{
  "schema": "aibast-canonical-transcripts/1.0",
  "captured_at": "2026-10-07T01:24:41.123728+00:00",
  "solution": "@aibast-agents-library/customer-sentiment-churn",
  "onepager": "37. Customer Sentiment and Churn Prediction Agent one-pager.pptx",
  "brainstem_version": "0.6.16",
  "loaded_tools_after_capture": [
    "CustomerSentimentChurnAgent"
  ],
  "agent_sources": [
    {
      "path": "agents/@aibast-agents-library/financial_services_stacks/customer_sentiment_churn_stack/customer_sentiment_churn_agent.py",
      "sha256": "f1313fa0dbbe637f3031d4789da99b312bbb274616cf231e5fb1e831c2b5064d"
    }
  ],
  "case_file": "tests/demo_cases/customer-sentiment-churn.json",
  "case_file_sha256": "a6492a47ecd283147fcd1723df00ce5ac24cbf21a2ad8d61d8bdc2bf0879fffe",
  "strict_isolation": true,
  "transcripts": [
    {
      "case_id": "CSC-01",
      "persona": "Customer Success Lead",
      "onepager_promise": "Aggregate sentiment across touchpoints to track trends before churn signals escalate",
      "prompt": "Analyze customer sentiment across our banking portfolio and identify accounts at highest churn risk.",
      "assistant_response": "Here's the portfolio snapshot (synthetic demo data, human review required):\n\n**Sentiment split** (180K customers): 40% Promoters, 45% Passives, 15% Detractors (up 18% this month). Top complaints: digital banking, fee transparency, wait times.\n\n**Highest churn risk:** 2,400 customers / $8.2M revenue at risk \u2014 840 critical (>80%), 1,560 high (60-80%).\n\n**Top 5 highest-value at-risk:**\n| Priority | Customer | Deposits | Risk | Key Signal |\n|---|---|---|---|---|\n| 1 | Robert Martinez | $890K | 84% | Downloaded competitor's app |\n| 2 | Sarah Thompson | $450K | 78% | Deposit pattern change |\n| 3 | James Lee | $340K | 76% | 3 complaints |\n| 4 | Marcus Johnson | $280K | 64% | 5 complaints |\n| 5 | Priya Sharma | $220K | 62% | Reduced logins |\n\nCombined: $2.18M deposits / $35K revenue at risk.\n\nWant me to pull the early warning signals to watch for, or generate retention strategies for these five?",
      "agent_logs": "[CustomerSentimentChurnAgent] > **SYNTHETIC DEMO DATA \u2014 HUMAN REVIEW REQUIRED.** Fictional customer signals only. Scores are prioritization heuristics, not facts about a real person; no outreach, offer, fee change, or account action has occurred.\n\n# Customer Sentiment Dashboard\n\n180K customers analyzed. 2,400 high churn risk representing $8.2M annual revenue.\n\n| Sentiment | Customers | % |\n|---|---|---|\n| Promoters | 72,000 | 40% |\n| Passives | 81,000 | 45% |\n| Detractors | 27,000 | 15% |\n\n## High Risk\n\n- **Critical (>80%):** 840 customers, $2.8M revenue, $3,333 avg\n- **High (60-80%):** 1,560 customers, $5.4M revenue, $3,462 avg\n\n**Top Negative Drivers:** Digital banking (2,400 mentions), Fee transparency (1,800 mentions), Wait times (1,200 mentions).\n\n**Alert:** Negative sentiment up 18% this month.\n\n## Highest-Value At-Risk Sample (interaction sentiment)\n\nSample interactions analyzed: 12 (Positive 0, Neutral 3, Negative 9)\n\n| Customer | Segment | NPS | Churn Probability | Recent Sentiment |\n|---|---|---|---|---|\n| Robert Martinez (CUST-8001) | Affluent | 4 | 84% (critical) | Negative |\n| Sarah Thompson (CUST-8003) | Affluent | 6 | 78% (high) | Neutral |\n| James Lee (CUST-8004) | Emerging Affluent | 3 | 76% (high) | Negative |\n| Marcus Johnson (CUST-8002) | Mass Market | 4 | 64% (high) | Negative |\n| Priya Sharma (CUST-8005) | Emerging Affluent | 7 | 62% (high) | Neutral |\n\nNext step: see the early warning signals?\n[CustomerSentimentChurnAgent] > **SYNTHETIC DEMO DATA \u2014 HUMAN REVIEW REQUIRED.** Fictional customer signals only. Scores are prioritization heuristics, not facts about a real person; no outreach, offer, fee change, or account action has occurred.\n\n# Churn Prediction Report: Highest-Value At-Risk Customers\n\nTop 5 at-risk profiles prepared.\n\n### Robert Martinez (CUST-8001) - Priority 1\n\n- $890K deposits, $12K annual revenue\n- Risk: 84% (critical); review score 60/100\n- Signals: Competitor app download (Summit National Bank), Reduced logins\n- Last survey: Detractor (4/10) - \"Your app is years behind\"\n- Recent contact: Asked about wire fees\n- Segment: Affluent, tenure 12 years, products: checking, savings, investment\n\n## Top 5 Summary\n\n| Priority | Customer | Deposits | Revenue | Risk | Key Signal |\n|---|---|---|---|---|---|\n| 1 | R. Martinez (CUST-8001) | $890K | $12K | 84% | competitor app |\n| 2 | S. Thompson (CUST-8003) | $450K | $8K | 78% | deposit changed |\n| 3 | J. Lee (CUST-8004) | $340K | $6K | 76% | 3 complaints |\n| 4 | M. Johnson (CUST-8002) | $280K | $5K | 64% | 5 complaints |\n| 5 | P. Sharma (CUST-8005) | $220K | $4K | 62% | reduced logins |\n\n**Combined Risk:** $2.18M deposits, $35K annual revenue at risk. Replacement cost if lost: $16K.\n\n## Churn Indicators Reference\n\n- **Low Nps** (weight: 25): NPS score below 5 indicates detractor status\n- **Declining Transactions** (weight: 20): Monthly transactions below half of the segment average\n- **High Complaints** (weight: 20): 3+ complaints in last 12 months\n- **Low Engagement** (weight: 15): Digital engagement score below 30\n- **Single Product** (weight: 10): Only one active product\n- **Stale Survey** (weight: 10): Last survey response over 90 days before the demo date\n\nThese scores prioritize review; they do not predict an individual outcome.\n\nNext step: generate retention strategies?",
      "expected_agent": "CustomerSentimentChurnAgent",
      "must_include": [
        "180K customers analyzed",
        "$8.2M",
        "Negative sentiment up 18%"
      ],
      "model": "claude-sonnet-5",
      "requested_model": "claude-sonnet-5",
      "passed": true
    },
    {
      "case_id": "CSC-02",
      "persona": "Retention Specialist",
      "onepager_promise": "Predict churn risk to pinpoint critical and high-risk customers",
      "prompt": "Show me profiles of our highest-value at-risk customers.",
      "assistant_response": "Here's the top-5 at-risk, highest-value customer profiles (all synthetic data, flagged for human review before any action):\n\n**1. Robert Martinez (CUST-8001) \u2014 Priority 1**\n- $890K deposits / $12K annual revenue, 84% risk (critical)\n- Downloaded a competitor's app, reduced logins, detractor survey (\"app is years behind\")\n- 12-year tenure, checking/savings/investment\n\n**2. Sarah Thompson (CUST-8003)** \u2014 $450K / $8K revenue, 78% risk \u2014 deposit activity changed\n\n**3. James Lee (CUST-8004)** \u2014 $340K / $6K revenue, 76% risk \u2014 3 complaints logged\n\n**4. Marcus Johnson (CUST-8002)** \u2014 $280K / $5K revenue, 64% risk \u2014 5 complaints logged\n\n**5. Priya Sharma (CUST-8005)** \u2014 $220K / $4K revenue, 62% risk \u2014 reduced digital engagement\n\n**Combined exposure:** $2.18M in deposits / $35K annual revenue at risk (replacement cost ~$16K if lost).\n\nWant me to pull retention strategies for any of these, or set up outreach tracking?",
      "agent_logs": "[CustomerSentimentChurnAgent] > **SYNTHETIC DEMO DATA \u2014 HUMAN REVIEW REQUIRED.** Fictional customer signals only. Scores are prioritization heuristics, not facts about a real person; no outreach, offer, fee change, or account action has occurred.\n\n# Churn Prediction Report: Highest-Value At-Risk Customers\n\nTop 5 at-risk profiles prepared.\n\n### Robert Martinez (CUST-8001) - Priority 1\n\n- $890K deposits, $12K annual revenue\n- Risk: 84% (critical); review score 60/100\n- Signals: Competitor app download (Summit National Bank), Reduced logins\n- Last survey: Detractor (4/10) - \"Your app is years behind\"\n- Recent contact: Asked about wire fees\n- Segment: Affluent, tenure 12 years, products: checking, savings, investment\n\n## Top 5 Summary\n\n| Priority | Customer | Deposits | Revenue | Risk | Key Signal |\n|---|---|---|---|---|---|\n| 1 | R. Martinez (CUST-8001) | $890K | $12K | 84% | competitor app |\n| 2 | S. Thompson (CUST-8003) | $450K | $8K | 78% | deposit changed |\n| 3 | J. Lee (CUST-8004) | $340K | $6K | 76% | 3 complaints |\n| 4 | M. Johnson (CUST-8002) | $280K | $5K | 64% | 5 complaints |\n| 5 | P. Sharma (CUST-8005) | $220K | $4K | 62% | reduced logins |\n\n**Combined Risk:** $2.18M deposits, $35K annual revenue at risk. Replacement cost if lost: $16K.\n\n## Churn Indicators Reference\n\n- **Low Nps** (weight: 25): NPS score below 5 indicates detractor status\n- **Declining Transactions** (weight: 20): Monthly transactions below half of the segment average\n- **High Complaints** (weight: 20): 3+ complaints in last 12 months\n- **Low Engagement** (weight: 15): Digital engagement score below 30\n- **Single Product** (weight: 10): Only one active product\n- **Stale Survey** (weight: 10): Last survey response over 90 days before the demo date\n\nThese scores prioritize review; they do not predict an individual outcome.\n\nNext step: generate retention strategies?",
      "expected_agent": "CustomerSentimentChurnAgent",
      "must_include": [
        "Robert Martinez",
        "$2.18M",
        "prioritize review"
      ],
      "model": "claude-sonnet-5",
      "requested_model": "claude-sonnet-5",
      "passed": true
    },
    {
      "case_id": "CSC-03",
      "persona": "Relationship Manager",
      "onepager_promise": "Generate tailored retention and outreach plans",
      "prompt": "Prepare options for Marcus that I can review before anyone contacts him or changes a fee.",
      "assistant_response": "Here are draft options for Marcus Johnson (CUST-8002) \u2014 nothing has been sent or applied, purely for your review:\n\n**Recommended strategy:** Relationship manager call, framed around his fee disputes and complaints, within 3 days (modeled 55% success).\n\n**Options to choose from:**\n| Action | What it does | Cost | Modeled success |\n|---|---|---|---|\n| Complaint Resolution | Escalate to service recovery team | $50 | 65% |\n| Product Bundle | Discounted bundle w/ waived fees | $200 | 60% |\n| Personal Outreach | RM call | $25 | 55% |\n| Loyalty Bonus | Credit to account | $100 | 50% |\n| Fee Waiver | Waive maintenance fee 6 mo. | $72 | 45% |\n| Rate Upgrade | Premium savings rate 12 mo. | $150 | 35% |\n\nDraft talk track (not sent): *\"Mr. Johnson, I reviewed your recent fee disputes and want to walk you through what we can fix.\"*\n\nAll options need your sign-off, policy validation, and his consent before anything happens. Want me to set up the outreach assignment and tracking once you pick an option?",
      "agent_logs": "[CustomerSentimentChurnAgent] > **SYNTHETIC DEMO DATA \u2014 HUMAN REVIEW REQUIRED.** Fictional customer signals only. Scores are prioritization heuristics, not facts about a real person; no outreach, offer, fee change, or account action has occurred.\n\n# Retention Action Recommendations\n\n### Marcus Johnson (CUST-8002) Strategy\n\n- Approach: Relationship manager call\n- Message: \"We reviewed your fees and complaints\"\n- Offer: Complaint resolution with a fee review\n- Incentive: Waive monthly maintenance fees for 6 months ($72 value)\n- Urgency: Within 3 days\n- Success probability (modeled): 55%\n- Draft talk track: \"Mr. Johnson, I reviewed your recent fee disputes and want to walk you through what we can fix.\"\n\n\n## Available Actions Catalog\n\n| Action | Description | Cost | Success Rate |\n|---|---|---|---|\n| Fee Waiver | Waive monthly maintenance fees for 6 months | $72 | 45% |\n| Rate Upgrade | Offer premium savings rate for 12 months | $150 | 35% |\n| Personal Outreach | Schedule call with relationship manager | $25 | 55% |\n| Product Bundle | Offer discounted product bundle with waived fees | $200 | 60% |\n| Loyalty Bonus | Credit loyalty bonus to account | $100 | 50% |\n| Complaint Resolution | Escalate to service recovery team | $50 | 65% |\n\nEvery option requires relationship-manager review, policy validation, customer consent where applicable, and approved execution. No customer was contacted and no offer was made.\n\nNext step: assign the outreach and set up tracking?",
      "expected_agent": "CustomerSentimentChurnAgent",
      "must_include": [
        "Marcus Johnson",
        "No customer was contacted"
      ],
      "model": "claude-sonnet-5",
      "requested_model": "claude-sonnet-5",
      "passed": true
    },
    {
      "case_id": "CSC-04",
      "persona": "Head of Customer Experience",
      "onepager_promise": "Aggregate sentiment across touchpoints to track trends before churn signals escalate",
      "prompt": "Which segment is under its experience benchmark, and what should we investigate?",
      "assistant_response": "The **Affluent** segment has the widest gap: NPS of 5.0 vs. an 8.2 benchmark (3.2-point shortfall), plus product holdings lagging (3.0 vs 4.1). Mass Market also underperforms (4.0 vs 6.5 NPS) but on a single customer, so less statistically meaningful.\n\nWorth investigating for Affluent:\n- Whether service/advisory experience matches the tier they're paying for (relationship manager responsiveness, wait times)\n- Fee-to-value perception \u2014 affluent clients are sensitive to feeling \"nickel-and-dimed\"\n- Why product penetration is below benchmark \u2014 missed cross-sell or unmet needs driving them elsewhere\n- Any recent incidents (outages, errors, rate changes) concentrated in this segment\n\nWant me to pull churn predictions or early warning signals specifically for the affluent customers to see if this NPS gap is translating into flight risk?",
      "agent_logs": "[CustomerSentimentChurnAgent] > **SYNTHETIC DEMO DATA \u2014 HUMAN REVIEW REQUIRED.** Fictional customer signals only. Scores are prioritization heuristics, not facts about a real person; no outreach, offer, fee change, or account action has occurred.\n\n# Segment Analysis\n\n## Segment Benchmarks\n\n| Segment | Avg NPS | Avg Products | Avg Tenure | Avg Transactions |\n|---|---|---|---|---|\n| Affluent | 8.2 | 4.1 | 10 yrs | 55/mo |\n| Emerging Affluent | 7.0 | 3.2 | 5 yrs | 35/mo |\n| Mass Market | 6.5 | 2.0 | 4 yrs | 20/mo |\n| Small Business | 6.8 | 3.0 | 5 yrs | 90/mo |\n\n## Current Customer Performance vs Benchmark\n\n### Affluent (2 customers)\n\n- NPS: 5.0 (benchmark: 8.2)\n- Products: 3.0 (benchmark: 4.1)\n\n### Emerging Affluent (2 customers)\n\n- NPS: 5.0 (benchmark: 7.0)\n- Products: 3.0 (benchmark: 3.2)\n\n### Mass Market (1 customers)\n\n- NPS: 4.0 (benchmark: 6.5)\n- Products: 2.0 (benchmark: 2.0)\n",
      "expected_agent": "CustomerSentimentChurnAgent",
      "must_include": [
        "Mass Market",
        "benchmark"
      ],
      "model": "claude-sonnet-5",
      "requested_model": "claude-sonnet-5",
      "passed": true
    },
    {
      "case_id": "CSC-05",
      "persona": "Retention Specialist",
      "onepager_promise": "Score churn risk by unifying sentiment, behavior, and all activity",
      "prompt": "Yes, what are the early warning signals we should watch for?",
      "assistant_response": "Here's the early warning picture (synthetic demo data, human review required before any action):\n\n**4 predictive signals** (appear 30-60 days before churn):\n- Direct deposit change \u2192 78% churn rate\n- Reduced logins \u2192 72%\n- Competitor app download \u2192 68%\n- Balance decline >30% \u2192 54%\n\n**Dangerous combos:**\n- Competitor app + balance decline \u2192 89% churn\n- Deposit change + reduced logins \u2192 82%\n- 2+ complaints \u2192 67%\n\n**Firing right now:**\n- 47 customers with competitor app detected\n- 23 with transfers >$10K out\n- 12 with changed direct deposit\n\n**High-value at-risk today:** Robert Martinez ($890K, competitor app), Sarah Thompson ($450K, deposit changed), James Lee ($340K, 3 complaints).\n\nWant me to pull full churn-risk profiles on these three, or set up outreach tracking for the review team?",
      "agent_logs": "[CustomerSentimentChurnAgent] > **SYNTHETIC DEMO DATA \u2014 HUMAN REVIEW REQUIRED.** Fictional customer signals only. Scores are prioritization heuristics, not facts about a real person; no outreach, offer, fee change, or account action has occurred.\n\n# Early Warning Signals\n\n4 key predictive signals identified, shown 30-60 days before leaving.\n\n| Signal | Customers | Churn Rate |\n|---|---|---|\n| Reduced logins | 3,400 | 72% |\n| Competitor app download | 1,200 | 68% |\n| Direct deposit change | 890 | 78% |\n| Balance decline >30% | 2,100 | 54% |\n\n## Critical Combinations\n\n- Competitor app + balance decline: 89% churn\n- Deposit change + reduced logins: 82% churn\n- 2+ complaints: 67% churn\n\n## Real-Time Alerts Today\n\n- 47 customers: Competitor app detected\n- 23 customers: External transfers >$10K\n- 12 customers: Direct deposit changed\n\n**High-Value at Risk:** Robert Martinez ($890K, competitor app), Sarah Thompson ($450K, deposit changed), James Lee ($340K, 3 complaints)\n\nNext step: profile the highest-value at-risk customers?",
      "expected_agent": "CustomerSentimentChurnAgent",
      "must_include": [
        "Reduced logins",
        "89% churn",
        "Robert Martinez"
      ],
      "model": "claude-sonnet-5",
      "requested_model": "claude-sonnet-5",
      "passed": true
    },
    {
      "case_id": "CSC-06",
      "persona": "Customer Success Lead",
      "onepager_promise": "Generate targeted retention strategies and recommended outreach plans",
      "prompt": "Yes, assign the outreach and set up tracking.",
      "assistant_response": "Outreach plan is staged \u2014 nothing's been sent yet, it's all pending your confirmation in the CRM.\n\n**Assignments (proposed):**\n| Customer | Owner | Channel | Due |\n|---|---|---|---|\n| R. Martinez | VP Chen | Phone | Today 2 PM |\n| S. Thompson | Sr. RM Johnson | Video | Tomorrow |\n| J. Lee | Sr. RM Williams | In-person | Tomorrow |\n| M. Johnson | RM Patel | Phone | Day 3 |\n| P. Sharma | RM Garcia | Phone+email | Day 3 |\n\n**Tracking set up:** real-time contact attempts, RM-logged outcomes, immediate offer-acceptance tracking, 30-day balance/activity monitoring.\n\n**Escalation rules:** no contact in 48h \u2192 manager alert; declined offer \u2192 VP escalation; balance withdrawal \u2192 instant notification.\n\nNothing goes out to customers until you approve each brief/offer in the CRM \u2014 want me to pull up the talking points for any one of them first?",
      "agent_logs": "[CustomerSentimentChurnAgent] > **SYNTHETIC DEMO DATA \u2014 HUMAN REVIEW REQUIRED.** Fictional customer signals only. Scores are prioritization heuristics, not facts about a real person; no outreach, offer, fee change, or account action has occurred.\n\n# Proposed Outreach Assignments and Tracking\n\nOutreach plan ready for the manager to confirm: all 5 high-priority customers have a proposed owner, channel and due time.\n\n| Customer | Assigned To | Channel | Due |\n|---|---|---|---|\n| R. Martinez (CUST-8001) | VP Chen | Phone | Today 2 PM |\n| S. Thompson (CUST-8003) | Sr. RM Johnson | Video call | Tomorrow |\n| J. Lee (CUST-8004) | Sr. RM Williams | In-person | Tomorrow |\n| M. Johnson (CUST-8002) | RM Patel | Phone | Day 3 |\n| P. Sharma (CUST-8005) | RM Garcia | Phone + email | Day 3 |\n\n**Prepared for release (not sent):** customer briefs and talk tracks drafted, offers queued for approval, calendar holds proposed.\n\n## Tracking\n\n- Contact attempts: real-time\n- Outcomes: RM updates\n- Offer acceptance: immediate\n- Account activity: 30-day monitoring\n\n## Escalation Rules\n\n- No contact in 48 hours -> alert manager\n- Declines offer -> escalate to VP\n- Balance withdrawal -> immediate notification\n\nNo customer was contacted, no offer was made, and no assignment, brief or calendar hold was sent; the manager confirms the plan in the CRM.",
      "expected_agent": "CustomerSentimentChurnAgent",
      "must_include": [
        "VP Chen",
        "Escalation Rules",
        "No customer was contacted"
      ],
      "model": "claude-sonnet-5",
      "requested_model": "claude-sonnet-5",
      "passed": true
    }
  ]
}