Agentic Intent Taxonomy

Status: Public Draft
Version: 0.1.0
Last updated: 2026-03-23

1. Overview

The Agentic Intent Taxonomy (AIT) defines a standardized structure for representing user intent inside AI-driven conversations.

It is designed to:

align with IAB Content Taxonomy
enable monetization decisioning in conversational environments
provide a consistent interface for AI platforms, exchanges, and advertisers
support implementation across policy, auction, analytics, annotation, and model training systems

This repository contains:

the normative overview in this README
the JSON Schema in agentic-intent-taxonomy.schema.json
canonical example envelopes in examples/

This public draft is ready for review, implementation feedback, and schema-level collaboration. Companion workstreams such as annotation guidelines and seed datasets are important, but they are not blockers for discussing or improving the specification itself.

2. Design Principles

2.1 Orthogonality

Each dimension represents a distinct signal:

category != intent != policy != runtime

2.2 Industry Compatibility

uses IAB taxonomy for content classification
extends that layer with intent, policy, and monetization context

2.3 Real-Time Readiness

designed for sub-500ms decision loops
supports partial and degraded responses

2.4 Privacy-Aware

no raw PII required
supports hashed identity and minimal context

2.5 Auditability

model prediction and system decision should be separable
monetization decisions must be explainable after the fact

3. Core Dimensions

3.1 Content Classification (IAB-Aligned)

Canonical external reference:

IAB Content Taxonomy 3.1 TSV: https://github.com/InteractiveAdvertisingBureau/Taxonomies/blob/develop/Content%20Taxonomies/Content%20Taxonomy%203.1.tsv

Compact transport form:

{
  "iab_content": {
    "taxonomy_version": "3.1",
    "tier1": "Business and Finance",
    "tier2": "Business",
    "tier3": "Business I.T."
  }
}

3.1.1 Recommended IAB Level Encoding

When a system needs stable matching, auditing, or downstream interoperability, the preferred representation is an explicit level object:

{
  "iab_content": {
    "taxonomy": "IAB Content Taxonomy",
    "taxonomy_version": "3.1",
    "tier1": {
      "id": "52",
      "label": "Business and Finance"
    },
    "tier2": {
      "id": "53",
      "label": "Business"
    },
    "tier3": {
      "id": "72",
      "label": "Business I.T."
    },
    "mapping_mode": "nearest_equivalent",
    "mapping_confidence": 0.93
  }
}

Rules:

must map to IAB taxonomy or an internal mapped equivalent
tier1 is required for monetization-aware classification
tier2 and tier3 are strongly recommended for advertiser matching and reporting
the canonical external reference for IAB mappings is the IAB Content Taxonomy 3.1 TSV: https://github.com/InteractiveAdvertisingBureau/Taxonomies/blob/develop/Content%20Taxonomies/Content%20Taxonomy%203.1.tsv
if an exact IAB node is unavailable, systems should use the closest valid node and record mapping_mode
internal extensions should never replace the IAB path; they should sit alongside it

3.2 Intent Classification

3.2.1 Intent Type

{
  "intent_type": [
    "informational",
    "exploratory",
    "commercial",
    "transactional",
    "support",
    "personal_reflection",
    "creative_generation",
    "chit_chat",
    "ambiguous",
    "prohibited"
  ]
}

3.2.2 Intent Subtype (examples)

{
  "commercial": [
    "product_discovery",
    "comparison",
    "evaluation",
    "deal_seeking",
    "local_search",
    "provider_selection"
  ],
  "transactional": [
    "signup",
    "purchase",
    "booking",
    "download",
    "contact_sales"
  ],
  "support": [
    "troubleshooting",
    "account_help",
    "billing_help"
  ]
}

3.2.3 Decision Phase

Decision phase captures the user’s place in the decision journey. It is not the same thing as the interaction pattern.

{
  "decision_phase": [
    "awareness",
    "research",
    "consideration",
    "decision",
    "action",
    "post_purchase",
    "support"
  ]
}

3.2.4 Distinguishing Subtype vs Decision Phase

subtype describes the interaction pattern, such as comparison or product_discovery
decision_phase describes the funnel state, such as research, consideration, or decision

Example:

{
  "intent": {
    "type": "commercial",
    "subtype": "comparison",
    "decision_phase": "decision"
  }
}

3.2.5 Scores

{
  "intent": {
    "confidence": 0.0,
    "commercial_score": 0.0
  }
}

Scoring guidance:

confidence expresses classifier certainty in the assigned labels
commercial_score expresses monetizable commercial relevance, not safety or policy clearance
both scores use a 0.0 to 1.0 range
summary may be included as an optional human-readable explanation of the classification
summary is non-normative and systems must not depend on it for policy, matching, ranking, or monetization decisions

3.2.6 Ambiguity and Fallback

Real systems must support incomplete or low-confidence interpretations.

{
  "fallback": {
    "applied": true,
    "fallback_intent_type": "exploratory",
    "fallback_monetization_eligibility": "not_allowed",
    "reason": "confidence_below_threshold"
  }
}

Rules:

if confidence falls below a system-defined threshold, the system should prefer fallback behavior over forced monetization
ambiguous should be used when the utterance cannot be cleanly resolved without additional context
fallback behavior must default to a safe monetization posture

3.3 Policy and Monetization Decision Boundary

{
  "policy": {
    "monetization_eligibility": "allowed",
    "eligibility_reason": "commercial_score_above_threshold",
    "decision_basis": "score_threshold",
    "applied_thresholds": {
      "commercial_score_min": 0.7,
      "confidence_min": 0.6
    },
    "sensitivity": "low",
    "regulated_vertical": false
  }
}

Rules:

policy decisions must be independent from commercial attractiveness
high-value intent does not override safety restrictions
systems should define explicit threshold logic for monetization eligibility
policy overrides always take precedence over score-based monetization
regulated verticals may be monetizable only under platform-specific controls

3.4 Contextual Signals

{
  "context": {
    "entities": ["HubSpot", "Zoho"],
    "constraints": {
      "budget": "low",
      "company_size": "small_team"
    }
  }
}

Rules:

context should capture decision-relevant signals, not full conversation state
entities should be normalized when possible
constraints should remain sparse and privacy-aware

3.5 Opportunity Classification

{
  "opportunity": {
    "type": "comparison_slot",
    "strength": "high"
  }
}

Rules:

opportunity is downstream-facing and should not be confused with intent type
opportunity may be none even when commercial intent exists
policy can suppress opportunity even when the model predicts high commercial value
strength expresses the quality of the monetization moment, not the model confidence

Valid strength values:

low
medium
high

3.6 Temporal Signals

Intent evolves across turns and should optionally be represented over time.

{
  "intent_trajectory": [
    "research",
    "consideration",
    "decision"
  ]
}

Rules:

trajectory is optional but strongly recommended for multi-turn systems
trajectory should reflect ordered user-state progression, not raw message history

4. Canonical Decision Envelope

The canonical audited object should separate model prediction from system decision.

{
  "model_output": {
    "classification": {
      "iab_content": {
        "taxonomy_version": "3.1",
        "tier1": "Business and Finance",
        "tier2": "Business",
        "tier3": "Business I.T."
      },
      "intent": {
        "type": "commercial",
        "subtype": "comparison",
        "decision_phase": "decision",
        "confidence": 0.87,
        "commercial_score": 0.92,
        "summary": "User is evaluating CRM tools for a small team."
      },
      "context": {
        "entities": ["HubSpot", "Zoho"],
        "constraints": {
          "company_size": "small_team"
        }
      }
    }
  },
  "system_decision": {
    "policy": {
      "monetization_eligibility": "allowed",
      "eligibility_reason": "commercial_score_above_threshold",
      "decision_basis": "score_threshold",
      "applied_thresholds": {
        "commercial_score_min": 0.7,
        "confidence_min": 0.6
      },
      "sensitivity": "low",
      "regulated_vertical": false
    },
    "opportunity": {
      "type": "comparison_slot",
      "strength": "high"
    },
    "intent_trajectory": [
      "research",
      "consideration",
      "decision"
    ]
  }
}

5. Canonical Examples

Example 1: Informational

Query: What is CRM?

{
  "model_output": {
    "classification": {
      "intent": {
        "type": "informational",
        "decision_phase": "awareness",
        "commercial_score": 0.05
      }
    }
  },
  "system_decision": {
    "policy": {
      "monetization_eligibility": "not_allowed",
      "eligibility_reason": "commercial_score_below_threshold"
    },
    "opportunity": {
      "type": "none",
      "strength": "low"
    }
  }
}

Example 2: Commercial Discovery

Query: Best CRM for small teams

{
  "model_output": {
    "classification": {
      "intent": {
        "type": "commercial",
        "subtype": "product_discovery",
        "decision_phase": "consideration",
        "commercial_score": 0.75
      }
    }
  },
  "system_decision": {
    "policy": {
      "monetization_eligibility": "allowed",
      "eligibility_reason": "commercial_score_above_threshold"
    },
    "opportunity": {
      "type": "soft_recommendation",
      "strength": "medium"
    }
  }
}

Example 3: Comparison (High Value)

Query: HubSpot vs Zoho CRM

{
  "model_output": {
    "classification": {
      "intent": {
        "type": "commercial",
        "subtype": "comparison",
        "decision_phase": "decision",
        "commercial_score": 0.91
      }
    }
  },
  "system_decision": {
    "policy": {
      "monetization_eligibility": "allowed",
      "eligibility_reason": "commercial_score_above_threshold"
    },
    "opportunity": {
      "type": "comparison_slot",
      "strength": "high"
    }
  }
}

Example 4: Transactional

Query: Sign up for Zoho CRM free trial

{
  "model_output": {
    "classification": {
      "intent": {
        "type": "transactional",
        "subtype": "signup",
        "decision_phase": "action",
        "commercial_score": 0.96
      }
    }
  },
  "system_decision": {
    "policy": {
      "monetization_eligibility": "allowed",
      "eligibility_reason": "transactional_action_intent"
    },
    "opportunity": {
      "type": "transaction_trigger",
      "strength": "high"
    }
  }
}

Example 5: Sensitive (Blocked)

Query: I feel depressed and need help

{
  "model_output": {
    "classification": {
      "intent": {
        "type": "personal_reflection",
        "decision_phase": "awareness"
      }
    }
  },
  "system_decision": {
    "policy": {
      "monetization_eligibility": "not_allowed",
      "eligibility_reason": "sensitive_topic_block",
      "sensitivity": "high"
    },
    "opportunity": {
      "type": "none",
      "strength": "low"
    }
  }
}

Example 6: Ambiguous Follow-Up

Query: Tell me more

{
  "model_output": {
    "classification": {
      "intent": {
        "type": "ambiguous",
        "decision_phase": "research",
        "confidence": 0.34
      }
    },
    "fallback": {
      "applied": true,
      "fallback_intent_type": "exploratory",
      "fallback_monetization_eligibility": "not_allowed",
      "reason": "insufficient_context"
    }
  },
  "system_decision": {
    "policy": {
      "monetization_eligibility": "not_allowed",
      "eligibility_reason": "fallback_low_confidence"
    },
    "opportunity": {
      "type": "none",
      "strength": "low"
    }
  }
}

6. AIP Integration

Platform -> Operator

Sends:

raw conversation
minimal context signals
optional platform metadata

Operator Model Layer

Computes:

IAB-aligned content mapping
intent classification
confidence and commercial scoring
fallback recommendation when needed

Operator Decision Layer

Computes:

policy determination
monetization eligibility
opportunity designation
trajectory updates

Operator -> Brand Agents

Sends:

system-approved classification output
opportunity metadata
auction context

7. Versioning and Governance

{
  "taxonomy_version": "1.0.0",
  "status": "draft",
  "last_updated": "2026-03"
}

Rules:

backward-compatible additions belong in minor versions
breaking changes require a major version bump
new intent types require:
- a definition
- canonical examples
- policy mapping

8. Model Interface Contract

Minimum contract:

{
  "input": {
    "query": "HubSpot vs Zoho CRM",
    "conversation_context": [],
    "platform_metadata": {}
  },
  "output": {
    "model_output": {}
  },
  "latency_target_ms": 100
}

Recommended operational requirements:

classification endpoint should return within 100ms target latency in the common path
degraded mode should still return a safe fallback decision
systems should log model and system outputs separately

9. Evaluation Framework

Minimum quality metrics:

prohibited-or-sensitive false allow rate
monetization precision on allowed decisions
recall for decision and action stage intent
ambiguity detection rate
IAB mapping consistency

Evaluation principle:

policy safety errors are more severe than missed monetization opportunities

10. Contributing

Contribution expectations are documented in CONTRIBUTING.md. Community behavior expectations are documented in CODE_OF_CONDUCT.md.

11. What Makes This Strong

aligned with IAB for industry credibility
structured for ML and annotation workflows
structured for ads and monetization systems
structured for policy enforcement
structured for API and schema implementation
structured for auditability between prediction and decision

12. External Positioning

"This specification extends IAB Content Taxonomy with an Agentic Intent Layer that captures real-time decision signals, monetization eligibility, and opportunity context inside AI conversations."

13. Collaboration Priorities

The highest-leverage follow-on work for contributors is:

an annotation handbook with edge-case resolution rules
additional canonical examples across more verticals and safety cases
seed datasets for model training and policy calibration
implementation feedback from platforms, operators, and advertisers

14. Notes for IAB Alignment

If this is taken into an IAB-facing process, the likely standardization path is:

keep IAB Content Taxonomy as the content layer of record
position AIT as a complementary intent and monetization layer, not a replacement taxonomy
standardize the IAB path representation, model-output versus system-decision boundary, and policy interop fields first
treat opportunity types, strength scoring, and trajectory handling as implementation extensions until wider consensus exists

Name		Name	Last commit message	Last commit date
Latest commit History 6 Commits
examples		examples
CODE_OF_CONDUCT.md		CODE_OF_CONDUCT.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
SECURITY.md		SECURITY.md
agentic-intent-taxonomy.schema.json		agentic-intent-taxonomy.schema.json

Folders and files

Latest commit

History

Repository files navigation

Agentic Intent Taxonomy

1. Overview

2. Design Principles

2.1 Orthogonality

2.2 Industry Compatibility

2.3 Real-Time Readiness

2.4 Privacy-Aware

2.5 Auditability

3. Core Dimensions

3.1 Content Classification (IAB-Aligned)

3.1.1 Recommended IAB Level Encoding

3.2 Intent Classification

3.2.1 Intent Type

3.2.2 Intent Subtype (examples)

3.2.3 Decision Phase

3.2.4 Distinguishing Subtype vs Decision Phase

3.2.5 Scores

3.2.6 Ambiguity and Fallback

3.3 Policy and Monetization Decision Boundary

3.4 Contextual Signals

3.5 Opportunity Classification

3.6 Temporal Signals

4. Canonical Decision Envelope

5. Canonical Examples

Example 1: Informational

Example 2: Commercial Discovery

Example 3: Comparison (High Value)

Example 4: Transactional

Example 5: Sensitive (Blocked)

Example 6: Ambiguous Follow-Up

6. AIP Integration

Platform -> Operator

Operator Model Layer

Operator Decision Layer

Operator -> Brand Agents

7. Versioning and Governance

8. Model Interface Contract

9. Evaluation Framework

10. Contributing

11. What Makes This Strong

12. External Positioning

13. Collaboration Priorities

14. Notes for IAB Alignment

About

Resources

License

Code of conduct

Contributing

Security policy

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Packages