Continuous Cost Monitor

Continuous Cost Monitor Skill

Tracks actual infrastructure costs against estimated costs. Alerts when actual spend drifts from projection, identifies cost drivers, and suggests corrective actions.

When to Use

Use this skill to:

1. Detect cost creep — infrastructure spending slowly increases over time
2. Root-cause cost spikes — which service changed? traffic? resource use?
3. Compare actual vs. estimated — am I spending what I projected?
4. Trigger optimization cycles — when do I need to re-optimize?
5. Alert on budget overruns — notify team before exceeding budget

Input

Provide estimated costs + actual spend data:

{
  "estimated_monthly_cost": 8500,
  "actual_monthly_cost": 9200,
  "cost_breakdown_estimated": {
    "backend_compute": 1200,
    "database": 1800,
    "cache": 200,
    "cdn": 300,
    "monitoring": 300,
    "storage": 500,
    "other": 400
  },
  "cost_breakdown_actual": {
    "backend_compute": 1500,
    "database": 2100,
    "cache": 200,
    "cdn": 350,
    "monitoring": 400,
    "storage": 450,
    "other": 200
  },
  "traffic_metrics": {
    "dau_estimated": 100000,
    "dau_actual": 110000,
    "query_volume_estimated": 500000,
    "query_volume_actual": 650000
  },
  "measurement_period_days": 30,
  "budget_threshold": 9000
}

Output

A cost monitoring report:

{
  "period": "April 2026",
  "estimated_vs_actual": {
    "estimated": 8500,
    "actual": 9200,
    "variance": 700,
    "variance_percent": 8.2,
    "status": "YELLOW (over by $700)"
  },
  "cost_breakdown_analysis": {
    "backend_compute": {
      "estimated": 1200,
      "actual": 1500,
      "variance": 300,
      "variance_percent": 25,
      "driver": "Traffic 10% higher than expected (110k vs 100k DAU)"
    },
    "database": {
      "estimated": 1800,
      "actual": 2100,
      "variance": 300,
      "variance_percent": 16.7,
      "driver": "Query volume 30% higher (650k vs 500k queries). Likely N+1 queries or missing cache."
    },
    "cache": {
      "estimated": 200,
      "actual": 200,
      "variance": 0,
      "variance_percent": 0,
      "status": "On budget"
    },
    "cdn": {
      "estimated": 300,
      "actual": 350,
      "variance": 50,
      "variance_percent": 16.7,
      "driver": "Traffic 10% higher (correlates with DAU increase)"
    },
    "monitoring": {
      "estimated": 300,
      "actual": 400,
      "variance": 100,
      "variance_percent": 33,
      "driver": "Increased metric cardinality (added new dashboards or higher scrape frequency)"
    },
    "storage": {
      "estimated": 500,
      "actual": 450,
      "variance": -50,
      "variance_percent": -10,
      "status": "Better than expected"
    },
    "other": {
      "estimated": 400,
      "actual": 200,
      "variance": -200,
      "variance_percent": -50,
      "status": "Significantly better than expected"
    }
  },
  "root_cause_analysis": {
    "primary_driver": "Database query volume 30% higher than expected",
    "contributing_factors": [
      "Traffic 10% higher (110k vs 100k DAU) — accounts for ~$300",
      "Inefficient queries — N+1 patterns or missing caching — accounts for ~$200",
      "Monitoring cost creep — added dashboards — accounts for ~$100"
    ],
    "confidence": 0.85
  },
  "trend_analysis": {
    "month_1_actual": 8100,
    "month_2_actual": 8800,
    "month_3_actual": 9200,
    "trend": "Increasing",
    "monthly_growth": 5.5,
    "projected_next_month": 9700,
    "annualized_growth": 66
  },
  "alerts": [
    {
      "level": "YELLOW",
      "message": "Cost overrun: +$700 (+8.2%) vs. estimate",
      "action": "Review database query patterns and cache hit rates"
    },
    {
      "level": "ORANGE",
      "message": "Database costs growing (16.7% overrun). Query volume 30% higher.",
      "action": "Profile slow queries, implement caching, optimize N+1 patterns"
    },
    {
      "level": "ORANGE",
      "message": "Monitoring cost increasing (33% overrun). Cardinality growth?",
      "action": "Audit new metrics added. Disable unused dashboards."
    },
    {
      "level": "YELLOW",
      "message": "Cost trend: +5.5% per month. At this rate, exceed $10k budget in 2 months.",
      "action": "Run cost optimization (reduce scope or increase efficiency)"
    }
  ],
  "budget_status": {
    "budget": 9000,
    "actual": 9200,
    "over_budget": true,
    "headroom": -200,
    "trajectory": "Will exceed budget next month at current trend"
  },
  "recommendations": [
    {
      "action": "Optimize database queries",
      "potential_savings": 200,
      "effort": "Medium (1-2 weeks)",
      "confidence": "High"
    },
    {
      "action": "Review monitoring cardinality",
      "potential_savings": 100,
      "effort": "Low (1-2 days)",
      "confidence": "High"
    },
    {
      "action": "Consider caching layer for read-heavy queries",
      "potential_savings": 150,
      "effort": "Medium (1 week)",
      "confidence": "Medium"
    },
    {
      "action": "Scale traffic analysis — is 110k DAU sustainable?",
      "potential_savings": 300,
      "effort": "Low (planning only)",
      "confidence": "Medium"
    }
  ],
  "escalation_actions": {
    "if_over_10_percent": "Run `/architect:optimize-costs` to identify quick wins",
    "if_trend_continues": "Trigger roadmap planning discussion (may need rearchitecture)",
    "if_exceeded_budget": "Notify finance and product teams; discuss scope/timeline changes"
  }
}

Monitoring Categories

1. Variance Detection

Green (0-5% overrun):

• Expected traffic fluctuation
• No action needed
• Continue monitoring

Yellow (5-15% overrun):

• Actionable overrun
• Identify root cause
• Plan optimization (1-2 weeks)

Orange (15-30% overrun):

• Significant drift
• Immediate investigation required
• Implement fixes within 1 week

Red (>30% overrun):

• Major unexpected costs
• Emergency response
• May need scaling or rearchitecture

2. Service-Level Analysis

For each service, calculate:
- Estimated cost per unit ($/DAU, $/query, $/GB)
- Actual cost per unit
- Variance ratio = actual / estimated

If variance > 1.2 (20% overrun):
  - Service is performing worse than expected
  - Prioritize optimization

Example:

Database estimated: $18/1000 queries = $1.8k at 100k DAU
Database actual: $21/1000 queries = $2.1k at 110k DAU (30% more queries!)

Root cause: Queries increased 30% (not linear with traffic).
Hypothesis: N+1 pattern, missing cache, or change in feature usage.

3. Trend Analysis

Linear growth (stable):

Month 1: $8.1k
Month 2: $8.8k
Month 3: $9.2k
Growth: ~$0.5k/month

Projection: Will hit $10k at month 8
Action: Schedule optimization for month 5-6 (preventive)

Exponential growth (concerning):

Month 1: $8.1k
Month 2: $9.2k (+13%)
Month 3: $10.5k (+14%)

Projection: Will hit $15k at month 6
Action: Emergency optimization needed this month

Seasonal patterns:

Traffic spikes certain days (weekends, holidays):
- Normal: expected pattern
- Action: Adjust budget for peak days

Unexpected spike (viral event, marketing campaign):
- Investigate: did marketing team run unplanned campaign?
- Document: for future capacity planning

4. Budget Tracking

Budget: $9,000/month
Threshold red: $10,000 (11% over)
Threshold orange: $9,500 (5.5% over)
Threshold yellow: $9,200 (2.2% over)

Current: $9,200 (YELLOW)
Action: Monitor closely; if hits $9,500, trigger optimization

Monitoring Algorithm

run_monthly_cost_monitor():
  1. Collect actual costs from cloud providers (AWS, Vercel, etc.)
  
  2. Calculate total variance
     total_variance = actual_total - estimated_total
     variance_percent = (total_variance / estimated_total) × 100
  
  3. For each service:
     service_variance = actual[service] - estimated[service]
     if service_variance > 15% of estimated[service]:
       flag as "needs investigation"
  
  4. Analyze trends
     growth_rate = (month_3 - month_1) / (month_1 × 2)
     if growth_rate > 10% per month:
       projection = estimated × (1 + growth_rate) ^ months_remaining
       if projection > budget:
         alert "cost trajectory exceeds budget"
  
  5. Correlate with metrics
     for each service_with_variance:
       if traffic increased:
         expected_increase = service_estimated × (traffic_actual / traffic_estimated)
         unexplained_variance = service_actual - expected_increase
         if unexplained_variance > 0:
           "query efficiency degraded"
       if traffic flat:
         "efficiency lost (not traffic-driven)"
  
  6. Generate alerts + recommendations
     sort recommendations by impact × ease
     for each, suggest: action, savings, effort, when to implement

Real-World Monitoring Examples

Example 1: Normal Growth

Month 1 (Jan): $8.1k actual vs $8k estimated (+1.25%)
Month 2 (Feb): $8.8k actual vs $8.3k estimated (+6%)
Month 3 (Mar): $9.2k actual vs $8.5k estimated (+8.2%)

Traffic: 100k → 105k → 110k DAU (linear +5% per month)
Cost growth: +5.5% per month

Status: ✅ HEALTHY
- Cost growth correlates with traffic
- Variance stays under 10%
- Trend is linear (not exponential)

Action: Continue monitoring. No intervention needed yet.

Example 2: Unexpected Database Spike

Month 1 (Jan): Database $1.7k (on budget)
Month 2 (Feb): Database $1.9k (+11%, expected for +5% traffic)
Month 3 (Mar): Database $2.4k (+26% vs Feb, but traffic only +5%)

Root cause investigation:
- Query volume: 600k in March (was 500k in Jan) → +20%
- Traffic: 110k DAU (was 100k in Jan) → +10%
- Mismatch: queries grew 2× traffic growth

Hypothesis: New feature (user notifications?) doing N+1 queries
Evidence: Queries per user jumped from 5 to 6

Action: Profile notification feature, add caching, optimize queries
Expected savings: -$300/mo (-12% database cost)
Timeline: 1 week implementation

Example 3: Monitoring Cost Creep

Month 1: Monitoring $250 (expected)
Month 2: Monitoring $300 (+20%, no traffic increase)
Month 3: Monitoring $400 (+33% vs expected)

Analysis:
- Dashboards added: 3 new custom dashboards
- Metric cardinality: Up from 500 → 2000 metrics

Root cause: Team added detailed metrics for each user segment
Cost impact: $1.50/metric when cardinality grows

Action: Disable unused dashboards, consolidate metrics
Expected savings: -$100/mo
Timeline: 2 days (administrative)

Example 4: Capacity-Driven Growth (Healthy)

Month 1: 100k DAU, $8.1k
Month 2: 150k DAU, $9.8k
Month 3: 200k DAU, $11.2k

Analysis:
- Cost per DAU: $81 → $65 → $56 (improving!)
- Reason: Economies of scale (database utilization higher)

Status: ✅ EFFICIENT SCALING
- Variance: +15% of estimated (acceptable for 2× traffic growth)
- Cost per unit improving (good sign)
- Trend: Sustainable

Action: Continue at current architecture. No optimization needed.
Next major action: Plan rearchitecture at 500k DAU.

Example 5: Warning: Budget Trajectory

Budget: $9,000/month
Month 1: $8.1k
Month 2: $8.9k
Month 3: $9.4k

Trend: +$650/month growth
Projection at month 6: $11.3k (26% over budget)

Status: 🔴 ALERT — on collision course with budget

Immediate actions (this week):
1. Profile database for N+1 queries (-$300 potential)
2. Review monitoring cardinality (-$100 potential)
3. Implement caching for hot queries (-$200 potential)
Total potential: -$600/mo (brings us to $8.8k, under budget)

Timeline: 2-3 weeks
If not implemented: Will exceed budget in 5 weeks

Integration with Other Commands

In /architect:cost-estimate

// After estimating costs:
setup_monitoring(estimated_costs)
  → Track actual vs estimated
  → Monthly alerts on variance
  → Feed into optimization cycles

In /architect:optimize-costs

// When suggesting optimizations:
if actual_variance > 10%:
  identify_root_cause()
  → Shows which optimizations address the root cause
  → Prioritizes by impact × ease

Example: "Database 20% over expected. Root cause: queries up 30%.
Suggested optimizations: add caching (-$200), optimize queries (-$300)"

New command: /architect:monitor-costs

/architect:monitor-costs [--period monthly|weekly|quarterly]
→ Generates cost monitoring report
→ Compares actual vs estimated
→ Flags variance, identifies drivers
→ Suggests optimizations

/architect:monitor-costs --alert-threshold 5000
→ Trigger optimization if costs exceed $5k

Limitations

Accuracy factors:

• Cloud provider APIs don't report instantly (1-3 day lag typical)
• Cost allocation (shared resources) hard to attribute perfectly
• Pricing changes (provider raises rates mid-month) unpredictable
• One-time costs (data transfer, API calls) volatile

When monitoring less reliable:

• First month (no baseline to compare)
• During major infrastructure changes (hard to isolate variance)
• With new features affecting load patterns (unknown relationship to cost)

Recommendation: Collect 3 months of data before making major decisions based on variance trends. Single-month spikes are often noise.

Related Skills

• cost-optimizer/ — identifies optimization opportunities
• cost-whatif-simulator/ — models impact of changes before implementing
• constraint-solver/ — ties cost monitoring to project constraints
• stack-swap-simulator/ — assesses cost impact of architecture changes