Edit

Get customer predictions

Applies to: Dynamics 365 Customer Insights

Note

  • The tools and examples in this documentation illustrate current capabilities. Tool names, descriptions, parameters, response formats, and available functionality can change over time.
  • When you build agents, integrations, or orchestration logic, don't hard-code dependencies on specific tool names, parameter structures, tool metadata, or response schemas. Instead, design your solution to discover and use available tools at runtime.
  • As the platform evolves, Microsoft might add, remove, rename, or modify tools and tool metadata. Designing against capabilities rather than specific tool identifiers helps maintain compatibility across service updates.

Use this capability when you have a customer's unified ID and want the machine-learning scores Customer Insights has computed for them.

What it does

For one unified customer, this capability returns the machine-learning predictions Customer Insights models have scored for them—such as churn probability, predicted lifetime value, propensity, and next-best product suggestions. Each result shows the model that produced it, the model's provider, and the scored value.

Predictions are forward-looking scores. For descriptive KPIs like total purchases or order counts, use measures instead. Resolve the customer ID first with find_customer, get_customer_by_contact, or get_customer_by_source_id. If the customer has never been scored, the assistant returns an empty list.

Try prompts like

  • What's this customer's churn risk?
  • Show the Customer Insights predictions for this customer.
  • How likely is this customer to churn, and what's their predicted lifetime value?
  • What's this customer's propensity score?
  • What products does Customer Insights recommend for this customer?

What you'll see in chat

The assistant replies with a text list of the customer's prediction scores, each with its originating model, the model's provider, and the scored value. If the customer has never been scored, it tells you the list is empty.

Helpful tips

  • Resolve the customer ID first with find_customer, get_customer_by_contact, or get_customer_by_source_id.
  • Predictions are forward-looking scores (churn, predicted lifetime value, propensity)—for descriptive KPIs like total spend, ask for measures.
  • For predictions alongside segments and measures in one view, ask for the full Customer 360.

What happens next

You can keep working with the same customer:

  • What's this customer's total purchases and average order value?
  • Which segments is this customer in?
  • Show me everything Customer Insights knows about this customer.

Does this change data?

No, reading a customer's predictions doesn't change data.

This capability only reads model scores from Customer Insights. It never creates, updates, or deletes records.

Prerequisites

This tool is available on the Dynamics 365 Customer Insights MCP server. See the availability note at the top of this page for details. Customer Insights must be configured for your environment with prediction models scored, and you need permission to read customer predictions.

Tool summary

Property Value
User-facing name Get Customer Predictions
Internal tool name get_customer_predictions
Purpose Returns the machine-learning scores Customer Insights models compute for one unified customer

Tool behavior

Returns the machine-learning scores Customer Insights models compute per unified customer (msdynci_prediction): churn probability, predicted lifetime value, propensity, and next-best product suggestions. Each row carries the originating model, its provider, and a scored numeric result. Use it for forward-looking risk and opportunity figures, not descriptive traits. Returns an empty list when the customer has never been scored.

Annotations

Annotation Value Meaning
readOnlyHint true Read-only; this tool doesn't modify data.
destructiveHint Not set Not applicable (read-only tool).
idempotentHint Not set Not applicable (read-only tool).
openWorldHint Not set Uses default (queries Customer Insights through Dataverse).

Input concepts

Customer

Input Description Required
customerId The Customer Insights unified customer ID (32-character hex). Yes

Response and UI behavior

This tool returns a text response.

This MCP tool is not backed by an app-in-chat component.

Response type

Text list of prediction scores.

Each entry includes the model, its provider, and the scored value; the list is empty when the customer has never been scored.

Routing notes

Use get_customer_predictions for:

  • A request about a known customer's forward-looking risk or opportunity (churn, predicted lifetime value, propensity, recommendations).

Don't use get_customer_predictions when:

  • The prompt hasn't identified a customer—resolve first with find_customer, get_customer_by_contact, or get_customer_by_source_id.
  • The user wants descriptive KPIs (total spend, order counts)—route to get_customer_measures.
  • The user wants the full customer picture—route to get_customer_details.
Tool Relationship
get_customer_measures Returns aggregated KPIs instead of forward-looking scores
get_customer_details Returns predictions alongside the full Customer 360
get_customer_segments Lists the customer's segment memberships
find_customer Resolves the customer ID this tool needs

Data mutation classification

Read-only.

This tool only reads model prediction scores from Customer Insights and never changes data.

  • Last updated on