4.2.2. Licensing and Cost Models
💡 First Principle: AI costs follow two patterns: subscription (fixed monthly per user) or consumption (variable per use). CFOs who need budget predictability want subscriptions; teams experimenting with variable workloads prefer consumption. Mismatching the model to the need either wastes money (paying for unused subscriptions) or creates budget surprises (consumption costs spiking). Match the cost pattern to the use pattern.
Microsoft Copilot add-on licensing:
- Model: Per-user, per-month subscription
- Includes: Priority access to Copilot across the M365 apps, Microsoft Graph and Work IQ grounding
- Budget impact: Predictable monthly cost based on user count
- Consideration: Not all users may need Copilot—license strategically
Copilot Studio pricing:
- Model: Consumption-based, billed in Copilot Credits—the Copilot Studio maker/user license itself is free
- How you pay: Tenant-level Copilot Credit capacity packs (prepaid) and/or pay-as-you-go—not a per-user subscription
- Budget impact: Scales with agent usage; credits are consumed per agent response and action
- Consideration: Treating it as a flat per-user cost invites budget surprises as usage grows ("messages" were renamed Copilot Credits in September 2025)
Azure AI Services pricing:
- Model: Consumption-based (pay per use)
- Examples: Per API call, per processed image, per transcribed minute
- Budget impact: Variable cost based on usage
- Consideration: Can scale dramatically with volume—monitor and set limits
Microsoft Foundry pricing:
- Model: Combination of compute, model usage, and services
- Components: Model inference (per token), training compute, hosting
- Budget impact: Can be significant for custom model development
- Consideration: Development costs plus ongoing operational costs
| Solution | Pricing Model | Predictability | Best For |
|---|---|---|---|
| Microsoft Copilot | Per-user/month | High | Predictable productivity enhancement |
| Copilot Studio | Copilot Credits (capacity packs or pay-as-you-go) | Medium | Custom agents; cost scales with agent usage |
| Azure AI Services | Pay-per-use | Low | Variable workloads, development |
| Foundry | Compute + usage | Low | Custom AI product development |
Managing AI spend:
- Budget total cost, not just licenses: training, change management, integration, and ongoing support add to license or consumption fees.
- Judge value, not price: compare cost per business outcome; a higher-priced option with better capabilities, security, or support can deliver more value.
- Monitor actively: track usage patterns, value delivered per use case, and cost per outcome.
- Reclaim shelfware: licenses purchased but never used are wasted spend—review usage and reassign or drop them.
- Negotiate for growth: when usage will expand over several years, seek volume discounts and committed-use pricing (for example through an Enterprise Agreement).
⚠️ Exam Trap: Questions about budget predictability often have per-user licensing as the answer (predictable) versus consumption-based pricing (variable). Match the organization's need for predictability to the appropriate model.
Reflection Question: A CFO wants to invest in AI but is concerned about unpredictable costs. They're considering both Microsoft Copilot and building a custom solution in Azure. How would you advise them on cost predictability?
