Microsoft First‑Party Autonomous Agents for Dynamics 365 Customer Service and Sales

Technical overview and configuration guide (October 2025)

Summary Lede:
This article provides a technical overview of Microsoft’s first‑party autonomous agents for Dynamics 365 Customer Service, Contact Center, and Sales. It outlines purpose, scope, and expected outcomes for three audiences—end users (agents and sellers), business managers, and IT administrators—to support fit assessment and operational planning. For each agent (Customer Intent, Case Management, Customer Knowledge Management, and Sales Agent), it documents prerequisites, enablement steps, key configuration choices, and guardrails, including identity, channel authentication, data residency, and Power Platform pay‑as‑you‑go consumption. The guidance assumes a staged rollout in non-production environments and notes the preview status and regional availability where applicable. Integration points with Copilot Studio, Dynamics 365, and Microsoft 365 are identified, along with ALM, monitoring, and KPIs for rollout (e.g., AHT, FCR, and suggestion acceptance).

Agents in scope

• Customer Intent Agent (Customer Service & Contact Center): Discovers, maintains, and applies a library of customer intents from historic and live interactions; supports both self‑service and agent‑assist.
• Case Management Agent (Customer Service): Automates case lifecycle—creation, enrichment, updates, follow‑ups, and optional autonomous resolution.
• Customer Knowledge Management Agent (Customer Service): Autogenerates and maintains knowledge articles from cases and conversations, in real time and from historical data, with analytics.
• Sales Agent (Microsoft 365 Copilot for Sales, works with Dynamics 365 Sales or Salesforce): Researches leads, drafts outreach, follows up, can operate autonomously to qualify and hand off.

Status note Customer Service/Contact Center agents and Sales Agent capabilities referenced here include preview features with staged regional rollout. Please ensure preview features are treated appropriately in production planning.

Abstracts: purpose and benefits by persona

Customer Intent Agent

Purpose for end users (service representatives and supervisors)
The Customer Intent Agent helps service representatives by automatically identifying the intent behind customer messages, reducing the need for manual triage and allowing them to focus on resolution rather than classification. For supervisors, it provides visibility into intent trends and routing accuracy, enabling better workload distribution and process optimization. Overall, it ensures that both roles spend less time on repetitive tasks and more time on high-value activities, while maintaining consistency in how customer requests are handled

Benefit for business managers
Shortens handle time by streamlining the identification and resolution of common contact reasons. This system provides a structured, reusable taxonomy that facilitates efficient routing of inquiries, playbook development, and the creation of knowledge articles. As topics and customer needs evolve, this taxonomy remains relevant and adaptable, ensuring that the information provided to both agents and customers is up-to-date and effective in resolving issues promptly.

Benefit for IT managers
A centrally managed intent library serves as a unified repository for customer intents, which can be integrated with both chat and voice communication channels. This approach ensures the intent library is consistently applied across various interaction points, thereby promoting a standardized approach to intent recognition and response generation. By utilizing a governed rollout strategy, organizations can effectively manage the deployment of this library, ensuring that updates and modifications are systematically controlled and monitored. This eliminates the need for custom natural language processing (NLP) solutions tailored to individual queues, thereby reducing complexity and maintenance overhead. Instead, the centralized library enables a more streamlined and efficient handling of customer interactions, thereby enhancing the overall service delivery process.

Case Management Agent

Purpose for end users
Reduces manual effort for routine case administration tasks by automating several stages of the case lifecycle. The agent creates case records directly from authenticated chat and voice conversations as well as from incoming email messages. It pre-populates case fields by extracting relevant information from the conversation or email context, reducing data entry errors and ensuring consistent field values. At the conversation wrap-up, the agent updates the case record with additional details captured during the interaction, including resolution notes and follow-up requirements. When configured to do so, the agent can also send follow-up communications to customers based on case status or predefined triggers. In environments where autonomous resolution is enabled, the agent drafts resolution emails using case context and available knowledge, and—subject to configured business rules and approval workflows—can send those emails and close the case without human intervention. This automation enables the service representative to shift their focus from administrative overhead to higher-value customer interactions.

Benefit for business managers
Reduces the time service representatives spend on non-value-added administrative tasks by automating routine case lifecycle operations. Representatives can allocate more time to direct customer engagement and complex problem-solving rather than manual data entry and case status updates. The automation improves adherence to service-level agreements by ensuring that case records are created, updated, and progressed through defined stages without delay or manual intervention. Data quality across case records improves because the agent applies consistent field mappings and extraction logic, reducing the incidence of missing required fields, typographical errors, and inconsistent categorization that commonly occur during manual case creation. This standardization also simplifies downstream reporting and analytics, as case data becomes more reliable and complete across the organization.

Benefit for IT managers
Provides a configuration-driven approach to case lifecycle automation that leverages native Dynamics 365 platform capabilities, eliminating the need for custom development. The agent integrates with existing automatic record creation and update (ARC) rules to handle inbound email-to-case scenarios, utilizes authenticated channel infrastructure for chat and voice conversations to ensure proper identity context, and exposes administrative controls through the Customer Service admin center that allow IT teams to enable or disable autonomous behaviors per channel and per line of business. All automated actions — such as case creation, field updates, resolution emails, and case closure—generate audit trail entries in the standard Dynamics 365 audit log, providing full traceability of which operations were performed by the agent versus human users. For scenarios that require programmatic invocation outside the standard channel flows (for example, cases created from custom portals, mobile applications, or external systems writing directly to Dataverse), the agent exposes a business event API endpoint (msdyn_invokeCaseProcessingAgent) that external processes can call via HTTP to trigger case enrichment and resolution logic on demand. The multi-tier automation model allows organizations to increase agent autonomy progressively: in semi-autonomous mode, the agent drafts resolution communications and presents them to a human representative for review and approval before sending; in fully autonomous mode, the agent evaluates case context against configured business rules and, when conditions are satisfied, sends the resolution email and closes the case without human intervention. This staged approach enables IT teams to validate agent behavior, refine field mappings and decision thresholds, and build organizational confidence in a controlled manner before expanding the scope of unattended automation.

Customer Knowledge Management Agent

Purpose for end users
Reduces manual effort required to maintain the knowledge base by automatically generating draft articles from resolved cases, email exchanges, and transcripts of chat or voice conversations. The agent operates in two modes: real-time creation, which triggers article generation when a case is resolved or a conversation ends, and historical creation, which processes batches of past case records to extract patterns and common resolutions that warrant documentation. Service representatives and knowledge authors no longer need to manually extract relevant information from case notes, conversation logs, or email threads to draft new articles or update existing content. Instead, the agent analyzes the source content, identifies key resolution steps and context, applies configured compliance filters (such as personally identifiable information scrubbing), and produces a structured draft article that includes a title, summary, detailed resolution steps, and suggested keywords based on the intent and case attributes. The draft is routed through a configurable review workflow, allowing designated knowledge authors or subject matter experts to approve the article as-is, edit the content before publishing, or split a single draft into multiple focused articles if the source material covers several distinct topics. This workflow ensures that human oversight remains part of the publication process while significantly reducing the time knowledge authors spend on initial content creation and formatting.

Benefit for business managers
Reduces the elapsed time required to create, review, and publish knowledge articles from weeks to hours or days by automating the extraction of resolution information from case records and conversation transcripts. This compression of the knowledge lifecycle directly improves first-contact resolution rates because new solutions and workarounds become available to service representatives and self-service channels more quickly after they are first discovered during customer interactions. The automation also enhances content consistency across channels—chat, voice, email, and portal self-service—because all knowledge articles are generated from the same source material using standardized templates and extraction logic, rather than relying on individual authors to interpret manually and document resolutions in varying formats. Supervisors and knowledge managers gain access to quality and usage insights through built-in analytics dashboards that track key metrics, including article generation volume, review cycle time, publication rate, article view counts, and feedback scores from both service representatives and customers. These insights enable data-driven decisions about which topics require more detailed documentation, which articles should be retired or consolidated, and where knowledge gaps exist in the current library. The feedback loop supports continuous improvement of the knowledge base by identifying patterns in high-performing articles and surfacing opportunities to refine the agent’s extraction logic, compliance filters, and template structure based on actual usage data and representative input.

Benefit for IT managers.
Provides a native Dynamics 365 platform integration that eliminates the requirement for custom knowledge management middleware or third-party content generation tools. The agent operates within the established Dynamics 365 knowledge management framework, writing draft articles directly to the knowledge base entity in Dataverse and respecting all existing role-based security controls, field-level permissions, and record-sharing rules that govern knowledge article access and modification. IT teams configure compliance and data protection requirements through administrative controls that apply organization-wide policies to all generated content: the personally identifiable information (PII) scrubbing feature uses pattern matching and entity recognition to detect and redact sensitive customer data—such as email addresses, phone numbers, account identifiers, and custom-defined patterns—before the draft article is written to storage, ensuring that published knowledge content does not inadvertently expose customer information to service representatives or self-service portal users who lack authorization to view such data. The agent exposes publishing workflow controls that allow administrators to define approval stages, assign review responsibilities to specific security roles or individual knowledge authors, and enforce quality gates before any auto-generated content becomes visible in knowledge search results or is surfaced to customer-facing channels. Capacity planning is supported through consumption telemetry available in the Power Platform admin center, where IT teams can monitor the number of articles generated per day, the average processing time per draft, and the total message units consumed by the agent’s language model calls; this data enables organizations to forecast monthly costs under the pay-as-you-go model and set budget alerts when usage approaches defined thresholds. For environments that require strict control over when and how the agent processes historical case data, administrators can configure batch processing schedules that limit historical article creation to off-peak hours, define maximum batch sizes to avoid overwhelming the review queue, and pause or resume historical processing jobs through the admin center without requiring developer intervention or custom workflow modifications.

Sales Agent (Copilot for Sales)

Purpose for end users (sellers & sales ops)
Reduces the manual effort required to research, contact, and qualify early-stage sales opportunities by automating several stages of the lead engagement process. The agent retrieves lead records from the connected customer relationship management (CRM) system—either Dynamics 365 Sales or Salesforce—and supplements that data with publicly available information from web sources such as company websites, press releases, and professional networking platforms to build a contextual profile of each prospect’s business environment, recent organizational changes, and potential pain points relevant to the organization’s product or service offerings. Using this combined dataset, the agent drafts initial outreach messages tailored to the specific lead’s industry, role, and inferred needs, applying the tone, messaging framework, and value propositions defined during agent configuration. When a lead responds to the initial email, the agent evaluates the content of the reply to determine whether the prospect has asked a factual question that can be answered from the configured knowledge base—such as product specifications, pricing tier availability, or implementation timelines—or whether the response indicates interest that requires human follow-up. For straightforward informational questions, the agent drafts a reply using the knowledge sources provided during setup. Depending on the configured autonomy level, the agent either presents the draft to the assigned seller for review and approval or sends the response directly. If the reply signals qualified interest—such as a request for a product demonstration, a pricing proposal, or a discussion of specific business requirements—the agent updates the lead record in the CRM system with qualification notes and interaction history, and either notifies the assigned seller that the lead is ready for human engagement or, when operating in fully autonomous mode, schedules a follow-up task and hands off the opportunity according to predefined lead scoring and assignment rules. In scenarios where a lead does not respond to the initial outreach within a configured time window, the agent executes a follow-up sequence based on the cadence and messaging rules defined during setup, sending subsequent emails at scheduled intervals until the lead responds, opts out, or reaches the maximum number of touches allowed by the organization’s sales engagement policy. This automation will enable sales teams to maintain consistent and timely contact with a larger volume of early-stage leads—particularly those in the long-tail segment that might otherwise receive delayed or no follow-up due to capacity constraints—without requiring sellers to research each prospect manually, draft individualized messages, or track follow-up schedules across hundreds of open opportunities.

Benefit for business managers
Addresses the operational challenge of limited sales capacity to engage early-stage leads by automating research, outreach, and initial qualification activities that would otherwise require manual effort from quota-carrying sellers. Organizations often face a coverage gap, where the volume of inbound leads and marketing-qualified accounts exceeds the sales team’s capacity to make timely first contact, particularly for leads in lower-priority segments or those requiring extended nurture cycles before they reach active buying stages. The agent closes this gap by maintaining a consistent contact cadence across the entire lead inventory, ensuring that every prospect receives at least a baseline level of engagement—initial research, personalized outreach, and automated follow-up—regardless of whether a human seller has the capacity to handle that lead immediately. This consistent engagement model reduces the incidence of leads aging out of service-level agreement windows without contact, decreases the time-to-first-touch metric across the pipeline, and prevents qualified opportunities from being lost to competitors simply because they were not contacted promptly. The automation also standardizes the quality and messaging framework applied during early-stage interactions: all outreach follows the same research methodology, applies the same value proposition templates, and executes the same multi-touch follow-up sequence, eliminating the variability that occurs when different sellers use inconsistent approaches or when high-volume periods force sellers to skip research steps or reduce message personalization. Business managers gain visibility into the effectiveness of this standardized engagement through telemetry dashboards that track metrics such as email open rates, response rates by lead source and industry segment, the average number of touches required to elicit a response, the percentage of engaged leads that meet qualification criteria for sales handoff, and—most critically—the conversion rate uplift measured by comparing leads that received agent-driven engagement against control groups that received manual outreach or no outreach at all. This data enables sales operations teams to refine outreach messaging, adjust follow-up cadence based on observed response patterns, identify which lead segments benefit most from automated engagement versus immediate human contact, and quantify the incremental pipeline value generated by the agent’s activities, supporting return-on-investment analysis and capacity planning decisions without relying on anecdotal evidence or broad assumptions about engagement effectiveness.

Benefit for IT managers
Provides a first-party integration that connects directly to the organization’s customer relationship management system—either Dynamics 365 Sales or Salesforce—using the same authentication, data access controls, and API infrastructure that Microsoft 365 Copilot for Sales already employs for read and write operations against CRM records. This integration model eliminates the need for IT teams to deploy and maintain separate middleware components, custom API gateways, or third-party lead engagement platforms that would otherwise be required to enable automated research, outreach, and follow-up activities across the lead lifecycle. Because the agent operates as a component of Microsoft 365 Copilot for Sales rather than as an external service, all CRM data access is governed by the existing role-based security policies, field-level permissions, and record-sharing rules configured in the connected CRM system; the agent inherits the data access scope of the user context under which it operates, ensuring that it cannot read lead records, contact information, or account data that would be restricted to a human seller in the same role. The agent’s behavior—including which knowledge sources it consults to answer prospect questions, which messaging templates and value propositions it applies during outreach, the tone and language style it uses in drafted emails, the follow-up cadence and maximum number of touches it executes before halting engagement, and the lead scoring thresholds that determine when a lead should be handed off to a human seller—is configured through administrative controls exposed in the Copilot for Sales interface rather than requiring custom code, scripting, or workflow definitions. This configuration-driven approach allows sales operations teams and IT administrators to define and modify agent behavior without developer involvement, using structured input fields, dropdown menus, and template editors to specify guardrails such as prohibited language patterns, mandatory compliance disclosures, opt-out handling procedures, and escalation triggers for scenarios where the agent encounters questions outside its configured knowledge scope or detects negative sentiment in a prospect’s reply. All agent activities—research queries executed against web sources, emails drafted and sent to leads, updates written to CRM records, and decisions to hand off leads to human sellers—are logged in the Microsoft 365 audit trail and the connected CRM system’s activity history, providing IT teams and compliance officers with a complete record of which actions were performed autonomously versus which required human approval, along with timestamps, user context, and the specific knowledge sources or decision rules that informed each action. The agent operates within the Microsoft 365 security and compliance boundary, meaning that all data processed during research, outreach drafting, and lead qualification—including prospect contact information, conversation transcripts, and any customer data retrieved from the CRM system—is subject to the same data loss prevention policies, information protection labels, retention rules, and cross-border data transfer controls that apply to other Microsoft 365 workloads such as Exchange Online, SharePoint, and Teams. This operational model ensures that organizations can apply consistent governance frameworks across all automated lead engagement activities, without needing to define separate policies for agent-driven versus human-driven sales processes. It enables IT teams to leverage existing investments in Microsoft Purview, Defender, and Entra ID, rather than provisioning additional compliance tooling or identity infrastructure, to secure and monitor agent operations.

Activation & configuration (environment‑centric)

Pre‑flight for all agents (recommended)

• Use a non‑production environment first; plan Dev/Test/Prod with solution‑aware ALM for customizations.
• Confirm data residency and compliance allowances for generative AI features in your tenant.
• Where required, link an Azure subscription for Power Platform pay‑as‑you‑go consumption and consider the setting to move data across regions for copilots/generative features.
• Use the preview Agent hub to plan and monitor adoption with rollout plans and insights.

Customer Intent Agent

Prerequisites

• Dynamics 365 Customer Service or Dynamics 365 Contact Center environment.
• Channel setup for targeted channels (chat/voice) if using real‑time conversation capabilities.

Enable & configure

• Open the admin experience and locate Customer Intent Agent.
• Seed the intent library from historical cases and conversations; review discovered intents and attributes.
• Self‑service integration: connect Copilot Studio customer‑facing agents to the intent library so the agent can detect intents, ask follow‑ups (slot filling), and query knowledge.
• Assisted service integration: enable intent‑based suggestions in the agent desktop so reps get the following questions/solutions and see intent changes mid‑conversation.
• Govern rollout: use rollout plans in Agent hub to scope to selected intents/queues; monitor handle time and suggestion acceptance.

Case Management Agent

Prerequisites

• Dynamics 365 Customer Service.
• For autonomous creation from conversations: authenticated chat/voice, workstreams, and queues configured. For email, automatic record creation and update (ARC) rules. Enable AI form‑fill assistance and, if applicable, consumption billing.

Enable & configure (creation & update)

• In Copilot Service admin center → Case settings → Case Management Agent (Manage), enable Autonomous case creation (per channel) and Autonomous case updates. Map conversation/email context to target case fields.
• Configure field mappings, prediction targets, and thresholds; validate on representative chats/emails.

Enable & configure (resolution)

• In Global settings for Case Management Agent, set the Application user for outbound email, default templates, and usage of Copilot template recommendations.
• Per line of business, set automation level:
  • Semi‑autonomous: draft resolution emails for human approval.
  • Full: AI agent sends emails and closes the case when conditions are satisfied.

Integration & extension

• Trigger enrichment/resolution programmatically via the business event msdyn_invokeCaseProcessingAgent (HTTP) to handle custom scenarios (e.g., portal cases or custom tables).

Customer Knowledge Management Agent

Prerequisites

• Dynamics 365 knowledge management configured; Copilot access to internal knowledge base enabled. Connection references for the real-time flow are configured as needed.

Enable & configure

• In Copilot Service admin center → Support experience → Knowledge → Customer Knowledge Management Agent (preview) → Manage. Enable Real‑time creation for Cases and/or Conversations (triggered on case resolution or chat close).
• Define scoping rules (which resolved cases/conversations trigger article creation).
• Enable Historical creation to mine past cases; configure compliance filters (e.g., PII scrubbing) and publishing behavior.
• Set up review workflows (approve/edit/split & publish) and enable knowledge insights dashboards to track quality and usage.

Sales Agent (Microsoft 365 Copilot for Sales)

Prerequisites

• Copilot for Sales enabled and connected to Dynamics 365 Sales or Salesforce.

Enable & configure

• Access Sales Agent in Copilot for Sales; confirm CRM connection and lead sources.
• Provide knowledge (company/product value propositions, pricing sheets) and define behavior (tone, follow‑up cadence, handover criteria).
• Start in assistive mode (seller in the loop) to validate research quality and email drafts, then agent autonomous engagement for long‑tail leads with strict guardrails.

Operational notes. Treat autonomous outreach like any outbound automation: enforce opt‑out, use suppression lists, align DKIM/SPF, and maintain CRM hygiene.

Governance, security, and rollout guidance

• Security & compliance Review security/privacy guidance and ensure mail‑sending identities and audit trails are correctly configured before enabling autonomous emails or record updates.
• Consumption & cost Some agent operations bill via Power Platform pay‑as‑you‑go—link an Azure subscription and monitor usage.
• Rollout discipline Use rollout plans to scope by intent, queue, or team; track KPIs (AHT, FCR, suggestion acceptance rate, auto‑resolution success)—iterate mappings and thresholds.
• Channels and authentication For autonomous creation from conversations, enable authenticated chat/voice and transcription (for voice) as required.
• ALM Manage configurations as solutions where possible; document mappings, templates, and automation levels per line of business.

Pilot recipes you can implement this week

• Accelerate case admin (no code). Enable Case Management Agent for autonomous updates only; keep creation/resolution manual. Map the 5–8 most error‑prone fields; audit predictions on a 50‑conversation/email sample.

• Seed an intent library for peak drivers. Turn on Customer Intent Agent; ingest 3–6 months of cases/chats; review top 20 intents and attributes. Roll out to one queue with Agent Hub controls.

• Evergreen knowledge on a hot topic. Enable Customer Knowledge Management Agent (real‑time), scoped to one product and an “outage” intent; route drafts to a single reviewer; monitor knowledge insights.

• Autonomous long‑tail lead nurture. Configure Sales Agent with strict guardrails to engage unassigned leads from a CSV, hand off to Sales Agent on positive replies, and monitor conversions.

Troubleshooting & operational guardrails

• Agent sends emails from the wrong identity. Verify the Application user in Case Management Agent global settings and mailbox/Server‑Side Sync configuration.
• No case created from chat. Confirm authenticated chat and that the mapped fields have sufficient context; review workstream/queue configuration.
• Duplicate knowledge drafts. Enable duplicate detection, tune similarity thresholds, and ensure historical creation rules don’t overlap.
• Intent suggestions seem off. Re‑run discovery with a broader sample, prune low‑quality transcripts, and re‑label a few seed examples; then re‑roll out with controls.

Link list — next steps & documentation

Customer Service & Contact Center (autonomous agents)

• Autonomous service agents (overview)
• Agent hub (preview)
• Customer Intent Agent (overview & setup)
• Case Management Agent – set up creation/update
• Case Management Agent – use creation/update
• Case Management Agent – set up resolution
• Case Management Agent – developer trigger (msdyn_invokeCaseProces"ingAge"t)
• Customer Knowledge Management Agent – manage
• Customer Knowledge Management Agent – knowledge insights
• Customer Knowledge Management Agent – review drafts
• Release plan entries (timelines):
  • Case Management Agent
  • Customer Knowledge Management Agent

Sales

• Sales Agent (overview/preview)
• Microsoft announcement (access & context)
• Release plan note (autonomous Sales Agent)

Copilot Studio

• Build an autonomous agent (training)
• Copilot Studio agent capabilities (context)

Conclusion

Microsoft’s first-party autonomous agents for Dynamics 365 Customer Service, Contact Center, and Sales represent a shift from assistive AI to task-completion automation within the enterprise service and sales lifecycle. The four agents covered—Customer Intent, Case Management, Customer Knowledge Management, and Sales Agent—address different operational bottlenecks: intent discovery and taxonomy maintenance, case lifecycle administration, knowledge base currency, and early-stage lead engagement. Each agent operates within the existing Dynamics 365 and Microsoft 365 security, compliance, and data governance boundaries, eliminating the need for external middleware or custom NLP infrastructure while respecting role-based access controls and audit requirements.

Successful adoption requires a staged approach. Start in non-production environments to validate field mappings, decision thresholds, and knowledge source quality before enabling autonomous behaviors in production. Use the Agent hub and rollout plans to scope initial deployments to controlled populations—specific queues, intents, or lead segments—and establish baseline KPIs such as average handle time, first-contact resolution, suggestion acceptance rate, and auto-resolution success. Monitor consumption telemetry through the Power Platform admin center to forecast costs under the pay-as-you-go model and adjust batch processing schedules or autonomy levels as needed.

Governance discipline is critical. Define clear policies for autonomous email sending, including application user identity, DKIM/SPF alignment, opt-out handling, and enforcement of suppression lists. Establish review workflows for knowledge article publication and outbound sales messaging to maintain brand consistency and regulatory compliance. Document configuration choices—field mappings, templates, automation levels per line of business, and integration points with Copilot Studio or external systems—as solution-aware components to support application lifecycle management across Dev/Test/Prod environments.

The pilot recipes provided offer practical entry points: accelerate case administration without touching creation or resolution, seed an intent library for your top contact drivers, automate knowledge article generation for a single product category, or engage long-tail leads that would otherwise receive no contact. Each scenario can be implemented within a week and produces measurable outcomes that inform broader rollout decisions.

As these agents move from preview to general availability and expand regional coverage, monitor the release plans for new capabilities, additional channel support, and integration enhancements. Treat autonomous agents as long-term investments in operational leverage: the initial configuration effort establishes the foundation for continuous improvement cycles driven by usage data, knowledge insights, and feedback from service representatives and sellers.