Almanac
Microsoft/d365csDynamics 365

Consultant-focused KB for Microsoft Dynamics 365 Customer Service: implementation notes, gotchas, and configuration decisions beyond the official docs — across case management, queues and routing, knowledge base, SLAs, omnichannel, agent experience, Copilot, analytics, and administration.

feature-copilot-case-conversation-summaries.mdv1 · history
CurrentApplies to EnterpriseUpdated 6 days agoSource Microsoft Learn

What it does

Copilot reads a case record and its linked activity, then writes a short narrative of what the customer's problem is and what has happened so far. The same engine produces conversation summaries for chat and voice sessions, generated when a rep joins, when the conversation closes, or on demand.

Key facts

  • Two separate features with two separate switches. Case summaries are Customer Service only; conversation summaries also exist in Contact Center. Both are configured under Summaries in the Copilot Service admin center, but you turn each on individually.
  • Case summaries need Ask a question enabled first under Copilot for questions and emails. Turning Summaries on without that gets you nothing.
  • The case summary draws on case title, customer, subject, product, priority, case type and description, plus linked email activities, notes and any conversation summaries attached to the case.
  • There is a minimum content threshold: roughly 50 tokens, which Microsoft describes as about 38 English words across the source fields excluding spaces. Below that, no summary is generated and the card sits empty.
  • The card renders collapsed by default on the case form in Copilot Service workspace, and its position follows the form layout. Custom case forms need extra work to show it.
  • Summaries can be generated for active, resolved and cancelled cases. Reps can copy, regenerate and rate the output.
  • Related record summaries show two related records by default, with a Show more related records control loading two more at a time.
  • Output format is an admin choice: paragraph or structured. Conversation summary format is configured separately under Manage format.
  • You can exclude noise from summaries — up to 10 email addresses and 3 disclaimers, headers or footers.
  • Regional availability differs between the two. Case summarisation is available across a wide list of regions; conversation summarisation is restricted to North America, Europe and the UK. Outside those, you're enabling cross-region data movement or you're not getting it.
  • AI agent (bot) portions of a conversation are not automatically folded into the conversation summary.

When to use / skip

Turn case summaries on. It is the single cheapest Copilot win in the product and the one that survives contact with a real service desk, because reps genuinely do inherit cases with forty timeline entries and no context. The caveat is that it only works if your case data is worth summarising: if your reps put everything in a two-word case title and resolve without notes, you'll demo a blank card.

Conversation summaries are worth it where you have transfers, consults or supervisors picking up escalations mid-conversation. If your omnichannel volume is low or conversations are short and single-touch, the value is thin and you're paying regional-availability tax for it. Skip the "create a case from the summary" option unless case creation from conversation is genuinely part of the process — it produces cases nobody owns.

Configuration decisions

  • Which of the three conversation summary triggers you enable: on join/transfer/consult, on conversation end, on demand — or some combination. Enabling all three is the default temptation and produces the most noise.
  • Paragraph versus structured format, per feature. Structured tends to read better for triage; paragraph reads better for handover.
  • Which case fields feed the summary, via Manage data. Custom fields carrying the real problem description need adding explicitly.
  • What to exclude: the standard legal footer and the shared mailbox address will otherwise appear in every summary.
  • Whether case summaries appear on custom case forms, and who does that work — it is not automatic.
  • Whether to allow case creation directly from a conversation summary, and which queue those cases land in.

Gotchas

  • The 38-word threshold catches almost every pilot. A test case with title "Test" and description "test case" will never produce a summary, and people conclude the feature is broken.
  • Conversation summary region support is narrower than case summary region support. Check both against your Dataverse region before you promise either, especially for APAC or Canada deployments.
  • Custom case forms silently lose the summary card. If the client's main case form is custom — it usually is — budget for that step.
  • The summary is generated on demand and not stored as a durable record on the case. If someone wants "the summary at the point of closure" as an audit artefact, that's a different conversation.
  • Bot-handled portions of a conversation don't automatically make it into the summary, so an escalated conversation can summarise as though it started at the human handover.
  • Turning on transcript recording of rep interactions with Copilot is a separate checkbox with its own data-retention implications. It's off by default for a reason; get it past the client's privacy people before flipping it.

Consultant notes

  • Demo it on real, messy data from the client's own migration set. A summary of a well-formed case is unimpressive; a summary of a 30-email escalation is what sells it.
  • Before go-live, check the exclusion list against the client's actual email signature and disclaimer text. Reps notice the legal boilerplate in the summary immediately and it undermines trust in the whole feature.
  • Push back on requests to "make the summary always accurate". Set the expectation that it's a reading aid, not a record, and that reps stay responsible for what they act on.
  • If the client wants summaries as part of a compliance or QA process, tell them plainly that on-demand generation is not an audit trail and steer them to the Quality Evaluation Agent instead.
  • Check the security role plumbing early. Custom rep roles missing the Copilot interaction privileges produce a summary card that fails without a useful error.

Worth revisiting if conversation summary region coverage widens, or if Microsoft changes the minimum-content threshold.

Was this accurate?