… 6 unchanged lines …
  - 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
    … 4 unchanged lines …
  
  ## 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: 2026-04-14 | Check this after cases are added as an intent discovery source or intent-based routing expands*
+ *Source last updated: 2026-08-27 | Check this after cases are added as an intent discovery source or intent-based routing expands*