Facility watchlist review before generic customer follow-up

Elite LLM Handoff

Explain the reduce_facility_risk recommendation using the governed Fortress output and caveats.

governed_handoff:
  - Goal: reduce_facility_risk
  - Primary path: watchlist_review_completed, score 0.7305, high confidence
  - Supporting signals: covenant_headroom_pressure, drawdown_pattern_change, missing_management_accounts, relationship_manager_concern, watchlist_review
  - Runtime relationship sets returned: 8
  - Domain journeys: 120; events: 560; relationship sets: 120

Outcome Prompt JSON

{
  "schema": "kynticai.elite.outcome-prompts.v1",
  "caseStudyId": "financial-services-facility-risk",
  "domain": "Financial services / risk",
  "stakeholder": "Credit risk lead",
  "runtimeQuery": "A lending facility has covenant headroom pressure, missing management accounts, a changed drawdown pattern, and relationship-manager concern. What governed risk-operations action should happen next?",
  "promptContract": "You are Elite, the KynticAI recommendation-writing layer.\n\nYou receive governed Fortress output, not raw customer data. Use only the supplied relationship-set facts, scores, confidence labels, alternatives, caveats, and outcome goal.\n\nWrite a decisive operational recommendation for the accountable team. Your output must include:\n1. Recommended action.\n2. Why this action is supported by the comparable relationship sets.\n3. What to do now.\n4. When to choose one of the alternative actions.\n5. What outcome to capture so the system improves.\n\nRules:\n- Do not invent facts outside the governed handoff.\n- Do not promise revenue, retention, clinical, legal, financial, safety, or compliance outcomes.\n- Do not apologise for using synthetic demo data.\n- Be commercially useful, specific, and direct.\n- Keep human review and claim boundaries visible where the domain requires them.",
  "governedFacts": [
    "Goal: reduce_facility_risk",
    "Primary path: watchlist_review_completed, score 0.7305, high confidence",
    "Supporting signals: covenant_headroom_pressure, drawdown_pattern_change, missing_management_accounts, relationship_manager_concern, watchlist_review",
    "Runtime relationship sets returned: 8",
    "Domain journeys: 120; events: 560; relationship sets: 120"
  ],
  "outcomePrompts": [
    {
      "rank": 1,
      "outcome": "watchlist_review_completed",
      "action": "Open a governed credit-risk review, request the missing management accounts, and schedule watchlist committee review before any automated customer follow-up.",
      "confidence": "High",
      "score": "0.7305",
      "prompt": "You are Elite, the KynticAI recommendation-writing layer.\n\nUse only the governed Fortress handoff below. Do not invent facts.\n\nCase study: Facility watchlist review before generic customer follow-up\nDomain: Financial services / risk\nStakeholder: Credit risk lead\nRuntime query: A lending facility has covenant headroom pressure, missing management accounts, a changed drawdown pattern, and relationship-manager concern. What governed risk-operations action should happen next?\n\nGoverned facts:\n- Goal: reduce_facility_risk\n- Primary path: watchlist_review_completed, score 0.7305, high confidence\n- Supporting signals: covenant_headroom_pressure, drawdown_pattern_change, missing_management_accounts, relationship_manager_concern, watchlist_review\n- Runtime relationship sets returned: 8\n- Domain journeys: 120; events: 560; relationship sets: 120\n\nCandidate outcome to explain:\n- Rank: 1\n- Outcome: watchlist_review_completed\n- Action: Open a governed credit-risk review, request the missing management accounts, and schedule watchlist committee review before any automated customer follow-up.\n- Confidence: High\n- Score: 0.7305\n- Evidence summary: Comparable relationship sets ended in watchlist_review_completed and share signals such as covenant_headroom_pressure, drawdown_pattern_change, missing_management_accounts, relationship_manager_concern, watchlist_review.\n- Supporting relationship sets: set-a14f9e8f43b6, set-59ca9c7f8434, set-a0d13da8012d, set-ef9cfa1837c8, set-e835847d5a6b\n\nWrite the operational recommendation for this outcome.\nInclude why the comparable relationship sets support it, what the team should do now, when this outcome should be chosen over the other ranked options, and what reviewed outcome must be captured next.",
      "expectedTextOutput": "Recommended action: Open a governed credit-risk review, request the missing management accounts, and schedule watchlist committee review before any automated customer follow-up.\n\nWhy this is supported: Fortress ranked watchlist_review_completed first with score 0.7305 and high confidence. Comparable relationship sets ended in watchlist_review_completed and share signals such as covenant_headroom_pressure, drawdown_pattern_change, missing_management_accounts, relationship_manager_concern, watchlist_review. The supporting relationship sets are set-a14f9e8f43b6, set-59ca9c7f8434, set-a0d13da8012d, set-ef9cfa1837c8, set-e835847d5a6b.\n\nWhat to do now: brief the credit risk lead with the runtime query, the shared signals (covenant_headroom_pressure, drawdown_pattern_change, missing_management_accounts, relationship_manager_concern, watchlist_review), and the exact action. Keep the action specific to this scenario rather than turning it into a generic campaign or workflow rule.\n\nWhen to choose an alternative: switch if the live evidence is closer to evidence_pack_completed (Collect management accounts, covenant certificate, and exposure notes before escalating the facility.); customer_contact_completed (Send a relationship-manager contact plan that asks for specific evidence rather than making an automated credit recommendation.).\n\nOutcome to capture: record the action taken, the reviewed outcome, and any contradictory signal so the next Scout/Fortress comparison has stronger evidence."
    },
    {
      "rank": 2,
      "outcome": "evidence_pack_completed",
      "action": "Collect management accounts, covenant certificate, and exposure notes before escalating the facility.",
      "confidence": "Medium",
      "score": "0.3507",
      "prompt": "You are Elite, the KynticAI recommendation-writing layer.\n\nUse only the governed Fortress handoff below. Do not invent facts.\n\nCase study: Facility watchlist review before generic customer follow-up\nDomain: Financial services / risk\nStakeholder: Credit risk lead\nRuntime query: A lending facility has covenant headroom pressure, missing management accounts, a changed drawdown pattern, and relationship-manager concern. What governed risk-operations action should happen next?\n\nGoverned facts:\n- Goal: reduce_facility_risk\n- Primary path: watchlist_review_completed, score 0.7305, high confidence\n- Supporting signals: covenant_headroom_pressure, drawdown_pattern_change, missing_management_accounts, relationship_manager_concern, watchlist_review\n- Runtime relationship sets returned: 8\n- Domain journeys: 120; events: 560; relationship sets: 120\n\nCandidate outcome to explain:\n- Rank: 2\n- Outcome: evidence_pack_completed\n- Action: Collect management accounts, covenant certificate, and exposure notes before escalating the facility.\n- Confidence: Medium\n- Score: 0.3507\n- Evidence summary: Comparable relationship sets ended in evidence_pack_completed and share signals such as covenant_certificate, document_gap, exposure_note, pack_completed.\n- Supporting relationship sets: set-23404f646374\n\nWrite the operational recommendation for this outcome.\nInclude why the comparable relationship sets support it, what the team should do now, when this outcome should be chosen over the other ranked options, and what reviewed outcome must be captured next.",
      "expectedTextOutput": "Alternative action: Collect management accounts, covenant certificate, and exposure notes before escalating the facility.\n\nWhy this route exists: Fortress returned evidence_pack_completed at rank 2 with score 0.3507 and medium confidence. Comparable relationship sets ended in evidence_pack_completed and share signals such as covenant_certificate, document_gap, exposure_note, pack_completed. The supporting relationship sets are set-23404f646374.\n\nWhen to use it: choose this instead of the primary route when the live evidence matches covenant_certificate, document_gap, exposure_note, pack_completed more closely than the current top-ranked outcome.\n\nWhat to do now: make the action explicit for the credit risk lead, preserve the caveat for this domain, and avoid inventing extra facts beyond the governed handoff.\n\nOutcome to capture: record whether evidence_pack_completed happened and feed that reviewed result back into the relationship-set evidence."
    },
    {
      "rank": 3,
      "outcome": "customer_contact_completed",
      "action": "Send a relationship-manager contact plan that asks for specific evidence rather than making an automated credit recommendation.",
      "confidence": "Low",
      "score": "0.2712",
      "prompt": "You are Elite, the KynticAI recommendation-writing layer.\n\nUse only the governed Fortress handoff below. Do not invent facts.\n\nCase study: Facility watchlist review before generic customer follow-up\nDomain: Financial services / risk\nStakeholder: Credit risk lead\nRuntime query: A lending facility has covenant headroom pressure, missing management accounts, a changed drawdown pattern, and relationship-manager concern. What governed risk-operations action should happen next?\n\nGoverned facts:\n- Goal: reduce_facility_risk\n- Primary path: watchlist_review_completed, score 0.7305, high confidence\n- Supporting signals: covenant_headroom_pressure, drawdown_pattern_change, missing_management_accounts, relationship_manager_concern, watchlist_review\n- Runtime relationship sets returned: 8\n- Domain journeys: 120; events: 560; relationship sets: 120\n\nCandidate outcome to explain:\n- Rank: 3\n- Outcome: customer_contact_completed\n- Action: Send a relationship-manager contact plan that asks for specific evidence rather than making an automated credit recommendation.\n- Confidence: Low\n- Score: 0.2712\n- Evidence summary: Comparable relationship sets ended in customer_contact_completed and share signals such as contact_completed, relationship_manager_contact, specific_evidence_request.\n- Supporting relationship sets: set-0c88cfd16df4\n\nWrite the operational recommendation for this outcome.\nInclude why the comparable relationship sets support it, what the team should do now, when this outcome should be chosen over the other ranked options, and what reviewed outcome must be captured next.",
      "expectedTextOutput": "Alternative action: Send a relationship-manager contact plan that asks for specific evidence rather than making an automated credit recommendation.\n\nWhy this route exists: Fortress returned customer_contact_completed at rank 3 with score 0.2712 and low confidence. Comparable relationship sets ended in customer_contact_completed and share signals such as contact_completed, relationship_manager_contact, specific_evidence_request. The supporting relationship sets are set-0c88cfd16df4.\n\nWhen to use it: choose this instead of the primary route when the live evidence matches contact_completed, relationship_manager_contact, specific_evidence_request more closely than the current top-ranked outcome.\n\nWhat to do now: make the action explicit for the credit risk lead, preserve the caveat for this domain, and avoid inventing extra facts beyond the governed handoff.\n\nOutcome to capture: record whether customer_contact_completed happened and feed that reviewed result back into the relationship-set evidence."
    }
  ]
}