bi-agents/docs/BI-CCC-Implementation-Guide.md

BI-CCC Implementation Guide — Developer Reference

Audience: Engineering Team Date: February 2026 Status: Approved for implementation Current Maturity: 45/100 → Target: 85/100

---

Table of Contents

1. Overview & Current State
2. Phase 1 — Quick Wins (Weeks 1-4)
3. Phase 2 — Data Architecture (Weeks 5-10)
4. Phase 3 — Compliance & Predictive Analytics (Weeks 11-16)
5. Phase 4 — Platform Maturity (Weeks 17-24)
6. New KPIs to Implement
7. Architecture Reference
8. Code Refactoring Guide
9. Infrastructure & DevOps
10. Appendix — SQL Templates

---

1. Overview & Current State

What We Have Today

Stack: Node.js 20 + Express 4 + MySQL RDS (read-only) + SQLite + Chart.js
Deploy: Docker Compose + Netbird VPN
Auth: bcrypt + express-session (3 roles: admin/corporate/agente)

Critical Problems to Solve

Problem	Risk Level	Impact
Direct queries against production RDS	CRITICAL	Performance degradation, no historical data
queries.js = 1,786 lines monolith	HIGH	Unmaintainable, hard to test
admin-bi.js = 2,133 lines inline HTML/CSS/JS	HIGH	No component reuse, can't unit test
No alerting system	HIGH	Issues discovered manually, delayed response
No compliance/AML monitoring	CRITICAL	Regulatory exposure for forex platform
No ETL pipeline	HIGH	No snapshots, no transformation layer
No export/scheduled reports	MEDIUM	Manual report generation for stakeholders

---

2. Phase 1 — Quick Wins (Weeks 1-4)

2.1 Alerting System

Goal: Automated notifications when KPIs breach thresholds.

Stack: node-cron + Slack Webhooks + Nodemailer

File: src/alerts/alert-engine.js

// src/alerts/alert-engine.js
const cron = require('node-cron');
const { getPool } = require('../db-rds');
const { sendSlack, sendEmail } = require('./channels');

const ALERTS = [
  {
    name: 'volume-spike',
    cron: '*/30 * * * *', // every 30 min
    query: `
      SELECT
        COUNT(*) as current_count,
        (SELECT COUNT(*) / 7 FROM br_transaction_to_usa
         WHERE created_at >= DATE_SUB(NOW(), INTERVAL 7 DAY)
         AND created_at < DATE_SUB(NOW(), INTERVAL 30 MINUTE)) as avg_count
      FROM br_transaction_to_usa
      WHERE created_at >= DATE_SUB(NOW(), INTERVAL 30 MINUTE)
    `,
    condition: (row) => row.current_count > row.avg_count * 1.5 || row.current_count < row.avg_count * 0.5,
    severity: 'P0',
    channels: ['slack', 'email'],
    message: (row) => `Volume Alert: ${row.current_count} txs in last 30min (avg: ${Math.round(row.avg_count)})`
  },
  {
    name: 'spread-anomaly',
    cron: '*/15 * * * *',
    query: `
      SELECT
        AVG(exchange_rate - ptax) as current_spread,
        (SELECT AVG(exchange_rate - ptax) FROM br_transaction_to_usa
         WHERE status = 'finalizado'
         AND created_at >= DATE_SUB(NOW(), INTERVAL 7 DAY)) as avg_spread,
        (SELECT STDDEV(exchange_rate - ptax) FROM br_transaction_to_usa
         WHERE status = 'finalizado'
         AND created_at >= DATE_SUB(NOW(), INTERVAL 7 DAY)) as stddev_spread
      FROM br_transaction_to_usa
      WHERE status = 'finalizado'
      AND created_at >= DATE_SUB(NOW(), INTERVAL 15 MINUTE)
      HAVING COUNT(*) > 0
    `,
    condition: (row) => Math.abs(row.current_spread - row.avg_spread) > 2 * row.stddev_spread,
    severity: 'P0',
    channels: ['slack'],
    message: (row) => `Spread Anomaly: Current ${row.current_spread.toFixed(4)} vs avg ${row.avg_spread.toFixed(4)} (±2σ = ${(row.stddev_spread * 2).toFixed(4)})`
  },
  {
    name: 'provider-down',
    cron: '0 */4 * * *', // every 4 hours
    query: `
      SELECT pm.provider,
        MAX(t.created_at) as last_tx,
        TIMESTAMPDIFF(HOUR, MAX(t.created_at), NOW()) as hours_since
      FROM br_payment_methods pm
      LEFT JOIN br_transaction_to_usa t ON t.payment_method_id = pm.id
        AND t.created_at >= DATE_SUB(NOW(), INTERVAL 24 HOUR)
      GROUP BY pm.provider
      HAVING hours_since > 4 OR last_tx IS NULL
    `,
    condition: (rows) => rows.length > 0,
    severity: 'P0',
    channels: ['slack', 'email'],
    message: (rows) => `Provider Down: ${rows.map(r => `${r.provider} (${r.hours_since}h)`).join(', ')}`
  },
  {
    name: 'large-transaction',
    cron: '*/5 * * * *',
    query: `
      SELECT t.id_transaction_to_usa, t.amount_usd, c.nome, t.created_at
      FROM br_transaction_to_usa t
      JOIN conta c ON c.id_conta = t.id_conta
      WHERE t.amount_usd > 50000
      AND t.created_at >= DATE_SUB(NOW(), INTERVAL 5 MINUTE)
      AND t.status = 'finalizado'
    `,
    condition: (rows) => rows.length > 0,
    severity: 'P1',
    channels: ['slack'],
    message: (rows) => `Large Tx: ${rows.map(r => `$${r.amount_usd.toLocaleString()} from ${r.nome}`).join('; ')}`
  },
  {
    name: 'failed-tx-spike',
    cron: '*/15 * * * *',
    query: `
      SELECT
        SUM(CASE WHEN status IN ('refunded','on-hold') THEN 1 ELSE 0 END) as failed,
        COUNT(*) as total,
        SUM(CASE WHEN status IN ('refunded','on-hold') THEN 1 ELSE 0 END) / COUNT(*) * 100 as fail_rate
      FROM br_transaction_to_usa
      WHERE created_at >= DATE_SUB(NOW(), INTERVAL 1 HOUR)
      HAVING total > 0
    `,
    condition: (row) => row.fail_rate > 5,
    severity: 'P0',
    channels: ['slack', 'email'],
    message: (row) => `Failed Tx Spike: ${row.fail_rate.toFixed(1)}% failure rate (${row.failed}/${row.total}) in last hour`
  }
];

// Slack channel sender
async function sendSlack(webhook, message, severity) {
  const colors = { P0: '#ef4444', P1: '#f59e0b', P2: '#3b82f6' };
  await fetch(webhook, {
    method: 'POST',
    headers: { 'Content-Type': 'application/json' },
    body: JSON.stringify({
      attachments: [{
        color: colors[severity] || '#94a3b8',
        title: `BI-CCC Alert [${severity}]`,
        text: message,
        ts: Math.floor(Date.now() / 1000)
      }]
    })
  });
}

module.exports = { ALERTS, initAlerts };

Env vars to add:

SLACK_WEBHOOK_URL=https://hooks.slack.com/services/xxx/yyy/zzz
ALERT_EMAIL_TO=ops@cambioreal.com
SMTP_HOST=smtp.gmail.com
SMTP_USER=alerts@cambioreal.com
SMTP_PASS=app-password

2.2 PDF/Excel Export

Goal: Add export buttons to all dashboards.

Stack: Already have puppeteer and exceljs in querycambio — move to shared dependency.

File: src/export/report-generator.js

// src/export/report-generator.js
const puppeteer = require('puppeteer');
const ExcelJS = require('exceljs');

async function exportDashboardToPDF(url, sessionCookie) {
  const browser = await puppeteer.launch({
    headless: true,
    args: ['--no-sandbox']
  });
  const page = await browser.newPage();
  await page.setCookie({ name: 'connect.sid', value: sessionCookie, domain: 'localhost' });
  await page.goto(url, { waitUntil: 'networkidle0' });

  // Hide nav, expand all sections
  await page.evaluate(() => {
    document.querySelector('.console-nav')?.remove();
    document.querySelector('header')?.remove();
  });

  const pdf = await page.pdf({
    format: 'A4',
    landscape: true,
    printBackground: true,
    margin: { top: '1cm', bottom: '1cm', left: '1cm', right: '1cm' }
  });

  await browser.close();
  return pdf;
}

async function exportToExcel(data, columns, sheetName = 'Report') {
  const workbook = new ExcelJS.Workbook();
  const sheet = workbook.addWorksheet(sheetName);

  sheet.columns = columns.map(col => ({
    header: col.label,
    key: col.key,
    width: col.width || 15,
    style: col.style || {}
  }));

  // Header styling
  sheet.getRow(1).font = { bold: true, size: 11 };
  sheet.getRow(1).fill = { type: 'pattern', pattern: 'solid', fgColor: { argb: '7600BE' } };
  sheet.getRow(1).font = { bold: true, color: { argb: 'FFFFFF' } };

  data.forEach(row => sheet.addRow(row));

  return workbook;
}

module.exports = { exportDashboardToPDF, exportToExcel };

API endpoints to add in server.js:

// Export endpoints
app.get('/admin/api/export/bi-pdf', requireRole('admin'), async (req, res) => {
  const { start, end } = req.query;
  const pdf = await exportDashboardToPDF(
    `http://localhost:${PORT}/admin/bi?start=${start}&end=${end}`,
    req.cookies['connect.sid']
  );
  res.setHeader('Content-Type', 'application/pdf');
  res.setHeader('Content-Disposition', `attachment; filename=BI-Report-${start}-${end}.pdf`);
  res.send(pdf);
});

app.get('/admin/api/export/clients-excel', requireRole('admin'), async (req, res) => {
  const data = await queries.getTopClients();
  const workbook = await exportToExcel(data, [
    { key: 'nome', label: 'Client', width: 30 },
    { key: 'volume_usd', label: 'Volume (USD)', width: 18 },
    { key: 'transactions', label: 'Transactions', width: 15 },
    { key: 'revenue', label: 'Revenue (USD)', width: 18 },
    { key: 'health_score', label: 'Health Score', width: 15 }
  ], 'Top Clients');

  res.setHeader('Content-Type', 'application/vnd.openxmlformats-officedocument.spreadsheetml.sheet');
  res.setHeader('Content-Disposition', 'attachment; filename=top-clients.xlsx');
  await workbook.xlsx.write(res);
});

2.3 Provider Performance Dashboard

Goal: Compare all payin/payout providers in one view.

File: src/admin-providers.js

Query: src/queries/provider-queries.js

-- Provider performance (payin)
SELECT
  pm.provider,
  COUNT(*) as total_transactions,
  SUM(CASE WHEN t.status = 'finalizado' THEN 1 ELSE 0 END) as completed,
  SUM(CASE WHEN t.status = 'refunded' THEN 1 ELSE 0 END) as refunded,
  ROUND(SUM(CASE WHEN t.status = 'finalizado' THEN 1 ELSE 0 END) / COUNT(*) * 100, 2) as success_rate,
  ROUND(SUM(t.amount_usd), 2) as total_volume_usd,
  ROUND(AVG(t.amount_usd), 2) as avg_ticket,
  ROUND(AVG(TIMESTAMPDIFF(HOUR, t.created_at, t.date_received_br)), 1) as avg_settlement_hours,
  ROUND(AVG(t.exchange_rate - t.ptax), 4) as avg_spread
FROM br_transaction_to_usa t
JOIN br_payment_methods pm ON pm.id = t.payment_method_id
WHERE t.created_at BETWEEN ? AND ?
GROUP BY pm.provider
ORDER BY total_volume_usd DESC;

-- Provider performance (payout)
SELECT
  ex.provider,
  COUNT(et.id) as total_transactions,
  SUM(CASE WHEN et.status = 'completed' THEN 1 ELSE 0 END) as completed,
  ROUND(SUM(CASE WHEN et.status = 'completed' THEN 1 ELSE 0 END) / COUNT(*) * 100, 2) as success_rate,
  ROUND(SUM(et.amount_brl), 2) as total_volume_brl
FROM pagamento_br_export_transactions et
JOIN pagamento_br_exports ex ON ex.id = et.export_id
WHERE et.created_at BETWEEN ? AND ?
GROUP BY ex.provider
ORDER BY total_volume_brl DESC;

2.4 Failed Transaction Analysis

File: src/admin-failed-transactions.js

-- Failed transactions breakdown
SELECT
  t.status,
  pm.provider,
  COUNT(*) as count,
  ROUND(SUM(t.amount_usd), 2) as volume_lost,
  DATE(t.created_at) as date
FROM br_transaction_to_usa t
JOIN br_payment_methods pm ON pm.id = t.payment_method_id
WHERE t.status IN ('refunded', 'on-hold')
AND t.created_at BETWEEN ? AND ?
GROUP BY t.status, pm.provider, DATE(t.created_at)
ORDER BY date DESC, count DESC;

2.5 New KPIs for Corporate Dashboard

Add these to GET /corporate/api/kpis:

-- Transaction Success Rate
SELECT
  ROUND(SUM(CASE WHEN status = 'finalizado' THEN 1 ELSE 0 END) / COUNT(*) * 100, 2) as success_rate
FROM br_transaction_to_usa
WHERE created_at BETWEEN ? AND ?;

-- Refund Rate
SELECT
  ROUND(SUM(CASE WHEN status = 'refunded' THEN 1 ELSE 0 END) / COUNT(*) * 100, 2) as refund_rate
FROM br_transaction_to_usa
WHERE created_at BETWEEN ? AND ?;

-- Gross Take Rate
-- (Total Revenue / Total Volume)
-- Use existing revenue calculation, divide by total volume

---

3. Phase 2 — Data Architecture (Weeks 5-10)

3.1 Analytics Database Setup

Recommended: ClickHouse (best price/performance for analytical queries)

Alternative: TimescaleDB on PostgreSQL (easier migration path)

Option A: ClickHouse

# docker-compose.yml addition
services:
  clickhouse:
    image: clickhouse/clickhouse-server:24.1
    container_name: bi-clickhouse
    ports:
      - "8123:8123"  # HTTP
      - "9000:9000"  # Native
    volumes:
      - clickhouse-data:/var/lib/clickhouse
    environment:
      CLICKHOUSE_DB: bi_analytics
      CLICKHOUSE_USER: bi_user
      CLICKHOUSE_PASSWORD: ${CLICKHOUSE_PASSWORD}
    ulimits:
      nofile:
        soft: 262144
        hard: 262144

volumes:
  clickhouse-data:

ClickHouse Schema

-- Transactions fact table (payin)
CREATE TABLE bi_analytics.fact_payin (
  id UInt64,
  created_at DateTime,
  date_received DateTime,
  id_conta UInt64,
  client_name String,
  agente_id Nullable(UInt32),
  amount_brl Decimal(18,2),
  amount_usd Decimal(18,2),
  exchange_rate Decimal(12,6),
  ptax Decimal(12,6),
  spread Decimal(12,6),
  fee Decimal(12,2),
  iof Decimal(12,2),
  pfee Decimal(12,2),
  bonus_valor Decimal(12,2),
  taxa_cr Decimal(12,2),
  revenue Decimal(12,2),
  provider String,
  status String,
  tipo_envio String,
  tipo_pagamento String,
  is_checkout UInt8,
  merchant_id Nullable(UInt64),
  merchant_name Nullable(String),
  -- Calculated fields
  spread_pct Decimal(8,4),
  settlement_hours Nullable(Float32)
) ENGINE = MergeTree()
ORDER BY (created_at, id_conta)
PARTITION BY toYYYYMM(created_at);

-- Transactions fact table (payout)
CREATE TABLE bi_analytics.fact_payout (
  id UInt64,
  created_at DateTime,
  id_conta UInt64,
  sender_conta_id Nullable(UInt64),
  client_name String,
  valor_usd Decimal(18,2),
  valor_brl Decimal(18,2),
  ptax Decimal(12,6),
  cotacao Decimal(12,6),
  revenue Decimal(12,2),
  tipo_envio String,
  provider String,
  status String
) ENGINE = MergeTree()
ORDER BY (created_at, id_conta)
PARTITION BY toYYYYMM(created_at);

-- Daily snapshots (materialized)
CREATE TABLE bi_analytics.daily_metrics (
  date Date,
  product String, -- 'cambioPay', 'cambioPayReverso', 'cambioCheckout'
  total_transactions UInt32,
  completed_transactions UInt32,
  failed_transactions UInt32,
  total_volume_usd Decimal(18,2),
  total_volume_brl Decimal(18,2),
  total_revenue Decimal(18,2),
  avg_spread Decimal(8,4),
  avg_ticket_usd Decimal(12,2),
  unique_clients UInt32,
  new_clients UInt32,
  avg_settlement_hours Nullable(Float32)
) ENGINE = MergeTree()
ORDER BY (date, product);

-- Client health snapshots
CREATE TABLE bi_analytics.client_health_daily (
  date Date,
  id_conta UInt64,
  client_name String,
  health_score UInt8,
  churn_risk String,  -- 'low','medium','high','critical'
  lifetime_volume Decimal(18,2),
  lifetime_revenue Decimal(18,2),
  lifetime_transactions UInt32,
  last_transaction_date Nullable(Date),
  days_since_last_tx UInt16,
  monthly_velocity Decimal(12,2)
) ENGINE = MergeTree()
ORDER BY (date, id_conta)
PARTITION BY toYYYYMM(date);

3.2 ETL Pipeline

File: src/etl/pipeline.js

// src/etl/pipeline.js
const cron = require('node-cron');
const mysql = require('../db-rds');
const clickhouse = require('../db-clickhouse');
const logger = require('../etl/logger');

const BATCH_SIZE = 5000;

class ETLPipeline {

  // Run every hour — sync new transactions
  async syncPayinTransactions() {
    const jobId = logger.startJob('sync-payin');

    try {
      // Get last synced ID
      const [lastRow] = await clickhouse.query(
        'SELECT max(id) as last_id FROM bi_analytics.fact_payin'
      );
      const lastId = lastRow?.last_id || 0;

      // Fetch new records from MySQL
      const [rows] = await mysql.query(`
        SELECT
          t.id_transaction_to_usa as id,
          t.created_at,
          t.date_received_br as date_received,
          t.id_conta,
          c.nome as client_name,
          ac.agente_id,
          t.amount_brl,
          t.amount_usd,
          t.exchange_rate,
          t.ptax,
          (t.exchange_rate - t.ptax) as spread,
          t.fee,
          t.iof,
          COALESCE(t.pfee, 0) as pfee,
          COALESCE(t.bonus_valor, 0) as bonus_valor,
          COALESCE(t.taxa_cr, 0) as taxa_cr,
          pm.provider,
          t.status,
          t.tipo_envio,
          t.tipo_pagamento,
          CASE WHEN t.cobranca_id IS NOT NULL THEN 1 ELSE 0 END as is_checkout,
          e.id as merchant_id,
          e.nome_empresa as merchant_name
        FROM br_transaction_to_usa t
        JOIN conta c ON c.id_conta = t.id_conta
        LEFT JOIN ag_contas ac ON ac.conta_id = t.id_conta
        LEFT JOIN br_payment_methods pm ON pm.id = t.payment_method_id
        LEFT JOIN br_cb_cobranca cb ON cb.id = t.cobranca_id
        LEFT JOIN br_cb_empresas e ON e.id = cb.empresa_id
        WHERE t.id_transaction_to_usa > ?
        ORDER BY t.id_transaction_to_usa
        LIMIT ?
      `, [lastId, BATCH_SIZE]);

      if (rows.length === 0) {
        logger.endJob(jobId, 'success', 0);
        return;
      }

      // Calculate revenue for each row
      const enriched = rows.map(row => ({
        ...row,
        revenue: calculatePayinRevenue(row),
        spread_pct: row.ptax > 0
          ? ((row.exchange_rate - row.ptax) / row.exchange_rate * 100)
          : 0,
        settlement_hours: row.date_received
          ? (new Date(row.date_received) - new Date(row.created_at)) / 3600000
          : null
      }));

      // Insert into ClickHouse
      await clickhouse.insert('bi_analytics.fact_payin', enriched);

      logger.endJob(jobId, 'success', rows.length);
    } catch (err) {
      logger.endJob(jobId, 'error', 0, err.message);
      throw err;
    }
  }

  // Run daily at 1 AM — build daily aggregates
  async buildDailyMetrics(date) {
    const jobId = logger.startJob('daily-metrics');
    const targetDate = date || new Date(Date.now() - 86400000).toISOString().split('T')[0];

    await clickhouse.query(`
      INSERT INTO bi_analytics.daily_metrics
      SELECT
        toDate(created_at) as date,
        CASE
          WHEN is_checkout = 1 THEN 'cambioCheckout'
          ELSE 'cambioPay'
        END as product,
        count() as total_transactions,
        countIf(status = 'finalizado') as completed_transactions,
        countIf(status IN ('refunded', 'on-hold')) as failed_transactions,
        sum(amount_usd) as total_volume_usd,
        sum(amount_brl) as total_volume_brl,
        sum(revenue) as total_revenue,
        avg(spread_pct) as avg_spread,
        avg(amount_usd) as avg_ticket_usd,
        uniq(id_conta) as unique_clients,
        uniqIf(id_conta, /* first tx logic */) as new_clients,
        avg(settlement_hours) as avg_settlement_hours
      FROM bi_analytics.fact_payin
      WHERE toDate(created_at) = '${targetDate}'
      GROUP BY date, product
    `);

    logger.endJob(jobId, 'success');
  }
}

// Revenue calculation (mirrors existing logic)
function calculatePayinRevenue(row) {
  const noFeeProviders = ['ouribank', 'bs2'];
  const feeAjustado = noFeeProviders.includes(row.provider) ? 0 : (row.fee || 0);

  if (row.ptax <= 0) return 0;

  const revenue = (row.amount_brl - feeAjustado) / row.ptax
    - (row.pfee || 0)
    - (row.amount_usd + (row.bonus_valor || 0) - (row.taxa_cr || 0));

  return Math.round(revenue * 100) / 100;
}

module.exports = ETLPipeline;

Cron schedule (add to server.js or separate worker):

const ETLPipeline = require('./src/etl/pipeline');
const etl = new ETLPipeline();

// Hourly: sync new transactions
cron.schedule('0 * * * *', () => etl.syncPayinTransactions());
cron.schedule('5 * * * *', () => etl.syncPayoutTransactions());

// Daily at 1 AM: build aggregates
cron.schedule('0 1 * * *', () => etl.buildDailyMetrics());

// Daily at 2 AM: build client health snapshots
cron.schedule('0 2 * * *', () => etl.buildClientHealthSnapshots());

3.3 Data Quality Monitoring

File: src/etl/data-quality.js

// Run after every ETL job
async function validateDataQuality() {
  const checks = [];

  // 1. Row count comparison
  const [mysqlCount] = await mysql.query(
    'SELECT COUNT(*) as cnt FROM br_transaction_to_usa WHERE status = "finalizado"'
  );
  const [chCount] = await clickhouse.query(
    "SELECT count() as cnt FROM bi_analytics.fact_payin WHERE status = 'finalizado'"
  );
  checks.push({
    name: 'row-count-payin',
    expected: mysqlCount.cnt,
    actual: chCount.cnt,
    pass: Math.abs(mysqlCount.cnt - chCount.cnt) < 100 // tolerance for in-flight
  });

  // 2. Null checks on critical fields
  const [nullCheck] = await clickhouse.query(`
    SELECT
      countIf(ptax = 0 OR ptax IS NULL) as null_ptax,
      countIf(amount_brl = 0) as zero_amount,
      countIf(provider = '') as empty_provider
    FROM bi_analytics.fact_payin
    WHERE created_at >= today() - 1
  `);
  checks.push({
    name: 'null-critical-fields',
    pass: nullCheck.null_ptax === 0 && nullCheck.zero_amount === 0
  });

  // 3. Revenue reconciliation
  const [revCheck] = await clickhouse.query(`
    SELECT
      abs(sum(revenue) - ${expectedRevenue}) / ${expectedRevenue} * 100 as deviation_pct
    FROM bi_analytics.fact_payin
    WHERE toDate(created_at) = yesterday()
  `);
  checks.push({
    name: 'revenue-reconciliation',
    pass: revCheck.deviation_pct < 1 // less than 1% deviation
  });

  return checks;
}

---

4. Phase 3 — Compliance & Predictive Analytics (Weeks 11-16)

4.1 AML Transaction Monitoring

File: src/compliance/aml-monitor.js

New route: GET /admin/compliance

-- Threshold monitoring: transactions > $3,000
SELECT
  c.nome, t.id_conta, t.amount_usd, t.created_at, pm.provider
FROM br_transaction_to_usa t
JOIN conta c ON c.id_conta = t.id_conta
JOIN br_payment_methods pm ON pm.id = t.payment_method_id
WHERE t.amount_usd > 3000
AND t.status = 'finalizado'
AND t.created_at BETWEEN ? AND ?
ORDER BY t.amount_usd DESC;

-- Structuring detection: multiple transactions just below threshold per client per day
SELECT
  c.nome,
  t.id_conta,
  DATE(t.created_at) as tx_date,
  COUNT(*) as tx_count,
  SUM(t.amount_usd) as daily_total,
  AVG(t.amount_usd) as avg_amount,
  MIN(t.amount_usd) as min_amount,
  MAX(t.amount_usd) as max_amount
FROM br_transaction_to_usa t
JOIN conta c ON c.id_conta = t.id_conta
WHERE t.status = 'finalizado'
AND t.amount_usd BETWEEN 2500 AND 3000
AND t.created_at BETWEEN ? AND ?
GROUP BY t.id_conta, c.nome, DATE(t.created_at)
HAVING tx_count >= 2
ORDER BY daily_total DESC;

-- Unusual velocity: clients with transaction frequency spike
SELECT
  c.nome,
  t.id_conta,
  COUNT(*) as recent_count,
  (SELECT COUNT(*) FROM br_transaction_to_usa t2
   WHERE t2.id_conta = t.id_conta
   AND t2.created_at BETWEEN DATE_SUB(?, INTERVAL 30 DAY) AND ?) / 30 as daily_avg,
  SUM(t.amount_usd) as recent_volume
FROM br_transaction_to_usa t
JOIN conta c ON c.id_conta = t.id_conta
WHERE t.status = 'finalizado'
AND t.created_at BETWEEN ? AND ?
GROUP BY t.id_conta, c.nome
HAVING recent_count > daily_avg * 3
ORDER BY recent_count DESC;

4.2 IOF Reconciliation Dashboard

-- IOF verification: compare calculated vs expected
SELECT
  DATE(t.created_at) as date,
  pm.provider,
  COUNT(*) as transactions,
  SUM(t.iof) as total_iof_charged,
  SUM(t.amount_usd * t.exchange_rate * 0.0038) as expected_iof_standard,
  SUM(t.iof) - SUM(t.amount_usd * t.exchange_rate * 0.0038) as iof_delta,
  ROUND(
    ABS(SUM(t.iof) - SUM(t.amount_usd * t.exchange_rate * 0.0038))
    / NULLIF(SUM(t.amount_usd * t.exchange_rate * 0.0038), 0) * 100, 2
  ) as delta_pct
FROM br_transaction_to_usa t
JOIN br_payment_methods pm ON pm.id = t.payment_method_id
WHERE t.status = 'finalizado'
AND t.created_at BETWEEN ? AND ?
GROUP BY DATE(t.created_at), pm.provider
HAVING delta_pct > 0.5
ORDER BY date DESC;

4.3 Volume Forecasting

Stack: Python microservice with Prophet or Node.js with timeseries-analysis

File: src/analytics/forecast.js (Node approach)

// Simple exponential smoothing for Node.js
// For production, consider a Python sidecar with Prophet

function exponentialSmoothing(data, alpha = 0.3) {
  const result = [data[0]];
  for (let i = 1; i < data.length; i++) {
    result.push(alpha * data[i] + (1 - alpha) * result[i - 1]);
  }
  return result;
}

function forecast(historicalData, periodsAhead = 30) {
  // historicalData = [{ date, value }, ...]
  const values = historicalData.map(d => d.value);
  const smoothed = exponentialSmoothing(values);

  // Simple trend projection
  const recent = smoothed.slice(-7);
  const trend = (recent[recent.length - 1] - recent[0]) / recent.length;

  const lastValue = smoothed[smoothed.length - 1];
  const predictions = [];

  for (let i = 1; i <= periodsAhead; i++) {
    const lastDate = new Date(historicalData[historicalData.length - 1].date);
    lastDate.setDate(lastDate.getDate() + i);
    predictions.push({
      date: lastDate.toISOString().split('T')[0],
      predicted: Math.max(0, lastValue + trend * i),
      lower: Math.max(0, lastValue + trend * i - lastValue * 0.15),
      upper: lastValue + trend * i + lastValue * 0.15
    });
  }

  return predictions;
}

module.exports = { forecast };

API endpoint:

app.get('/admin/api/bi/forecast', requireRole('admin'), async (req, res) => {
  const { metric = 'volume', days = 30 } = req.query;

  // Get 90 days of historical data from analytics DB
  const historical = await clickhouse.query(`
    SELECT date, sum(total_volume_usd) as value
    FROM bi_analytics.daily_metrics
    WHERE date >= today() - 90
    GROUP BY date
    ORDER BY date
  `);

  const predictions = forecast(historical, parseInt(days));
  res.json({ historical, predictions });
});

4.4 Churn Prediction Model

Evolve the static Health Score into a weighted model:

function predictChurnRisk(client) {
  const weights = {
    recency: 0.30,     // days since last transaction
    frequency: 0.25,   // transactions per month trend
    monetary: 0.20,    // volume trend (growing/declining)
    velocity: 0.15,    // rate of change in transaction frequency
    engagement: 0.10   // product diversity, ticket variation
  };

  // Recency score (0-100, higher = less risk)
  const recencyScore = Math.max(0, 100 - (client.days_since_last_tx * 2));

  // Frequency score: compare last 30d vs previous 30d
  const freqRatio = client.tx_count_30d / Math.max(client.tx_count_prev_30d, 1);
  const frequencyScore = Math.min(100, freqRatio * 50);

  // Monetary score: volume growth
  const volRatio = client.volume_30d / Math.max(client.volume_prev_30d, 1);
  const monetaryScore = Math.min(100, volRatio * 50);

  // Velocity: acceleration of transaction frequency
  const velocityScore = client.tx_acceleration > 0 ? 70 + client.tx_acceleration * 10 : 30;

  // Engagement: uses multiple products?
  const engagementScore = client.products_used * 33;

  const totalScore =
    recencyScore * weights.recency +
    frequencyScore * weights.frequency +
    monetaryScore * weights.monetary +
    Math.min(100, velocityScore) * weights.velocity +
    Math.min(100, engagementScore) * weights.engagement;

  return {
    score: Math.round(totalScore),
    risk: totalScore < 25 ? 'critical' : totalScore < 45 ? 'high' : totalScore < 65 ? 'medium' : 'low',
    factors: {
      recency: { score: recencyScore, weight: weights.recency },
      frequency: { score: frequencyScore, weight: weights.frequency },
      monetary: { score: monetaryScore, weight: weights.monetary },
      velocity: { score: Math.min(100, velocityScore), weight: weights.velocity },
      engagement: { score: Math.min(100, engagementScore), weight: weights.engagement }
    }
  };
}

---

5. Phase 4 — Platform Maturity (Weeks 17-24)

5.1 Scheduled Report Generation

File: src/reports/scheduler.js

const cron = require('node-cron');
const { exportDashboardToPDF, exportToExcel } = require('../export/report-generator');
const { sendEmail } = require('../alerts/channels');

const REPORT_SCHEDULE = [
  {
    name: 'Daily Executive Digest',
    cron: '0 8 * * 1-5', // 8 AM weekdays
    recipients: ['exec@cambioreal.com'],
    generator: async () => {
      const data = await queries.getDailyDigest();
      return {
        subject: `BI Daily Digest — ${new Date().toLocaleDateString()}`,
        html: buildDigestHTML(data),
        attachments: [] // inline, no PDF
      };
    }
  },
  {
    name: 'Weekly Board Summary',
    cron: '0 9 * * 1', // Monday 9 AM
    recipients: ['board@cambioreal.com'],
    generator: async () => {
      const end = new Date().toISOString().split('T')[0];
      const start = new Date(Date.now() - 7 * 86400000).toISOString().split('T')[0];
      const pdf = await exportDashboardToPDF(
        `http://localhost:3080/admin/bi?start=${start}&end=${end}`
      );
      return {
        subject: `BI Weekly Report — Week of ${start}`,
        html: '<p>Weekly BI report attached.</p>',
        attachments: [{ filename: `BI-Weekly-${start}.pdf`, content: pdf }]
      };
    }
  },
  {
    name: 'Monthly Agent Commissions',
    cron: '0 10 1 * *', // 1st of month, 10 AM
    recipients: ['finance@cambioreal.com'],
    generator: async () => {
      const data = await queries.getMonthlyCommissions();
      const workbook = await exportToExcel(data, commissionColumns, 'Commissions');
      const buffer = await workbook.xlsx.writeBuffer();
      return {
        subject: `Agent Commissions — ${new Date().toLocaleDateString('en', { month: 'long', year: 'numeric' })}`,
        html: '<p>Monthly commission report attached.</p>',
        attachments: [{ filename: 'commissions.xlsx', content: buffer }]
      };
    }
  }
];

5.2 Agent Commission Integration

Move commission calculation from separate scripts (comissao-valorfx/, comissao-d4u/) into BI-CCC:

New route: GET /admin/commissions

Features:

• Auto-calculate commissions per agent per period
• Approval workflow (admin reviews → approves → generates Excel)
• Commission dispute tracking
• Agent ROI: commission paid vs. revenue generated by agent's clients

// src/queries/commission-queries.js

// ValorFx (Agent 76) — 50/50 rateio
async function calculateValorFxCommission(start, end) {
  const [rows] = await pool.query(`
    SELECT
      t.id_transaction_to_usa,
      t.created_at,
      c.nome as client_name,
      t.amount_brl,
      t.amount_usd,
      t.exchange_rate,
      t.ptax,
      t.fee,
      t.iof,
      t.pfee,
      t.bonus_valor,
      t.taxa_cr,
      pm.provider
    FROM br_transaction_to_usa t
    JOIN conta c ON c.id_conta = t.id_conta
    JOIN ag_contas ac ON ac.conta_id = t.id_conta
    JOIN br_payment_methods pm ON pm.id = t.payment_method_id
    WHERE ac.agente_id = 76
    AND t.status = 'finalizado'
    AND t.created_at BETWEEN ? AND ?
    ORDER BY t.created_at
  `, [start, end]);

  return rows.map(row => {
    const noFeeProviders = ['ouribank', 'bs2'];
    const feeAdj = noFeeProviders.includes(row.provider) ? 0 : row.fee;

    const spreadLiqUsd = (row.amount_brl - feeAdj) / row.ptax
      - row.pfee - row.amount_usd + row.bonus_valor - row.taxa_cr;

    const iofEficiencia = row.provider === 'RIPPLE'
      ? row.iof
      : row.iof * 0.425;

    const rateio = (iofEficiencia + spreadLiqUsd) / 2;

    return { ...row, spreadLiqUsd, iofEficiencia, rateio };
  });
}

// D4U — 20% do spread
async function calculateD4UCommission(start, end) {
  const d4uContas = [257174, 333262, 349175, 352427];
  // ... similar structure, 20% of revenue
}

5.3 Frontend Migration Path

Recommended approach: Gradual migration using htmx + Alpine.js

Step 1: Extract UI Template Components

src/
├── components/
│   ├── kpi-card.js          // buildKPICard(title, value, change, color)
│   ├── chart-wrapper.js     // buildChartWrapper(id, type, config)
│   ├── data-table.js        // buildDataTable(columns, data, options)
│   ├── filter-bar.js        // buildFilterBar(filters)
│   └── layout.js            // buildPageLayout(title, nav, content)

Step 2: Add htmx for Progressive Enhancement

<!-- Instead of full-page reload, use htmx for partial updates -->
<div hx-get="/admin/api/bi/kpis?start=${start}&end=${end}"
     hx-trigger="load, change from:#date-filter"
     hx-target="#kpi-grid"
     hx-swap="innerHTML">
  Loading...
</div>

Step 3: Introduce Alpine.js for Client Interactivity

<div x-data="{ tab: 'overview' }">
  <button @click="tab = 'overview'" :class="{ active: tab === 'overview' }">Overview</button>
  <button @click="tab = 'revenue'" :class="{ active: tab === 'revenue' }">Revenue</button>

  <div x-show="tab === 'overview'">...</div>
  <div x-show="tab === 'revenue'">...</div>
</div>

5.4 Multi-Language Support

File: src/i18n/index.js

const translations = {
  en: {
    'nav.dashboard': 'Dashboard',
    'nav.bi': 'BI Executive',
    'nav.clients': 'Clients',
    'kpi.volume': 'Volume',
    'kpi.transactions': 'Transactions',
    'kpi.revenue': 'Revenue',
    'kpi.avgTicket': 'Avg Ticket',
    'kpi.spread': 'Spread',
    'kpi.healthScore': 'Health Score',
    'filter.period': 'Period',
    'filter.from': 'From',
    'filter.to': 'To',
    'status.completed': 'Completed',
    'status.refunded': 'Refunded',
    'status.pending': 'Pending'
  },
  pt: {
    'nav.dashboard': 'Painel',
    'nav.bi': 'BI Executivo',
    'nav.clients': 'Clientes',
    'kpi.volume': 'Volume',
    'kpi.transactions': 'Transações',
    'kpi.revenue': 'Receita',
    'kpi.avgTicket': 'Ticket Médio',
    'kpi.spread': 'Spread',
    'kpi.healthScore': 'Score de Saúde',
    'filter.period': 'Período',
    'filter.from': 'De',
    'filter.to': 'Até',
    'status.completed': 'Finalizado',
    'status.refunded': 'Estornado',
    'status.pending': 'Pendente'
  }
};

function t(key, locale = 'en') {
  return translations[locale]?.[key] || translations.en[key] || key;
}

module.exports = { t, translations };

---

6. New KPIs to Implement

Financial KPIs

KPI	SQL/Formula	Priority	Add To
Net Revenue Retention	(Start MRR + Expansion - Contraction - Churn) / Start MRR	P0	BI Executive
Revenue per Transaction	SUM(revenue) / COUNT(*)	P0	Corporate
Gross Take Rate	SUM(revenue) / SUM(volume_usd)	P0	BI Executive
Provider Cost per $1K	SUM(provider_fees) / (SUM(volume) / 1000)	P1	Provider Dashboard
LTV	AVG(monthly_revenue) × AVG(lifespan_months)	P1	Client 360
Revenue Concentration (HHI)	SUM(client_share^2)	P1	BI Executive
MRR	SUM(revenue) WHERE client active in consecutive months	P2	BI Executive

Operational KPIs

KPI	SQL	Priority	Add To
Transaction Success Rate	SUM(IF status='finalizado',1,0) / COUNT(*)	P0	Corporate
Avg Settlement Time	AVG(TIMESTAMPDIFF(HOUR, created_at, date_received_br))	P0	Provider Dashboard
Provider Uptime	COUNT(DISTINCT HOUR(created_at)) / 24 per provider	P1	Provider Dashboard
Refund Rate	SUM(IF status='refunded',1,0) / COUNT(*)	P1	Corporate
PTAX Deviation	AVG(ABS(exchange_rate - ptax) / ptax)	P1	BI Executive
Peak Hour Analysis	GROUP BY HOUR(created_at)	P2	Corporate

Client KPIs

KPI	Logic	Priority	Add To
Client Activation Rate	First-tx clients / New signups	P0	BI Executive
Cross-Product Adoption	Clients using 2+ products / Total active	P1	BI Executive
Time to First Transaction	AVG(first_tx - signup)	P1	BI Executive
Reactivation Rate	Previously churned → active again	P2	Client 360

Compliance KPIs

KPI	Logic	Priority	Add To
Threshold Tx Count	Transactions > $3K, $10K per client per day	P0	Compliance
Structuring Detection	Multiple txs between $2.5K-$3K same day	P0	Compliance
SAR Filing Rate	SARs filed per period	P0	Compliance
IOF Reconciliation Delta	Calculated vs expected IOF	P1	Compliance
Cross-Border Volume per Corridor	Volume by origin/destination	P1	Compliance

---

7. Architecture Reference

Current Architecture

Browser → Express (SSR HTML + inline JS/CSS) → MySQL RDS (direct)
                                              → SQLite (auth)

Target Architecture

Browser (htmx + Alpine.js + Chart.js)
    │
    ▼
Express API Gateway
    │
    ├── REST API layer (JSON)
    │     ├── /api/bi/*         → BI Service
    │     ├── /api/clients/*    → Client Service
    │     ├── /api/providers/*  → Provider Service
    │     ├── /api/compliance/* → Compliance Service
    │     ├── /api/agents/*     → Agent Service
    │     └── /api/export/*     → Export Service
    │
    ├── SSR HTML layer (pages)
    │     └── Uses components/ for reusable UI blocks
    │
    ├── Analytics DB (ClickHouse)
    │     └── fact tables + daily aggregates + client snapshots
    │
    ├── ETL Worker (node-cron)
    │     ├── Hourly: sync transactions
    │     ├── Daily: build aggregates + snapshots
    │     └── Data quality checks
    │
    ├── Alert Engine (node-cron)
    │     ├── Volume/spread monitors
    │     ├── Provider health checks
    │     └── Channels: Slack + Email
    │
    ├── Report Generator (Puppeteer + ExcelJS)
    │     ├── Scheduled: daily/weekly/monthly
    │     └── On-demand: PDF/Excel export
    │
    ├── MySQL RDS (source — read-only via VPN)
    │
    └── SQLite (auth + user preferences)

New File Structure

bi-agentes/
├── server.js                      # Route registration + middleware only
├── src/
│   ├── routes/
│   │   ├── bi.routes.js           # /admin/bi, /admin/api/bi
│   │   ├── client.routes.js       # /admin/cliente, /admin/api/clientes
│   │   ├── corporate.routes.js    # /corporate, /corporate/api
│   │   ├── provider.routes.js     # /admin/providers (NEW)
│   │   ├── compliance.routes.js   # /admin/compliance (NEW)
│   │   ├── commission.routes.js   # /admin/commissions (NEW)
│   │   ├── export.routes.js       # /admin/api/export (NEW)
│   │   ├── agent.routes.js        # /dashboard
│   │   └── auth.routes.js         # /login, /logout
│   │
│   ├── services/
│   │   ├── revenue.service.js     # Revenue calculation logic
│   │   ├── client.service.js      # Client health, churn, LTV
│   │   ├── provider.service.js    # Provider analytics
│   │   ├── compliance.service.js  # AML, threshold, structuring
│   │   ├── commission.service.js  # Agent commissions
│   │   ├── forecast.service.js    # Volume/revenue forecasting
│   │   └── metrics.service.js     # KPI calculations
│   │
│   ├── queries/
│   │   ├── payin.queries.js       # br_transaction_to_usa queries
│   │   ├── payout.queries.js      # pagamento_br queries
│   │   ├── client.queries.js      # Client-related queries
│   │   ├── provider.queries.js    # Provider-related queries
│   │   ├── compliance.queries.js  # AML/compliance queries
│   │   └── commission.queries.js  # Commission queries
│   │
│   ├── components/
│   │   ├── kpi-card.js
│   │   ├── chart-wrapper.js
│   │   ├── data-table.js
│   │   ├── filter-bar.js
│   │   ├── alert-badge.js
│   │   └── layout.js
│   │
│   ├── etl/
│   │   ├── pipeline.js            # Main ETL orchestrator
│   │   ├── sync-payin.js          # Payin transaction sync
│   │   ├── sync-payout.js         # Payout transaction sync
│   │   ├── build-aggregates.js    # Daily/weekly/monthly rollups
│   │   ├── build-snapshots.js     # Client health snapshots
│   │   ├── data-quality.js        # Validation checks
│   │   └── logger.js              # ETL job logging
│   │
│   ├── alerts/
│   │   ├── alert-engine.js        # Alert definitions + scheduler
│   │   ├── channels.js            # Slack, Email, SMS senders
│   │   └── alert-rules.js         # Alert configurations
│   │
│   ├── export/
│   │   ├── report-generator.js    # PDF + Excel generation
│   │   └── scheduler.js           # Scheduled report delivery
│   │
│   ├── i18n/
│   │   ├── index.js               # Translation helper
│   │   ├── en.json
│   │   └── pt.json
│   │
│   ├── db-rds.js                  # MySQL connection (unchanged)
│   ├── db-clickhouse.js           # ClickHouse connection (NEW)
│   ├── db-local.js                # SQLite (unchanged)
│   ├── auth.js                    # Auth middleware (unchanged)
│   ├── cache.js                   # Cache layer (unchanged)
│   └── ui-template.js             # Base template (refactored)
│
├── public/
│   ├── chart.umd.min.js
│   ├── htmx.min.js                # NEW
│   ├── alpine.min.js              # NEW
│   ├── login.html
│   └── assets/
│
├── data/
│   └── agentes.db
│
├── scripts/
│   ├── seed-admin.js
│   ├── seed-agente.js
│   ├── migrate-clickhouse.js      # NEW: schema setup
│   └── backfill-analytics.js      # NEW: historical data load
│
├── Dockerfile
├── docker-compose.yml             # + clickhouse service
├── docker-entrypoint.sh
└── .env

---

8. Code Refactoring Guide

Priority 1: Break up queries.js (1,786 lines)

Split by domain:

queries.js →
  ├── queries/payin.queries.js      (~500 lines)
  ├── queries/payout.queries.js     (~300 lines)
  ├── queries/client.queries.js     (~400 lines)
  ├── queries/provider.queries.js   (~200 lines — NEW)
  ├── queries/commission.queries.js (~200 lines — from external scripts)
  └── queries/compliance.queries.js (~200 lines — NEW)

Migration strategy: Keep queries.js as a facade that re-exports from sub-modules. This avoids breaking existing imports.

// queries.js (facade — no breaking changes)
const payin = require('./queries/payin.queries');
const payout = require('./queries/payout.queries');
const client = require('./queries/client.queries');

module.exports = {
  ...payin,
  ...payout,
  ...client
};

Priority 2: Extract UI Components from admin-bi.js (2,133 lines)

Identify repeating patterns and extract to src/components/:

// BEFORE (admin-bi.js, inline)
html += `<div class="hero-card">
  <div class="hero-label">${label}</div>
  <div class="hero-value">${value}</div>
  <div class="hero-change ${change > 0 ? 'up' : 'down'}">${change}%</div>
</div>`;

// AFTER (components/kpi-card.js)
function buildKPICard({ label, value, change, color = 'green' }) {
  const direction = change > 0 ? 'up' : 'down';
  return `<div class="hero-card hero-${color}">
    <div class="hero-label">${label}</div>
    <div class="hero-value">${value}</div>
    <div class="hero-change ${direction}">${change > 0 ? '+' : ''}${change}%</div>
  </div>`;
}
module.exports = { buildKPICard };

Priority 3: Separate CSS into Files

Extract inline styles from ui-template.js into CSS files:

public/css/
├── variables.css    # CSS custom properties (dark/light theme)
├── layout.css       # Grid, nav, page structure
├── components.css   # Cards, tables, buttons
├── charts.css       # Chart containers, legends
└── responsive.css   # Media queries

---

9. Infrastructure & DevOps

Docker Compose Update

version: '3.8'

services:
  bi-ccc:
    build: .
    container_name: bi-ccc
    ports:
      - "3080:3080"
    environment:
      - MYSQL_URL=${MYSQL_URL}
      - USER_MYSQL=${USER_MYSQL}
      - PW_MYSQL=${PW_MYSQL}
      - SESSION_SECRET=${SESSION_SECRET}
      - CLICKHOUSE_URL=http://clickhouse:8123
      - CLICKHOUSE_DB=bi_analytics
      - SLACK_WEBHOOK_URL=${SLACK_WEBHOOK_URL}
      - ALERT_EMAIL_TO=${ALERT_EMAIL_TO}
    depends_on:
      - clickhouse
    volumes:
      - ./data:/app/data
    cap_add:
      - NET_ADMIN
      - SYS_RESOURCE

  clickhouse:
    image: clickhouse/clickhouse-server:24.1
    container_name: bi-clickhouse
    ports:
      - "8123:8123"
    volumes:
      - clickhouse-data:/var/lib/clickhouse
    environment:
      - CLICKHOUSE_DB=bi_analytics
      - CLICKHOUSE_USER=${CLICKHOUSE_USER}
      - CLICKHOUSE_PASSWORD=${CLICKHOUSE_PASSWORD}

volumes:
  clickhouse-data:

New Dependencies

{
  "dependencies": {
    "node-cron": "^3.0.3",
    "nodemailer": "^6.9.8",
    "puppeteer": "^22.0.0",
    "exceljs": "^4.4.0",
    "@clickhouse/client": "^0.2.10"
  }
}

Monitoring

Add basic health check endpoint:

app.get('/health', (req, res) => {
  const checks = {
    mysql: pool ? 'connected' : 'disconnected',
    clickhouse: chPool ? 'connected' : 'disconnected',
    uptime: process.uptime(),
    memory: process.memoryUsage(),
    cache: cache.stats()
  };
  const healthy = checks.mysql === 'connected';
  res.status(healthy ? 200 : 503).json(checks);
});

---

10. Appendix — SQL Templates

A. Net Revenue Retention (NRR)

WITH monthly_revenue AS (
  SELECT
    id_conta,
    DATE_FORMAT(created_at, '%Y-%m') as month,
    SUM(revenue) as mrr
  FROM bi_analytics.fact_payin
  WHERE status = 'finalizado'
  GROUP BY id_conta, DATE_FORMAT(created_at, '%Y-%m')
),
retention AS (
  SELECT
    a.month as current_month,
    SUM(a.mrr) as current_mrr,
    SUM(CASE WHEN b.mrr IS NOT NULL THEN LEAST(a.mrr, b.mrr) ELSE 0 END) as retained,
    SUM(CASE WHEN b.mrr IS NOT NULL AND a.mrr > b.mrr THEN a.mrr - b.mrr ELSE 0 END) as expansion,
    SUM(CASE WHEN b.mrr IS NOT NULL AND a.mrr < b.mrr THEN b.mrr - a.mrr ELSE 0 END) as contraction,
    SUM(CASE WHEN b.mrr IS NULL AND a.mrr IS NOT NULL THEN 0 ELSE b.mrr END) as churned
  FROM monthly_revenue a
  LEFT JOIN monthly_revenue b ON a.id_conta = b.id_conta
    AND b.month = DATE_FORMAT(DATE_SUB(STR_TO_DATE(CONCAT(a.month, '-01'), '%Y-%m-%d'), INTERVAL 1 MONTH), '%Y-%m')
  GROUP BY a.month
)
SELECT
  current_month,
  ROUND((retained + expansion - contraction) / NULLIF(retained + contraction, 0) * 100, 1) as nrr_pct
FROM retention
ORDER BY current_month;

B. Revenue Concentration (HHI)

SELECT
  ROUND(SUM(POW(client_share, 2)) * 10000) as hhi_index
FROM (
  SELECT
    id_conta,
    SUM(revenue) / (SELECT SUM(revenue) FROM bi_analytics.fact_payin WHERE created_at BETWEEN ? AND ?) as client_share
  FROM bi_analytics.fact_payin
  WHERE created_at BETWEEN ? AND ?
  AND status = 'finalizado'
  GROUP BY id_conta
) sub;
-- HHI < 1500 = competitive, 1500-2500 = moderate concentration, > 2500 = high concentration

C. Cross-Product Adoption

SELECT
  products_used,
  COUNT(*) as client_count,
  ROUND(COUNT(*) / SUM(COUNT(*)) OVER() * 100, 1) as pct
FROM (
  SELECT
    id_conta,
    COUNT(DISTINCT product) as products_used
  FROM (
    SELECT id_conta,
      CASE WHEN is_checkout = 1 THEN 'checkout' ELSE 'cambioPay' END as product
    FROM bi_analytics.fact_payin
    WHERE created_at >= DATE_SUB(NOW(), INTERVAL 90 DAY)
    UNION ALL
    SELECT id_conta, 'cambioPayReverso' as product
    FROM bi_analytics.fact_payout
    WHERE created_at >= DATE_SUB(NOW(), INTERVAL 90 DAY)
  ) all_products
  GROUP BY id_conta
) sub
GROUP BY products_used
ORDER BY products_used;

D. Structuring Detection

SELECT
  c.nome as client_name,
  t.id_conta,
  DATE(t.created_at) as tx_date,
  COUNT(*) as tx_count,
  ROUND(SUM(t.amount_usd), 2) as daily_total,
  ROUND(AVG(t.amount_usd), 2) as avg_amount,
  ROUND(STDDEV(t.amount_usd), 2) as stddev_amount,
  GROUP_CONCAT(ROUND(t.amount_usd, 2) ORDER BY t.created_at) as amounts
FROM br_transaction_to_usa t
JOIN conta c ON c.id_conta = t.id_conta
WHERE t.status = 'finalizado'
AND t.amount_usd BETWEEN 2000 AND 3000
AND t.created_at BETWEEN ? AND ?
GROUP BY t.id_conta, c.nome, DATE(t.created_at)
HAVING tx_count >= 2 AND daily_total > 3000
ORDER BY daily_total DESC;

---

Checklist de Entregas

Phase 1 (Weeks 1-4)

• Alert engine (src/alerts/) com 5+ regras configuradas
• Slack webhook integrado e testado
• Export PDF/Excel nos dashboards existentes
• Provider Performance dashboard funcional
• Failed Transaction view
• Success Rate + Refund Rate no Corporate Dashboard
• queries.js dividido em módulos (facade mantida)

Phase 2 (Weeks 5-10)

• ClickHouse rodando no Docker Compose
• ETL pipeline sincronizando payin + payout
• Daily aggregates sendo gerados automaticamente
• Client health snapshots diários
• Data quality checks passando
• BI-CCC lendo de ClickHouse (não mais direto do RDS)

Phase 3 (Weeks 11-16)

• Compliance dashboard com threshold monitoring
• Structuring detection funcional
• IOF reconciliation view
• Volume forecast aparecendo no BI Executive
• Churn prediction substituindo Health Score estático
• NRR, LTV, Take Rate no BI Executive

Phase 4 (Weeks 17-24)

• Scheduled reports (daily/weekly/monthly) sendo enviados
• Commission module integrado ao BI-CCC
• Revenue forecast para board reporting
• Frontend migrado para components + htmx
• i18n EN/PT-BR funcional
• API documentada
• Audit logging de acesso a dados

---

Nota: Este guia assume que o time manterá a stack Node.js/Express. Se houver decisão de migrar para Python (FastAPI) ou Go para o backend de dados, as queries e lógica de negócio se mantêm — muda apenas a camada de serviço.