AI in Microsoft ERP · Model Choice · Series 4 · #035
Microsoft’s “best model for the job” strategy continues to deepen. As of June 2026, you can pick Anthropic’s Claude as a model inside Copilot Chat – and Microsoft’s own documentation recommends it specifically for D365 ERP MCP agent scenarios.
The Model Choice Timeline
Why This Matters Specifically for D365 ERP Agent Work
The most concrete, practical detail here is one we’ve referenced earlier in this series but that deserves its own focused explanation: when Microsoft tells you, in its own official documentation, to use a specific non-Microsoft model for a specific Microsoft scenario, that’s worth paying attention to. The Dynamics 365 ERP MCP server documentation states plainly that Claude Sonnet 4.5 “provides a better success rate in Copilot Studio over other default models like GPT-4.1” for agents that connect to D365 finance and operations data.
If you’re building custom agents via Copilot Studio against your D365 environment, this is directly actionable: when configuring the orchestration model for your agent, select Claude Sonnet 4.5 if it’s available in your tenant, and fall back to GPT-5 (Chat) if it isn’t. This single configuration choice has a measurable impact on how reliably your agent performs ERP-connected tasks.
A Practical Framework: When to Choose Which Model
Model selection inside Copilot doesn’t need to be a complicated decision for most users (as Microsoft generally sets sensible defaults) but for power users and agent builders, here’s a reasonable way to think about it. For straightforward, fast, everyday questions (the kind covered by GPT-5.5 Instant), the default model is usually fine and optimized for low latency. For complex, multi-step reasoning tasks — building a financial analysis, drafting a detailed process design document, working through a nuanced governance question – Claude’s strength in instruction-following and structured, long-form output makes it a strong choice when it’s available to you. For agents connected to D365 ERP data via the MCP server specifically, Microsoft’s own guidance points to Claude Sonnet 4.5 as the recommended orchestration model.
This isn’t a universal ranking – model performance varies by task type and continues to evolve with each release. The point is that model choice now exists as a real lever, and understanding when to pull it is a skill worth developing as part of the “AI fluency” theme we returned to throughout Series 3.
📚 Go Deeper — Microsoft Resources
- Build an Agent with D365 ERP MCP — Includes Model Recommendation
- Choose an External Model as the Primary AI Model — Admin Guide
- Microsoft 365 Copilot Blog — Monthly Model and Feature Updates
Model choice inside Copilot is a quiet but meaningful development for D365 organizations — it means you don’t have to choose between “stay inside the Microsoft ecosystem” and “use the best AI tool for complex reasoning work.” For agent builders specifically, Microsoft’s own recommendation to use Claude Sonnet 4.5 for ERP MCP scenarios is worth implementing directly.
BB
Bobbi Bricker
ERP Capability Lead and D365 Functional Architect at Centric Consulting. Former controller. This series reflects fifteen + years in ERP (as an end user and a Consultant) and a genuine belief that AI, used thoughtfully, makes finance and operations teams more capable — not less. Reach out with questions, pushback, or war stories from your own organizations.
Thank you for reading!
Interested in learning more? Below are some of my latest posts:
Leave a Reply