Define one output and exclude the rest
Use one format containing decisions already made, questions still open, deadlines explicitly stated and the person who needs to confirm each item. Exclude reply drafting, mailbox rules, automatic categorisation and automatic follow-up. A single output makes it possible to compare results consistently and identify whether an error came from the prompt, the conversation or the review step.
Choose threads that participants can verify
Select recurring, long conversations whose context normally takes time to rebuild, such as a client scope revision or a cross-team handover. Exclude legal disputes, unusually sensitive material and cases with no informed reviewer. Record the current time-to-context and ask someone familiar with each conversation to establish the expected decisions and open points.
Confirm the working environment before the pilot
Check current licensing, mailbox and client requirements, then confirm that the selected users can access the relevant messages through their normal permissions. Do not expand permissions for the sake of the test. Microsoft 365 Copilot uses the access available to the user, so existing oversharing remains a governance issue that the MVP must expose rather than ignore.
Use a fixed prompt and a mandatory source check
Tell Copilot to use only the email thread and mark anything uncertain as unconfirmed. It must not invent an owner, decision or date. Before a summary enters a work record, the user checks each line against the original messages and available source links. Fluent wording is not evidence that the summary is complete or correct.
Measure time-to-context and correction quality
For every test, record the time from opening the thread to completing review, missing decisions, incorrect details, sentences requiring correction and whether the user would use the summary again. A small number of severe omissions should outweigh a larger number of fast outputs. Compare like-for-like threads instead of reporting a single average across unrelated work.
Make a stop, refine or continue decision after 30 days
Continue only if reviewed summaries reduce time-to-context without crossing the agreed correction threshold. Refine the format if errors cluster around owners, dates or ambiguous decisions. Stop if review takes longer than the original method or participants cannot reliably verify the source. Reply drafting and automated follow-up remain separate MVPs.
Interface and requirements checked 27 August 2026
Current Outlook setup and operating steps
These steps reflect current Outlook for Microsoft 365, new Outlook and Outlook on the web. Depending on the client, the control may read Summary by Copilot or Summarize. Reconfirm the tenant update channel, licence and admin policy before the pilot starts.
-
Confirm the account and mailbox
Verify an eligible Microsoft 365 subscription or Copilot entitlement, a Microsoft Entra ID account and an Exchange Online primary mailbox. Keep archive, group, shared and delegate mailboxes outside this in-product summary MVP.
-
Choose one verifiable conversation
Open a long conversation in the primary mailbox. Exclude S/MIME, Double Key Encryption, unusually sensitive content and any thread for which no informed participant can review the result.
-
Generate the summary at the top of the thread
Select Summary by Copilot or Summarize above the email conversation. Outlook has not moved this feature to the lower-right Dynamic Action Button used by Word and Excel.
-
Return to the messages and citations
Open any numbered citations and verify every decision, request, person and date. Attachment summaries in new Outlook are a navigation aid, not a substitute for reviewing the source file.
-
Save only the reviewed output
Move confirmed decisions, open questions, owners and deadlines into the fixed template, then record review time, corrections and omissions. Unconfirmed text does not become an official record.
View Microsoft Support’s current Outlook steps and official interface examples →
Practical next step
- Free AI & Workflow Project Planner: Define the thread type, output, reviewer and 30-day decision measures.