommendations by a specific agent