Table of Content
Microsoft continues to evolve Copilot Studio at a rapid pace. The October 2025 wave of updates focuses heavily on three themes that matter to enterprises:
- Higher agent quality and reliability
- Deeper integration with existing systems and data
- Better governance, analytics, and ROI visibility
In this article, we walk through the newest Copilot Studio capabilities and explain what they mean for your organization—from both an IT and business perspective. Throughout, you’ll find Impactory insights on how we help enterprises implement these features in a structured, low-risk way.
1. Agent Quality & Reliability
Automated Evaluations for Copilot Studio Agents
One of the most impactful updates this month is the introduction of automated agent evaluations in public preview. Instead of manually testing agents one conversation at a time, makers can now:
- Build evaluation sets from:
- Uploaded question–answer files
- Recent Test Pane queries
- Manually added cases
- AI-generated queries based on the agent’s topics and metadata
- Run these test sets directly from the agent or the Test Pane
- Review structured pass/fail results, detailed scoring, and drill down into which knowledge sources and topics were used
Under the hood, evaluations are powered by a flexible grader framework. Makers can choose from:
- Strict checks: Exact Match, Partial/Contains
- Semantic checks: Similarity, Intent Match
- AI-powered metrics: Relevance, completeness, groundedness
Where needed, expected answers can be defined manually or uploaded in bulk, ensuring that business-critical flows are validated against clearly defined standards.
What this means for your organization
- Systematic, repeatable testing rather than ad-hoc manual checks
- Clear quality baselines before going live with new agents
- Easier regression testing after content or model changes
- Earlier detection of knowledge gaps and misaligned responses
As multi-turn testing and additional graders roll out, this will become the backbone for continuous quality assurance of your AI agents.
Impactory Insight: Turn Evaluations into a Quality Framework, Not Just a Feature
Many enterprises underuse evaluation features because they are treated as a “nice-to-have” test tool rather than a formal QA process.
When we work with clients, we typically:
- Define quality criteria per use case
- For support agents: resolution accuracy, groundedness, tone compliance
- For internal assistants: completeness, privacy safeguards, escalation behavior
- Map those criteria to grader configurations
- Exact Match / Partial Match for transactional intents (password reset, order status)
- Semantic similarity and AI metrics for complex, generative answers
- Create reusable evaluation sets
- A baseline set per agent
- Additional sets per critical workflow (e.g., HR policies, finance approvals, IT troubleshooting)
- Integrate evaluations into change management
- Every major content, model, or connector change re-runs a defined evaluation suite
- Results become part of a release checklist
This transforms evaluations from “someone testing in a pane” into a proper quality gate.
Example Enterprise Scenario
A global customer service organization introducing Copilot Studio for L1 support creates:
- A core evaluation set of 200 high-volume questions
- Separate sets for billing, technical support, and account changes
Before each new deployment, these sets run automatically. Only if the pass rate exceeds an agreed threshold (e.g., 92%) is the updated agent promoted to production. This reduces the risk of regressions and unexpected behavior after content changes or model updates.
2. Model Performance & Choice
GPT-4.1 as Default and GPT-5 Models in Preview
Model selection is a strategic decision for every enterprise AI program. The October update brings two important changes:
- GPT-4.1 becomes the default model for all new agents from October 27, 2025
- Replaces GPT-4o as the default
- Shows meaningful improvements in both latency and response quality
- Existing agents can continue using GPT-4o until November 26, 2025, but can be upgraded manually
- GPT-5 family becomes available in more scenarios (public preview)
- GPT-5 Auto
- GPT-5 Chat
- GPT-5 Reasoning
These models can now be used not only in test environments but also in deployed agents, with the caveat that they are still in preview and not recommended for production-critical use yet.
What this means for your organization
- You gain access to faster and more capable models without losing continuity for existing agents.
- You can start exploring advanced reasoning and richer dialogue via GPT-5 models in low-risk scenarios.
- You need a model strategy, not just a “flip the switch to the newest thing” approach.
Impactory Insight: A Structured Model Strategy, Not Trial-and-Error
In practice, we recommend a tiered model selection framework:
- Stability Tier (Production-Critical)
- Use GA models (like GPT-4.1) for customer-facing and compliance-sensitive scenarios.
- Implement change controls: no model upgrades without evaluations and stakeholder approval.
- Innovation Tier (Pilot & Internal)
- Use preview models (GPT-5 family) for internal copilots, prototypes, or low-risk workflows.
- Test what GPT-5 Reasoning can unlock for complex multi-step tasks or knowledge synthesis.
- Experimentation Tier (Labs)
- Short experiments, A/B tests, and proof-of-concepts.
- Explore performance differences across models for your specific domain (e.g., logistics, healthcare, manufacturing).
We usually combine this with evaluation sets so that model changes are always backed by concrete, measurable improvements instead of subjective impressions.
Example Enterprise Scenario
A European financial services provider:
- Keeps customer support agents on GPT-4.1 to ensure stability and predictable behavior
- Pilots GPT-5 Reasoning for internal regulatory Q&A uses, where the benefit of deeper reasoning is high but the risk is lower
- Uses automated evaluations to compare GPT-4.1 and GPT-5 on complex policy interpretation tasks before making any broader rollout decision
3. Operational Efficiency & Execution Speed
Faster Agent Flows with Express Mode
Copilot Studio introduces Express mode (preview) for agent flows, designed to reduce execution time and avoid timeouts in flows that must respond within two minutes.
With Express mode enabled:
- Copilot Studio optimizes flows to be leaner and faster
- Flows are limited to:
- Under 100 actions
- Smaller payload sizes
- Best suited for logic-heavy, data-light scenarios
For flows that move large datasets or iterate over big arrays, makers should test both modes and evaluate trade-offs. Express mode is in public preview and is enabled by default, with a toggle available on the flow’s Overview page.
What this means for your organization
- Reduced risk of agent or app timeouts in interactive scenarios
- Better user experience where responses must feel near real-time
- A clearer separation between “chat-time logic” and “back-end batch processing”
Impactory Insight: Design Flows for Conversation, Not for ETL
We often see organizations overloading conversational flows with:
- Heavy data processing
- Large loops
- Reporting or export logic
Those operations are better handled by background workflows or downstream systems.
By “not for ETL”, we mean that conversational agent flows should avoid heavy data processing tasks such as moving large datasets, iterating through large arrays, or performing complex data transformations. These operations belong in dedicated data or integration pipelines, not in real-time conversational flows where speed, reliability, and responsiveness are critical.
Our recommendation:
- Use Express mode for:
- Decision trees
- Light lookups
- Orchestration of APIs for a single user question
- Keep heavy data operations in:
- Background flows
- Batch jobs
- Dedicated integration platforms
This separation yields more reliable user experiences and cleaner system architectures.
4. Richer User Interactions & Omnichannel CX
File Uploads for Custom Omnichannel Agents
Omnichannel CX refers to delivering a consistent and seamless customer experience across multiple channels—such as web chat, mobile applications, Teams, SMS, and contact center interfaces.
In Copilot Studio, omnichannel agents allow users to interact with the same AI assistant wherever they are, ensuring continuity, consistent answers, and higher service quality across the entire customer journey.
Copilot Studio now supports file uploads for custom agents in omnichannel scenarios. End users can upload:
- Images
- Documents
- Other supported file types (aligned with Microsoft 365 Copilot file support, up to 5MB unless restricted by admins)
These files can then be analyzed in real time, enabling richer use cases such as:
- Customers sharing receipts or invoices for claims
- Employees uploading forms, contracts, or screenshots during support interactions
- Agents performing document-based reasoning (e.g., summarizing, extracting key fields, validating entries)
File uploads are enabled by default, and makers can refine supported file types in the agent manifest.
What this means for your organization
- More context-rich interactions, especially in customer service and internal support
- Faster resolutions, as users can show, not just describe, their issue
- A natural bridge between your existing document processes and AI-driven assistance
Impactory Insight: Treat Files as Data, Not Just Attachments
From an enterprise lens, file uploads are not a minor convenience feature—they’re a data and governance topic. When we support clients, we focus on:
- What kinds of files are allowed (e.g., PDFs, images, Office docs)
- Where files are stored and for how long
- Which agents are allowed to see or process which file types
- How sensitive information (PII, financial data, health data) is handled or redacted
We also help design flows that do something meaningful with the files, such as:
- Automatically extracting key fields from an invoice
- Checking a contract against a standard template
- Analysing a screenshot for typical error patterns
- Summarizing multi-page documents into structured answers
This is where file uploads move from “nice feature” to operational process improvement.
5. System Integration & Real-Time Data Use
Model Context Protocol (MCP) Resources
Copilot Studio has already supported MCP tools, which let agents call external systems to retrieve information or perform actions. With this update, Copilot Studio now also supports MCP resources in preview.
MCP resources allow agents to:
- Read external content such as:
- Files
- API responses
- Database records
- Treat them as file-like data objects that can be referenced in conversations
In practice, this means agents can:
- Access customer-specific data securely and contextually
- Read the latest policy documents, not a stale snapshot
- Use up-to-date API-based information (inventory, shipment status, entitlement data)
What this means for your organization
- Richer, real-time context in every conversation
- Less manual copying of data between systems
- The ability to keep knowledge sources in system of record, instead of duplicating them into static knowledge bases
Impactory Insight: Design MCP as a Layered Integration Fabric
We recommend thinking about MCP resources in three layers:
- Core Systems Layer
- ERP, CRM, ticketing, HRIS, line-of-business apps
- MCP resources surface relevant data (e.g., open tickets, current contract, order status)
- Knowledge & Policy Layer
- Policy documents, process guides, manuals
- Agents read these dynamically, so content stays in its authoritative source
- Contextual Interaction Layer
- The agent composes answers using a mix of conversational intelligence and real-time data from MCP resources.
Impactory typically helps clients:
- Identify priority systems to expose via MCP
- Define data contracts (what is exposed, under which conditions)
- Set up access controls so the right agents see the right resources
- Create evaluation sets to ensure the agent uses the data appropriately and remains grounded
This approach keeps governance under control while unlocking far more powerful use cases than a static FAQ-style copilot.
6. Analytics, Intelligence & ROI Measurement
Savings Analytics for Conversational Agents
Copilot Studio now extends its savings analytics beyond autonomous agents to include conversational agents. Organizations can:
- Configure time and cost savings settings per agent
- Time saved per interaction or completed workflow
- Cost equivalents (e.g., hourly rate of human agents)
- Let Copilot Studio automatically aggregate these metrics over time
- View how agents contribute to:
- Reduced manual work
- Faster resolution times
- Operational efficiencies
What this means for your organization
- A clearer business case for expanding or refining AI agents
- A consistent view of impact across different agent types
- Better prioritization of where to invest next
Thematic Analysis of User Questions
Another significant enhancement is automatic grouping of user questions into themes. For agents that use generative answers and have received at least 50 questions in the last seven days, Copilot Studio can:
- Cluster questions into themes (e.g., billing, password reset, product features)
- Show key metrics per theme:
- Question volume
- Answer rate
- User satisfaction (thumbs up/down)
- Allow drill-downs into:
- Specific user questions
- Corresponding agent responses
- The underlying metrics
What this means for your organization
- A manageable, “bird’s-eye view” of what users are really asking
- Rapid identification of content gaps or misunderstood topics
- A data-backed way to prioritize knowledge and process improvements
Impactory Insight: Turn Analytics into a Continuous Improvement Loop
We see the best results when organizations use these analytics not just as reports, but as inputs into a continuous improvement cycle:
- Measure
- Savings and usage
- Thematic breakdown of questions
- Satisfaction per theme
- Decide
- Which themes to improve (high volume, low satisfaction)
- Which flows or knowledge sources to refine
- Where to add new automations or escalations
- Act
- Update content, flows, or integrations
- Adjust evaluation sets to reflect new requirements
- Roll out improvements and re-measure
Impactory often helps clients create custom dashboards that integrate Copilot analytics with other business metrics (ticket volumes, handle times, NPS, etc.) to provide a unified view of impact.
7. Governance, Security & Enterprise Control
Controlling Org-Wide Sharing of Copilot Studio Lite Agents
As more employees start building their own agents using Copilot Studio lite (formerly the agent builder in Microsoft 365 Copilot), governance becomes crucial.
A new admin control in the Microsoft 365 Admin Center now allows organizations to:
- Restrict or disable organization-wide sharing of agents
- Choose who can share agents with “Anyone in your organization”:
- All users (default)
- No users
- Specific users or groups
- Ensure that existing access remains unchanged but future sharing respects the new policies
What this means for your organization
- Better control over which agents can be broadly accessed
- Reduced risk of unvetted agents spreading across the tenant
- A more structured path from “personal productivity agents” to approved enterprise agents
Impactory Insight: Balance Empowerment and Control
From a change-management perspective, Copilot Studio lite is a powerful citizen-development tool. At the same time, unmanaged proliferation of agents can create:
- Duplicated experiences
- Conflicting answers
- Compliance and data-access risks
We typically help clients design a governance model with:
- Clear tiers of agents
- Personal agents
- Team-level agents
- Organization-wide agents
- Approval workflows for moving an agent from one tier to the next
- Policies and templates for:
- Naming conventions
- Data access
- Evaluation and testing requirements
- Ownership and lifecycle management
This way, organizations support innovation at the edge while keeping risk and fragmentation under control.
Conclusion: Turning Copilot Studio Updates into Enterprise Advantage
The October 2025 Copilot Studio updates are not just incremental feature tweaks. Together, they form a stronger platform for:
- Building higher-quality, more reliable agents
- Choosing and testing the right models for each use case
- Running flows faster and more efficiently
- Creating richer, file-aware omnichannel experiences
- Connecting agents to real-time systems via MCP
- Measuring ROI and understanding user needs at scale
- Governing AI adoption responsibly across the organization
The challenge—and opportunity—for enterprises is to connect these capabilities into a coherent AI strategy across IT, operations, and business teams.
Next Step: Schedule a Copilot Studio Consultation with Impactory
If you want to go beyond reading update notes and actually embed these capabilities in your organization, we can help.
In a Copilot Studio Consultation, Impactory will:
- Review your current Copilot and Copilot Studio usage
- Identify priority use cases across departments (IT, HR, Finance, Customer Service, Operations)
- Assess your readiness for:
- Automated evaluations and quality frameworks
- Model strategy (GPT-4.1, GPT-5 previews)
- MCP-based integrations and data access
- Governance and analytics
- Outline a concrete 90-day action plan to:
- Improve existing agents
- Launch new, high-impact scenarios
- Put the right governance and analytics in place
blog
News from Impactory
Find out the latest from our company and stay up to date with everything worth knowing about our intelligent solutions and services from the multifaceted Microsoft Office world.
What is Microsoft Whiteboard?
Microsoft Whiteboard in Microsoft 365 is a free-form, digital new canvas where people, content, and ideas come together.
Microsoft Power Apps
Power Apps is a suite of apps, services, and connectors, as well as a data platform, that provides a rapid development environment to build custom apps for your business needs.
Microsoft 365 Learning Pathways – What’s In It For You
The Microsoft 365 Learning Pathways solution is a quick and easy entry point to Microsoft 365 knowledge & organizations’ training strategies.
What is Microsoft To Do?
Microsoft To Do is a valuable companion that simplifies your daily routines and helps you stay on top of your tasks.
SharePoint Viva Connections (formerly known as the Home site app)
With the new Homesite app, you will be able to integrate your intranet based on SharePoint Online right into Teams with a few clicks.
IMPACTORY
Your reliable, high-performance partner
We offer a wide range of consultancy services for the planning, introduction, and implementation of SharePoint, Microsoft 365, and hybrid applications. Benefit from our many years of experience in the industry.