How Do You Reduce Azure AI Services Costs Across Cognitive and OpenAI Workloads?

# List all Cognitive Services accounts in your subscription
az cognitiveservices account list \
  --query "[].{Name:name, Kind:kind, SKU:sku.name, Location:location, RG:resourceGroup}" \
  --output table

# Check usage metrics for a specific account (last 30 days)
az monitor metrics list \
  --resource /subscriptions/{sub-id}/resourceGroups/{rg}/providers/Microsoft.CognitiveServices/accounts/{account-name} \
  --metric "TotalCalls" \
  --interval PT1H \
  --start-time $(date -u -d '30 days ago' +%Y-%m-%dT%H:%M:%SZ) \
  --end-time $(date -u +%Y-%m-%dT%H:%M:%SZ) \
  --output table

# List custom models to identify candidates for replacement
az cognitiveservices account list \
  --query "[?kind=='CustomVision.Training' || kind=='CustomVision.Prediction']" \
  --output table

Decision Flowchart: Commitment Tier vs Pay-As-You-Go
=====================================================

1. Is monthly usage consistent (variance < 20%)?
   |
   ├── YES → Is monthly spend > $500 on this service?
   |          |
   |          ├── YES → Switch to commitment tier
   |          |          → Choose tier where usage ≥ 80% of commitment
   |          |
   |          └── NO  → Stay on pay-as-you-go
   |                     (commitment minimums may exceed your usage)
   |
   └── NO  → Is baseline usage (lowest month) > $300?
              |
              ├── YES → Commit to baseline level
              |          → Pay-as-you-go for burst above commitment
              |
              └── NO  → Stay on pay-as-you-go
                         → Re-evaluate quarterly

# Check current SKU and pricing tier
az cognitiveservices account show \
  --name my-computer-vision \
  --resource-group my-rg \
  --query "{Name:name, SKU:sku.name, Tier:sku.tier}" \
  --output table

# List available commitment plans for a service
az cognitiveservices commitment-plan list \
  --resource-group my-rg \
  --account-name my-computer-vision \
  --output table

# Create a commitment plan (example: Computer Vision)
az cognitiveservices commitment-plan create \
  --resource-group my-rg \
  --account-name my-computer-vision \
  --commitment-plan-name "cv-monthly-plan" \
  --hosting-model "Web" \
  --plan-type "ComputerVision.S1" \
  --auto-renew true \
  --current "{commitment-count:1,tier:'S1'}"

# Verify the commitment plan is active
az cognitiveservices commitment-plan show \
  --resource-group my-rg \
  --account-name my-computer-vision \
  --commitment-plan-name "cv-monthly-plan" \
  --output json

# List all Azure OpenAI deployments to identify batch candidates
az cognitiveservices account deployment list \
  --name my-openai-account \
  --resource-group my-rg \
  --query "[].{Name:name, Model:properties.model.name, Version:properties.model.version, SKU:sku.name, Capacity:sku.capacity}" \
  --output table

# Example output:
# Name            Model      Version    SKU        Capacity
# --------------  ---------  ---------  ---------  --------
# gpt4o-realtime  gpt-4o     2024-08    Standard   120
# gpt4o-batch     gpt-4o     2024-08    Standard   80
# embed-prod      text-embedding-3-large  1    Standard   350

# List all Cognitive Search services and their tiers
az search service list \
  --query "[].{Name:name, SKU:sku.name, Replicas:replicaCount, Partitions:partitionCount, RG:resourceGroup}" \
  --output table

# Check index statistics to see actual storage usage
az search index list \
  --service-name my-search-service \
  --resource-group my-rg \
  --output table

# Export index definition for migration to a lower tier
az search index show \
  --service-name my-search-service \
  --resource-group my-rg \
  --name my-index \
  --output json > index-definition.json

# Create a new search service at a lower tier
az search service create \
  --name my-search-service-s1 \
  --resource-group my-rg \
  --sku s1 \
  --location eastus \
  --replica-count 1 \
  --partition-count 1

# Create a budget with tiered alerts for AI services
az consumption budget create \
  --budget-name "ai-services-monthly" \
  --amount 10000 \
  --time-grain Monthly \
  --start-date 2026-02-01 \
  --end-date 2027-01-31 \
  --resource-group ai-services-rg \
  --category Cost \
  --notifications \
    '{
      "50pct": {
        "enabled": true,
        "operator": "GreaterThan",
        "threshold": 50,
        "contactEmails": ["[email protected]"],
        "contactRoles": ["Owner"]
      },
      "75pct": {
        "enabled": true,
        "operator": "GreaterThan",
        "threshold": 75,
        "contactEmails": ["[email protected]", "[email protected]"],
        "contactRoles": ["Owner", "Contributor"]
      },
      "90pct": {
        "enabled": true,
        "operator": "GreaterThan",
        "threshold": 90,
        "contactEmails": ["[email protected]", "[email protected]", "[email protected]"],
        "contactRoles": ["Owner"]
      },
      "100pct-forecast": {
        "enabled": true,
        "operator": "GreaterThan",
        "threshold": 100,
        "thresholdType": "Forecasted",
        "contactEmails": ["[email protected]", "[email protected]"],
        "contactRoles": ["Owner"]
      }
    }'

Priority Matrix: Azure AI Cost Optimization
=============================================

HIGH IMPACT, LOW EFFORT (Do First)
├── Switch eligible services to commitment tiers
├── Apply PTU reservations for steady OpenAI workloads
├── Switch non-prod storage from GRS to LRS
└── Set up tiered budget alerts

HIGH IMPACT, MEDIUM EFFORT (Do Next)
├── Right-size Cognitive Search tiers
├── Move batch-eligible workloads to Batch API
├── Switch idle PTU deployments to pay-as-you-go
└── Optimize compute hosting (Functions to consumption)

MEDIUM IMPACT, MEDIUM EFFORT (Plan For)
├── Replace custom models with prebuilt APIs
├── Optimize search index storage
├── Implement idle shutdown automation
└── Consolidate redundant service instances

URGENT (Retirement Deadlines)
├── LUIS → CLU migration (deadline: March 31, 2026)
└── Anomaly Detector replacement (deadline: October 1, 2026)