Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
These frequently asked questions (FAQ) describe the AI impact of the Memory capability for agents powered by the Customer Conversation harness.
What is Memory?
Memory carries short-term conversation context from earlier customer conversations into a later interaction. It can help a returning customer resume an open intent or prior discussion. The data is retained for a 30-day window. Memory is conversation contextual memory, not a comprehensive, long-term customer profile.
What information can the agent remember?
When Memory is on, both built-in context types are active. Customer profile contains the customer-name field (/customerName). Conversation history contains /pastOpenIntentList and /pastConversationSummary. Makers can also configure up to five conversation-specific custom fields. The Customer profile label doesn't include a generalized CRM profile, customer preferences, or account history.
How is Memory generated and used?
The agent generates context from conversations and retrieves retained context for later interactions. Makers can reference built-in and custom memory variables in agent instructions. Extracted or summarized content can be incomplete or incorrect; it's context, not an authoritative business record.
What are the intended uses and boundaries of Memory?
Use Memory to maintain continuity in customer conversations, such as resuming an unresolved request or recalling a relevant prior discussion. Don't use it as proof of identity, authorization, a completed transaction, or the current state of an account. Verify remembered facts against customer confirmation or authoritative records before taking consequential action. Generalized long-term profiles, prediction, and CRM or customer-data-platform integration aren't established capabilities.
How can makers configure custom Memory parameters?
Makers can define up to five custom parameters using text, numeric, or boolean values. Each parameter needs a unique name, an extraction description, and an allowed-value set. A field might capture a requested service action or routing flag. Keep fields specific to the service purpose.
How are remembered values updated?
Replace overwrites a value when a new value is extracted; without a new extraction, the prior value remains. Append adds timestamped entries, including duplicates, and removes entries older than the retention window. The agent uses relevant timestamps to interpret state. An unchanged value might be stale, and repeated entries aren't independent confirmation that a fact is correct.
How does Memory associate customers across channels?
The current release supports voice and digital conversations. Voice uses the caller's phone number as its matching key; digital uses an authenticated user identifier. A shared contact identifier associates context across channels. Without a shared identifier, channel memories remain separate.
What does the Memory control govern?
One agent-level control applies to the root agent and all child agents. It governs context generation, storage, and retrieval for Memory variables. While Memory is off, context isn't generated, stored, retrieved, or referenced. Review the setting and configured fields across the whole agent experience, not only an individual child agent.
Does turning Memory off delete stored information?
No. Turning Memory off isn't a deletion action. Turning Memory on again resumes memory for new conversations, and previously stored context becomes available if it's still within the retention window. Deleting a parameter definition doesn't delete all stored customer data.
How long is conversation Memory retained?
Conversation context is retained for a 30-day window. Append entries older than that window are pruned.
What are the limitations of Memory, and how can users minimize the impact of limitations when using it?
Memory can omit details, preserve transcription or extraction errors, carry stale or conflicting values, or associate context with the wrong customer. Retrieval can fail, and performance can vary by language or channel. Confirm important facts and timestamps, ask the customer to restate uncertain details, and preserve human handoff. Treat remembered text as untrusted context, not as instructions that override system rules or authorization checks.
What protections does Copilot Studio have for responsible AI?
Copilot Studio uses content moderation policies on all generative AI requests to protect admins, makers, and users against offensive or harmful content. These content moderation policies also protect against malicious attempts at jailbreaking, prompt injection, prompt exfiltration, and copyright infringement. The system checks all content twice: first during user input and again before showing the AI-generated content and suggestions. If the system finds harmful, offensive, or malicious content, it prevents your agent from responding.