In This Article
- Wave 3 and the Agentic Shift: Why This Matters Now
- From Teams Bots to Copilot Studio Agents
- Copilot Cowork: When AI Builds Your Spreadsheets and Presentations
- Four Agent Types Every Financial Institution Should Build
- What Teams Agents Actually Do All Day
- Agent 365: Governing AI Agents Like Employees
- Building Your First Teams Agent in 4 Weeks
- Frequently Asked Questions
Microsoft launched Copilot Wave 3 on March 9, 2026, and it changes the conversation about what Teams can do inside a financial institution. Wave 3 shifts Copilot from a conversational assistant into an agentic platform—meaning it no longer waits for someone to ask a question. It executes multi-step workflows across Microsoft 365 applications autonomously, with transparent checkpoints that keep humans in the loop where it matters.
For banks, credit unions, and mortgage companies that already live in Teams, this is not a theoretical upgrade. Your relationship managers are already chatting with operations staff there. Your compliance team already posts alerts there. Your loan officers already run pipeline meetings there. Wave 3 turns that communication hub into an automation platform where AI agents handle the coordination work that currently burns hours every day.
IDC projects 1.3 billion enterprise agents by 2028. Microsoft 365 Copilot surpassed 15 million paid seats as of the January 28, 2026 earnings call. The technology is production-ready. The question for financial institutions is no longer whether to deploy AI agents, but how fast you can get the first one running.
HumAI research shows agentic workflows deliver 30% to 50% faster business processes and a 25% to 40% reduction in low-value work. One global financial services firm cut audit reporting time by 92%. First West Credit Union in Canada deployed Copilot to eliminate hours of daily manual work for advisors. AIB in Ireland rolled out Copilot to over 10,000 employees, reporting measurable productivity gains in document processing and internal communications.
Wave 3 and the Agentic Shift: Why This Matters Now
Before Wave 3, Copilot in Teams was reactive. You asked it to summarize a meeting, draft a message, or find a document. It did those things well. But it waited for you to ask.
Wave 3 introduces Agent Mode across Word, Excel, PowerPoint, Outlook, and Teams. Copilot now takes autonomous actions within boundaries you define—scheduling follow-ups, routing approvals, updating records, and triggering workflows based on events happening across your Microsoft 365 environment.
Three capabilities landed this month that make agentic AI practical for regulated financial institutions:
Multi-Agent Orchestration
One agent can call other agents. A customer onboarding agent delegates to a document agent, a compliance check agent, and a notification agent. Each handles its specialty. The orchestration layer coordinates the full sequence without a human routing files between people.
Cross-App Workflow Execution
Agents now work across Word, Excel, PowerPoint, Outlook, and Teams without switching between applications. Ask an agent to prepare a quarterly review and it pulls performance data from Excel, assembles a summary in Word, builds the presentation in PowerPoint, and schedules the meeting in Outlook. The work happens in the background. You review the output when it is ready.
Human-in-the-Loop Controls
Copilot Studio's "request for information" action pauses an agent flow to collect a decision from a designated reviewer through Outlook, then resumes execution using their response. For financial institutions where compliance requires human sign-off on certain actions, this is not optional. It is the feature that makes autonomous agents acceptable to regulators and auditors.
From Teams Bots to Copilot Studio Agents
If your institution deployed Teams bots in the past few years, you already understand the value of automation inside Teams channels. What changed is the ceiling. Traditional bots followed scripts. Copilot Studio agents reason about context, access multiple data sources, and take actions across systems.
Traditional Teams Bot
- Script-based: if X, respond Y
- Single data source per interaction
- Cannot take actions across systems
- Requires a developer for every change
- No context awareness between sessions
Copilot Studio Agent (Wave 3)
- Reasons about context and intent
- Queries multiple systems simultaneously
- Takes actions: creates records, routes files, sends alerts
- No-code visual builder for operations managers
- Multi-agent orchestration with governance controls
Copilot Studio provides the no-code visual builder where you define triggers, actions, and decision logic. Your operations manager or IT administrator can build and modify agents without writing code. For integrations with specialized platforms, an implementation partner handles the initial connection, but ongoing agent management requires no coding skills.
The pricing structure supports gradual adoption. Copilot for Microsoft 365 Business dropped to $18 per user per month through June 2026. Copilot Studio starts at $200 per month for 25,000 agent credits, or you can use pay-as-you-go pricing. Agent 365, the governance layer, adds $15 per user per month when it goes generally available on May 1, 2026.
Copilot Cowork: When AI Builds Your Spreadsheets and Presentations
The most tangible Wave 3 feature is Copilot Cowork, and it is the easiest to understand because it does the work you already do—just faster.
Cowork is the execution layer for Microsoft 365. You describe what you need in plain language, and it plans the steps, works across applications, and delivers finished output for your review. It runs in the background while you do other things. Microsoft calls this "delegation over prompting" because you are not asking Copilot a question. You are handing it a task.
Excel: Build the Tracker
"Create a spreadsheet tracking our open account applications by branch, status, and days since submission." Cowork builds the Excel file, formats it, adds formulas for aging calculations, and populates it from data it finds in your SharePoint files and recent emails.
PowerPoint: Build the Presentation
"Prepare the quarterly review deck for the board." Cowork pulls performance data from Excel, assembles slides matching your brand kit, adds charts, and writes speaker notes. A presentation that takes someone two hours to build arrives in your Teams chat ready for review.
Outlook: Clean Up the Calendar
"Review my schedule this week and find time for the Henderson prep." Cowork scans your calendar, identifies low-priority meetings, proposes rescheduling, adds focus blocks, and sends decline notices—all after you approve the plan.
Word: Assemble the Report
"Compile this month's branch operations summary." Cowork gathers data from Excel spreadsheets, pulls context from recent meeting notes and emails, and assembles a formatted Word document with the narrative, tables, and key metrics your leadership team expects.
Here is what makes Cowork different from the Copilot you may have tried last year: it uses multiple AI models working together. Microsoft integrated Anthropic Claude alongside OpenAI models so Cowork can assign the right model to the right task automatically. Claude handles long-form analytical work. OpenAI models handle structured data processing. You never see this happening. You just get better output, faster.
Cowork Runs Inside Your Security Perimeter
Every action Cowork takes is governed by your existing Microsoft 365 security policies. Sensitivity labels follow documents. DLP policies cover Copilot prompts and responses. Conditional Access controls who can delegate what. For financial institutions, this means Cowork operates under the same compliance controls as your employees—because it is using the same identity and permission infrastructure.
Charles Lamanna, Microsoft's President of Business Applications, demonstrated Cowork analyzing a month of meetings with direct reports, compiling customer notes from a business trip, and generating a competitive analysis—delivering both a Word document and Excel spreadsheet—all running autonomously in the background. That is the kind of work that takes a branch operations manager an entire afternoon. Cowork finishes it while they are in their next meeting.
Four Agent Types Every Financial Institution Should Build
Financial institutions share four operational patterns that map directly to Teams agent capabilities. Whether you are a community bank processing commercial loans, a credit union managing member services, or a mortgage company running a pipeline, these agent types address the same coordination bottlenecks.
| Agent Type | What It Does | Where It Works | ROI Driver |
|---|---|---|---|
| Workflow | Routes work items to the next responsible party based on capacity, specialty, and product type | Teams, Outlook, SharePoint | Eliminates 4–8 hr handoff delays |
| Notification | Sends targeted alerts with deep links: document received, approval needed, deadline approaching | Teams, Outlook, SharePoint | Saves 15–20 min per transaction |
| Conversational | Answers status queries with live data from connected systems | Teams, SharePoint, Excel | Eliminates status-check meetings |
| Compliance | Monitors deadlines, tracks policy acknowledgments, and flags documentation gaps | Teams, SharePoint, Purview | Prevents audit findings |
Workflow agents watch for status changes and route work to the next responsible party. When a document package is complete, the agent assigns it to the right reviewer based on workload and specialty. When an approval comes back, the agent moves the file to the next step and notifies everyone involved. Every routing decision is timestamped and searchable—an audit trail that manual assignment never creates.
Notification agents send targeted alerts to the right people at the right time. Document received, approval needed, deadline in 48 hours, policy acknowledgment overdue. Each notification includes a deep link so the recipient goes straight to the relevant item without searching through systems.
Conversational agents answer questions from your team in natural language. "What is the status of the Johnson file?" The agent checks SharePoint for recent document activity, reviews the Teams channel history, and responds with a summary: current stage, next required action, who is responsible, and what the deadline is.
Compliance agents track deadlines, documentation requirements, and policy acknowledgments across your organization. Annual training certifications, security awareness completions, policy review sign-offs, document retention schedules. The agent flags what is overdue, who is responsible, and what the deadline is—before it becomes an audit finding.
Is Your M365 Environment Ready for AI Agents?
Teams agents depend on the data governance, permissions, and security architecture underneath them. ABT's AI Readiness Scan maps your tenant against the prerequisites for Copilot Studio deployment.
Take the AI Readiness Scan Talk to an ABT AI SpecialistWhat Teams Agents Actually Do All Day
New Employee Onboarding
Every financial institution hires people, and every financial institution knows the onboarding chaos that follows. A new employee starts Monday. IT needs to provision accounts. HR needs signed paperwork. Compliance needs training scheduled. The manager needs orientation meetings on the calendar. An onboarding agent coordinates all of it:
- Creates a SharePoint folder with the standard subfolder structure: offer letter, tax forms, direct deposit, emergency contacts, equipment checklist
- Posts a notification in the hiring manager's Teams channel with the new hire's name, start date, role, and a link to the onboarding folder
- Sends tasks to IT (provision M365 account, set up workstation), HR (benefits enrollment packet), and Compliance (schedule mandatory training modules)
- Emails the new employee a welcome message with first-day instructions, parking information, and a link to upload their completed forms
- Creates a 30-day onboarding checklist in Teams with milestones: day-one orientation, week-one check-in, 30-day review
The hiring manager gets a Teams notification saying "Onboarding started for Sarah Chen" with the folder already built, IT already notified, and the first two weeks of check-ins already on the calendar. They focus on welcoming the new team member instead of chasing five departments through email.
Annual Compliance Training Tracking
Every employee at a financial institution has mandatory training: BSA/AML awareness, cybersecurity fundamentals, fair lending, harassment prevention, information security. Someone in compliance is always chasing people to finish theirs. A training tracking agent eliminates that entire job.
Annual BSA/AML training is due by March 31. The agent maintains a completion checklist by employee, pulled from the training platform and updated in a shared Excel tracker. When an employee finishes a module, the agent updates their status and posts a confirmation in the compliance Teams channel. Two weeks before the deadline, anyone who has not started gets a friendly reminder. One week out, their manager gets notified. Three days out, the compliance officer gets an escalation with names.
The compliance officer's dashboard in Teams shows training completion across the entire organization. Green means done. Yellow means in progress. Red means not started. No one spends hours pulling completion reports, cross-referencing spreadsheets, or sending individual reminder emails. When examiners ask for training records, the data is already organized.
Meeting Prep and Report Assembly
This is the use case that sells itself because everyone recognizes the pain. A branch manager has a quarterly review with regional leadership tomorrow morning. Today, that means two hours of pulling numbers from Excel, copying them into PowerPoint, writing a narrative summary, and drafting the agenda email.
With a meeting prep agent, the manager says: "Prepare me for tomorrow's quarterly review with Sandra." The agent gathers recent performance data from shared Excel files, pulls context from the last quarterly review meeting notes, checks the Teams channel for relevant discussions, and delivers three things: a briefing document in Word, a presentation deck in PowerPoint, and a draft agenda email in Outlook. The manager reviews, adjusts, and sends—in 15 minutes instead of two hours.
The same pattern works for client relationship reviews, board meeting prep, audit committee presentations, and any recurring report that someone manually assembles from multiple sources today.
Weekly Operations Coordination
Every financial institution has a version of the Monday morning operations report. Someone spends an hour pulling data from spreadsheets, formatting the summary, and posting it to the management Teams channel. A coordination agent handles this automatically.
- Pulls the latest data from shared Excel workbooks in SharePoint: application volume, processing times, pending items by team member
- Assembles the weekly summary in Word with the formatting and structure leadership expects
- Posts the report to the management Teams channel at 8:00 AM Monday with a brief narrative highlighting anything unusual
- Tracks action items from the previous week's report and flags what is still open
The agent does not make operational decisions. It assembles the information, tracks the follow-ups, and surfaces what needs attention. For institutions managing Conditional Access and security policies, coordination agents add another layer of operational visibility without adding headcount.
Agent 365: Governing AI Agents Like Employees
Financial institutions cannot deploy autonomous agents without governance. Agent 365, generally available May 1, 2026, provides the enterprise control plane for managing AI agents the way you manage employees—with defined permissions, activity monitoring, and lifecycle management.
For financial institutions subject to examination, Agent 365 provides what auditors will ask for: a documented record of what each agent can access, what actions it can take, and a complete audit log of every action it did take. Microsoft Purview integration means sensitivity labels are enforced on data that agents touch. DLP policies now cover Copilot prompts and responses, preventing agents from surfacing or transmitting restricted data outside authorized boundaries.
Agent 365 costs $15 per user per month. It is also bundled in the new Microsoft 365 E7 Frontier Suite at $99 per user per month, which combines E5, Copilot, and Agent 365 into a single license.
Key Takeaway
Deploy agents and their governance layer together. An AI agent without Agent 365 is an unmanaged employee with system access and no audit trail. For regulated financial institutions, that is an examination finding waiting to happen.
Building Your First Teams Agent in 4 Weeks
Start with one agent that solves one problem. The notification agent is the best starting point because it delivers immediate value with the lowest implementation risk.
Week 1: Define
Map the business events that trigger notifications: new application, document uploaded, condition set, approval needed, deadline approaching. Decide who receives each alert and through which Teams channel.
Week 2: Build
Use Copilot Studio's visual builder. Connect SharePoint, Outlook, and your Teams channels. Configure notification templates with the data fields your team actually needs—not everything available, just what drives action.
Week 3: Pilot
Run the agent for one branch or one team. Collect feedback on notification frequency, content, and timing. Adjust before broader rollout. Test the human-in-the-loop controls for any notification that triggers a compliance-sensitive action.
Week 4: Deploy + Govern
Enable for all users. Register the agent in Agent 365 with defined permissions and monitoring. Monitor adoption in the Teams admin center. Plan your second agent based on where the first one reveals the next bottleneck.
After the notification agent is stable, build the workflow agent for task routing. Then the conversational agent for status queries. Each new agent builds on the data connections you established in the first one. Most financial institutions have all four agent types running within 90 days.
Teams agents do not replace your relationship managers, processors, or compliance officers. They handle the repetitive coordination work so your team can focus on the decisions and relationships that grow your institution.
For financial institutions already using AI governance frameworks, Teams agents fit within the boundaries you have already established. For those just beginning their AI journey, the notification agent is a low-risk entry point that demonstrates value before you tackle more complex agent types.
"Most financial institutions we work with have 3 to 5 manual handoff points per transaction that add days to their processing time. A single Teams workflow agent typically eliminates 2 to 3 of those handoffs in the first month."
Serving 750+ financial institutions since 1999
Build Teams Agents That Eliminate Manual Handoffs
ABT configures Copilot Studio, connects your Microsoft 365 environment, implements governance through Agent 365, and manages the entire stack long-term.
Frequently Asked Questions
No. Copilot Studio provides a no-code visual builder where operations managers can create agents using drag-and-drop triggers, actions, and decision logic. For complex integrations with core banking platforms or custom loan origination systems, an implementation partner may be needed for the initial API connection, but ongoing agent management requires no coding skills.
Copilot Cowork is the execution layer introduced in Wave 3 that handles multi-step tasks across Microsoft 365 applications. While Teams agents automate specific workflows like notifications and document tracking, Cowork handles broader delegation tasks like building Excel spreadsheets, assembling PowerPoint presentations, and compiling Word reports from multiple data sources. Cowork uses multiple AI models including Anthropic Claude and OpenAI, automatically selecting the best model for each task. Both Cowork and Teams agents operate within the same Microsoft 365 security perimeter.
Most financial institutions see measurable productivity gains within 30 to 60 days of deploying their first Teams agent. Notification agents eliminate 15 to 20 minutes of manual communication per transaction. Workflow agents reduce handoff delays by 4 to 8 hours per file. Full ROI across all agent types typically materializes within 6 to 12 months depending on transaction volume and implementation scope.
Yes. Teams agents operate within the Microsoft 365 security perimeter, which includes Entra ID authentication, Conditional Access policies, and data loss prevention through Purview. As of Wave 3, DLP policies cover Copilot prompts and responses. Agent 365, generally available May 2026, provides governance controls for monitoring agent activity, managing permissions, and auditing every action an agent takes. These controls support GLBA, BSA/AML, and FTC Safeguards Rule requirements.
Copilot Studio is where you build agents. Agent 365 is where you govern them. Copilot Studio provides the no-code visual builder for defining triggers, actions, and decision logic. Agent 365 provides the enterprise control plane for managing agent permissions, monitoring activity, enforcing security policies, and maintaining the audit trail that regulators expect. Financial institutions should deploy both together. Agent 365 goes generally available May 1, 2026 at $15 per user per month.