Planning in the AI assistant is only as good as the model behind it

Corporate performance management (CPM) and workforce planning and analytics vendors spent 2026 opening their planning data to AI assistants through Model Context Protocol (MCP) servers and APIs, much as Excel integrations once brought planning into the tool finance already used. The appeal is wider participation, because budget owners can check results against plan or test a change by asking a question instead of learning the planning tool or waiting on FP&A. Most of these connections can only retrieve data today. Planning scenarios, saving changes, and triggering actions through an assistant remain limited or on vendor roadmaps. By the end of 2027, Nucleus expects most CPM vendors to support retrieval, so the vendors that stand out will be those that let users save changes safely through an assistant. Value will depend on two conditions. The AI assistant must be able to reach a governed, current planning model with visible assumptions, and permissions and audit controls must apply to every person and agent that uses it. Buyers should hold AI access to the same standard they applied to Excel integrations and judge it by what it can safely do, not by the number of connections a vendor lists. Vendors that prove governed write-back, transparent assumptions, and clear pricing for assistant and agent access will hold their position as the planning model becomes the trusted source of numbers the rest of the business relies on.