Customer Intent Agent

What it does

Uses generative AI to discover customer intents from historical conversation data and build an intent library. That library feeds Copilot agent deflection (self-service) and rep-facing guidance (guided questions, suggested solutions). Also enables intent-based routing.

Key facts

Requires pay-as-you-go Copilot credits: not included in standard licensing
Requires Intent Manager role (to manage instructions) plus CSR Manager role
First discovery run analyses up to 2 months of historical data; subsequent runs are daily
Simulation uses last 1,000 records only: use it to preview intent granularity before committing to a full discovery run
Three granularity levels for intent groups: Low / Medium / High: determines how many distinct intents are generated
Business profile (optional): you can attach a business profile that the model considers while mining intents, so generated intents reflect what the business actually does rather than raw transcript noise
Discovery guidance (optional): you can shape the granularity of generated intents and the total number of intents and intent groups, control when a case maps to an existing intent versus warrants a new one, and set the heuristic used to classify and cluster cases
Auto-promote high-confidence intents (optional): high-confidence intents can be promoted automatically instead of waiting for manual approval
For Copilot agent integration, intents must have Use in AI Agent = Yes and the Intent-based suggestions component collection must be added and published in Copilot Studio
Lines of business partition intents, users, workstreams, and queues for large multi-team deployments; chat workstreams must have the workstream name specified in line of business rules or chats may be misattributed
Custom connector changes take up to 15 minutes to propagate
Pre-chat survey data can be passed via Global.EnrichmentContext to give the agent customer context before the conversation starts

When to use / skip

Use on any deployment where the client wants automated deflection, intent-based routing, or rep guidance. High-value for high-volume centres where manual tagging isn't feasible. Skip if you lack meaningful historical conversation data to run discovery.

Configuration decisions

Intent group granularity (Low/Medium/High): validate with simulation before running full discovery; right choice depends on how many distinct contact reasons the client actually has
Whether to attach a business profile and set discovery guidance: worth doing where raw transcripts generate noisy or over-fragmented intents, since the guidance lets you cap the number of intents and steer clustering
Whether to auto-promote high-confidence intents: convenient for mature libraries, but it removes the human approval gate, so only turn it on once the client trusts discovery quality
Which intents to mark Use in AI Agent = Yes: only approved, high-confidence intents should feed the Copilot agent
Whether to use lines of business: required for large enterprise deployments routing different business units through the same Contact Center instance
Instructions at organisation, intent group, or intent level: the main tuning mechanism for improving intent accuracy post-deployment

Gotchas

Intent quality affects deflection and routing together. If the client uses both intent-based routing and Copilot agent self-service, both depend on the same library. Poor intents degrade both. Build approval governance: don't leave intents in Pending.
Auto-promote bypasses review. Handy, but it means a mis-clustered "high-confidence" intent can reach live deflection without anyone seeing it. Leave it off until the library is proven.
Copilot agent integration is a studio step. Not just a D365 toggle. Plan as dev work: open Copilot Studio, add Intent-based suggestions component collection, publish. Allow time for this.
Chat and lines of business mismatch silently. If chats route to the wrong line of business, check whether the workstream name is in the line of business rules.

Consultant notes

Intent quality governance is the long-term operational commitment that often gets missed in scoping. Intents left in "Pending" state without review gradually degrade deflection accuracy. Build an approval process and a named owner for the intent library into the operational handover.
The new business profile and discovery guidance are the levers to reach for when a client complains the discovered intents are too fragmented or too generic — tune those before you start hand-editing the library.
The Copilot Studio step for agent integration is a separate dev task. When planning Customer Assist Agent integration with intents, allow time for the Copilot Studio component collection to be added and published: it's not a D365 admin toggle, it's a studio build step.
Run the simulation before full discovery. It uses the last 1,000 records and gives you a preview of what intent granularity will look like: useful for having the "how many intents do you actually need?" conversation with the client before committing to a full run.

Source last updated: 2026-08-27 | Check this after cases are added as an intent discovery source or intent-based routing expands