Navigation
CyberAdvisors Hub
Structure Copilot Guide Dev Planner Model Guide Solution Stack Agent Playbook $Agent Cost Calculator PII Sanitizer Agent Tracker Meeting Agenda Agent Intake
CyberAdvisors AI Toolkit · Internal Use
CyberAdvisors · Reference

Microsoft Copilot Guide

Deployment, licensing, PRU estimation and usage guidance for the Microsoft Copilot ecosystem.

// Select Copilot Variant
Microsoft · Productivity Suite
Microsoft 365 Copilot
Word · Excel · PowerPoint · Outlook · Teams · SharePoint · Loop
Add-On License
Licensing
M365 Business$30/user/mo
E3/E5 Add-On$30/user/mo
Model TokensIncluded
Min Users1
// Token costs absorbed by Microsoft · requires M365 Business Standard/Premium or E3/E5 base · no per-call billing
Word / Docs Excel / Data Outlook Teams ⊞ Long Context ◎ Graph-Grounded
In-app productivity tasks where the data already lives in Microsoft 365 — summarising long email threads, drafting responses in Outlook, turning meeting transcripts into action items in Teams, building first-draft decks in PowerPoint, and running natural-language queries against Excel data. Copilot reads your tenant's Microsoft Graph data (emails, files, meetings) so it has full context without you pasting anything.
Ideal Query Types
  • "Summarise this Teams meeting and list action items"
  • "Draft a reply to this email thread maintaining a professional tone"
  • "Create a slide deck from this Word document"
  • "Analyse this Excel table and identify the top trends"
  • "Find all emails from [client] about [project] this month"
  • "Rewrite this section to be more concise"
Strengths
  • Native integration — no copy/paste
  • Reads your real tenant data via Graph
  • No per-token billing surprises
  • Respects existing M365 permissions
  • Teams transcription + summarisation
  • Business Chat spans all M365 apps
Limits
  • $30/user/mo is expensive at scale
  • Quality tied to GPT-4o — not best-in-class reasoning
  • No control over underlying model
  • Poor at complex multi-step logic
  • Hallucinations do occur — verify outputs
  • Requires clean, well-structured data
App-by-App Capability
AppWhat Copilot Can DoWhat It Cannot DoUseful?
WordDraft, rewrite, summarise, extract, change tonePull live external data; complex calculations✓✓
ExcelNL queries, formula suggestions, trend analysis, chart creationMacros/VBA generation; real-time data feeds✓✓
PowerPointGenerate deck from Word doc/prompt, add design ideasCustom brand templates (hit or miss); complex layouts
OutlookSummarise threads, draft replies, meeting prep briefsSend emails autonomously; manage calendar rules✓✓
TeamsMeeting recap, action items, Q&A on transcriptReal-time captioning for all languages; join as bot✓✓
LoopGenerate content blocks, brainstorm, summariseDeep integration with external tools
SharePointAsk questions about site content, summarise pagesComplex site admin actions; bulk operations~
Business ChatCross-app: "What did I miss on [project] last week?"Actions across apps; external data without plugins✓✓
MSP-Specific Prompt Examples
Outlook
"Summarise the last 2 weeks of emails from Acme Corp and list any open issues they've raised."
Pulls from your mailbox via Graph — no pasting required. Great for client catch-ups.
Teams
"Recap today's meeting, extract all action items with owners and deadlines, and format as a table."
Works on Teams transcripts. Saves 10–15 min of note-taking per meeting.
Excel
"Identify which clients have the most open tickets this month and show it as a bar chart."
Best when your sheet is clean and well-labelled. Struggles with merged cells or messy data.
Word
"Rewrite the executive summary to be more concise and shift the tone to match a technical audience."
One of Copilot's strongest use cases — tone-aware rewriting in context.
Licensing Breakdown
M365 Business Standard
$12.50/user/mo + $30 Copilot add-on
Word, Excel, Teams, Outlook Copilot. Minimum viable for most SMB clients.
M365 Business Premium
$22/user/mo + $30 Copilot add-on
Adds Intune, Defender, Azure AD P1. Recommended for MSP-managed clients.
M365 E3 / E5
$36–$57/user/mo + $30 Copilot add-on
Enterprise compliance, advanced analytics. E5 has Defender P2 and Purview.
Copilot Free (Web)
Free — bing.com/chat
No M365 Graph data access. Just GPT-4o with web search. No enterprise context.
Microsoft · Low-Code AI Platform
Copilot Studio
Custom AI agents · Multi-channel deployment · Power Platform integration
Consumption + Seat
Pricing
Seat License$200/mo
Per Message$0.01
Generative AI msg$0.03
Incl. Messages25K/mo
// $200/mo tenant seat = 25,000 messages included · overage billed per message · generative AI responses (GPT-4) cost 3× a standard message
Custom Agents ⊕ Teams / Web / SharePoint No-Code Builder ◉ Power Platform ⊞ Knowledge Sources
Building custom AI agents without writing backend code. Point it at SharePoint sites, uploaded documents, or Dataverse tables as a knowledge base, then deploy the bot to Teams, a website, or a SharePoint page. Ideal for MSPs building client-facing helpdesk bots, internal IT FAQ agents, or onboarding assistants. Connects to 1,000+ connectors via Power Platform and supports multi-agent orchestration (Copilot Studio agents calling other agents).
Ideal Use Cases
  • IT helpdesk bot for client Teams channel
  • Onboarding FAQ agent (reads SharePoint docs)
  • Ticket-routing agent via ServiceNow/ConnectWise connector
  • HR policy Q&A bot for internal staff
  • Vendor price/availability lookup automation
  • Broadcast storm triage agent (multi-step workflow)
Strengths
  • Low-code — no LLM expertise needed
  • Native Teams & SharePoint deployment
  • Connects to Power Automate flows
  • Generative answers from your docs
  • Multi-agent orchestration supported
  • Azure AD auth baked in
Limits
  • $200/mo seat is steep for small bots
  • Generative messages burn budget fast
  • Complex logic needs Power Automate
  • Custom model selection limited
  • Debugging and testing UX is clunky
  • Latency on generative responses
Architecture Concepts
ComponentWhat It IsCost ImpactRequired?
TopicsConversation flows — trigger phrases → response logic1 message each✓ Core
Generative AnswersGPT-4 reads knowledge source and generates a reply3 messages eachOptional
Knowledge SourcesSharePoint sites, uploaded files, public URLsStorage onlyFor Gen AI
ActionsCall Power Automate, HTTP endpoint, or connector1 msg + flow costOptional
AuthenticationAzure AD (SSO) or manual auth per userNo extra costRecommended
ChannelsTeams, Web, SharePoint, Slack, custom endpointNo extra cost✓ Choose one+
Multi-AgentOne Studio agent orchestrates other Studio agentsMessages eachAdvanced only
Prompt / Trigger Examples
Trigger Phrase
"I can't connect to the VPN"
Route to a Topic that walks through VPN troubleshooting steps, then escalate to ticket if unresolved.
Generative Answer
"What's our password reset policy?"
Reads the IT policy SharePoint page and generates a contextual answer. Uses 3 messages — watch volume.
Action Call
"Log a ticket for my broken monitor"
Triggers a Power Automate flow that creates a ConnectWise or ServiceNow ticket with captured entity values.
Multi-Agent
"Investigate why this switch port is flapping"
Master agent delegates to a network-analysis sub-agent and a change-log-search sub-agent, synthesises results.
Microsoft / GitHub · Developer Tools
GitHub Copilot
VS Code · Visual Studio · JetBrains · Vim · Neovim · GitHub.com · CLI
Free Tier + Paid
Pricing
Free$0/mo
Individual$10/mo
Business$19/user/mo
Enterprise$39/user/mo
// Free: 2,000 completions/mo + 50 chat messages · Business adds policy controls + audit logs · Enterprise adds fine-tuning + Copilot Extensions
Code Completion Inline Chat ◈ Multi-file Edits CLI Assistant ⊕ Copilot Extensions
Accelerating code writing directly inside your IDE. Provides line and block autocomplete, inline chat to explain/refactor/test code in context, and a sidebar chat for broader questions. The Business and Enterprise tiers add multi-file edits ("Copilot Edits"), workspace-aware context, custom instructions, and Copilot Extensions (connects to external tools). For MSPs writing PowerShell, Python automation, or dashboards — the free tier alone is useful.
Ideal Query Types
  • "Write a PowerShell function to query AD for stale accounts"
  • "Explain what this regex does" (inline /explain)
  • "Refactor this function to be more readable"
  • "Generate unit tests for this Python class"
  • "Fix the bug on line 42" (inline /fix)
  • "What does this entire repo do?" (workspace context)
Strengths
  • Free tier is genuinely useful
  • Deeply integrated in VS Code
  • Context-aware of open files
  • Supports 30+ languages
  • CLI assistant for shell commands
  • Multi-file Copilot Edits (Business+)
Limits
  • Not a replacement for a senior dev
  • Can generate plausible but wrong code
  • Free: limited to 2K completions/mo
  • Struggles with complex architecture
  • No local model option (cloud only)
  • Org policy can restrict usage
Feature Comparison by Tier
FeatureFreeIndividual $10Business $19Enterprise $39
Code Completions2,000/moUnlimitedUnlimitedUnlimited
Copilot Chat (IDE)50 msgs/mo✓ Unlimited✓ Unlimited✓ Unlimited
Copilot Chat (GitHub.com)
Copilot Edits (multi-file)
CLI Assistant
Policy / Org Controls
Audit Logs
Copilot ExtensionsLimited
Fine-Tuning (custom model)
IP Indemnity
MSP-Specific Prompt Examples
PowerShell
"Write a script to export all AD users inactive for 90+ days to CSV with their last logon date."
Copilot will use Get-ADUser with a LastLogonDate filter. Always test in a dev OU first.
Python
"Add error handling and a retry loop to this API call function with exponential backoff."
Excellent at adding boilerplate patterns. Paste your existing function and ask inline with /fix or /edit.
Explain
Select a complex regex → right-click → "Copilot: Explain"
One of the most practical features. Breaks down patterns like IP validation or log parsing regex line by line.
CLI
gh copilot suggest "find all files modified in the last 24 hours and zip them"
CLI assistant suggests and explains shell commands. Type gh copilot explain for any command you don't recognise.
Copilot Studio · GitHub Copilot
PRU Overage Estimator
Premium Request Unit calculator — model your agent workload and forecast monthly cost
Interactive Tool
// Configure your plan, agent stages, and usage — results update in real time
Plan Configuration
Copilot Plan
Total licensed users
Auto-set by plan above
Agent Workload
Agent run profile
PRU-consuming steps per run
Avg daily invocations
Other Monthly PRU Usage
Non-agent usage per user / month
Model Multiplier
Primary model used by agent stages
GPT-4.1 / GPT-4o · Included
GPT-4o mini0.5×
Claude Sonnet 4 / GPT-4.1 Premium
o3-mini / Gemini 2.0 Flash1.5×
Claude Opus 4 / GPT-5
o3 / o1
Monthly Estimate
PRU per Agent Run
9
9 stages × 1× multiplier
Agent PRU / Month
990
9 PRU × 5 runs × 22 days × 1 user
Other Usage PRU / Month
616
Chat + code review + CLI
Total PRU Consumed
1606
vs. 1,500 included
0107%1,500 incl.
Overage PRU
106
× $0.04 = $4.24
Cost Breakdown
Subscription$39.00
Overage ($0.04/PRU)$4.24
Est. Monthly Total$43.24
 Approaching limit — over 80% of included PRUs consumed. Consider upgrading or throttling agent frequency.
OVERAGE ACTIVE — charges accrue at $0.04/PRU. Set a budget cap in GitHub billing settings.
Break-Even Runs/Day
Max agent runs/day before overage
given current other usage
Microsoft Copilot — Reality Check
What Copilot Can't Do
Gaps vs. ChatGPT · Claude · Grok — and what to use instead
Know the Gaps
Microsoft Copilot is an integration layer, not a general-purpose AI assistant. It excels when data lives in your M365 tenant — but standalone AI tools are still superior for everything outside that boundary.
// Head-to-Head Gaps — Copilot vs. Standalone AI
Deep / Open-Ended Reasoning
Complex multi-step logic, root cause chains, hypotheticals
Copilot Weak
✗ Copilot Cannot

M365 Copilot is optimised for retrieval and summarisation of existing tenant data. It does not chain multi-step logical deductions, explore edge cases unprompted, or work through novel problems without being anchored to a document or thread.

→ Use Instead
Claude OpusBest nuanced reasoning, handles ambiguous complex prompts with minimal hallucination
GPT-5 ReasoningBuilt-in thinking tokens for deep logic chains, math, and STEM problems
Grok 3Strong STEM reasoning with real-time context included
Real-Time Web Search & Current Events
Breaking news, live pricing, CVE advisories, stock data
Copilot Limited
✗ Copilot Cannot

M365 Copilot reads your tenant's internal data, not the live web. The Bing-connected Copilot (copilot.microsoft.com) does browse the web, but it has no access to your Microsoft Graph data — they're two separate products. CVE lookups, vendor pricing, and breaking news are always stale inside M365 Copilot.

→ Use Instead
Grok 3Live X/Twitter index + real-time web — best for breaking news and social signal monitoring
PerplexityAlways-on web search with inline citations — ideal for CVE lookups and vendor pricing
ChatGPT (Browse)GPT-4o with Bing grounding — good general web Q&A with source links
Long-Form Creative & Persuasive Writing
Marketing copy, narratives, proposals, thought leadership
Copilot Weak
✗ Copilot Cannot

Copilot rewrites and improves existing documents well, but it cannot produce genuinely novel long-form creative writing, compelling narratives, or persuasive copy from a blank slate. Its Word integration is reactive — it needs source material. Ask it to write a 1,500-word thought leadership article and you get bland, generic output.

→ Use Instead
Claude Sonnet / OpusIndustry-leading tone, voice consistency, and long-form structure — best for client proposals and reports
ChatGPT GPT-5Strong creative writing, good at matching brand voice from examples
Arbitrary API & External Data Integration
ConnectWise, Axcient, Datto, custom databases — without building a Studio agent
Copilot Very Limited
✗ Copilot Cannot

Without a custom Copilot Studio agent and connector, Copilot cannot reach outside the M365 ecosystem. Querying ConnectWise tickets, checking Axcient backup status, pulling Datto RMM alerts, or reading from a custom SQL database requires building a full Studio agent — which costs PRUs and dev time.

→ Use Instead
Claude (API + tools)Tool/function calling lets it query any API — ideal for custom MSP automation pipelines
ChatGPT (API + tools)GPT-4o function calling with broad plugin/connector ecosystem
Copilot StudioBuild a custom agent with Power Platform connectors — budget PRUs and dev time accordingly
General Code Generation & Debugging
PowerShell, Python, JavaScript — writing scripts from scratch
M365 Cannot
✗ Copilot Cannot

M365 Copilot is not a coding assistant. It can write simple Excel formulas and basic Office automation, but has no code editor integration, no file context, no terminal, and cannot write or debug PowerShell or Python at any meaningful scale. That requires GitHub Copilot — a completely separate license.

→ Use Instead
GitHub CopilotPurpose-built — IDE autocomplete, inline chat, multi-file edits, CLI assistant. Free tier available.
Claude (chat)Excellent for writing full scripts, explaining code, PowerShell/Python without needing an IDE
ChatGPTStrong coder with built-in code interpreter for running and testing Python inline
Model Selection & Customisation
Choosing which AI model runs, system prompts, temperature, fine-tuning
Copilot Cannot
✗ Copilot Cannot

M365 Copilot runs on whatever model Microsoft deploys. You cannot switch to Claude, choose a reasoning model, set a system prompt, adjust temperature, or fine-tune on your data. The model is a black box managed entirely by Microsoft with no configuration surface.

→ Use Instead
Claude (API)Full model selection (Haiku → Opus), system prompts, temperature control, tool use
OpenAI (API)Choose GPT-4.1, o3, or GPT-5 per task — full parameter control
Grok (API)Grok-3 / grok-3-mini selection with system prompts and real-time grounding toggle
Image Generation
Creating images, diagrams, visual assets from prompts
Copilot Cannot
✗ Copilot Cannot

M365 Copilot has no image generation inside the productivity suite. PowerPoint Designer is layout-based, not generative. Consumer Copilot.microsoft.com can generate images via DALL·E, but it's completely separate from your M365 tenant Copilot and not enterprise-governed.

→ Use Instead
ChatGPT (DALL·E / GPT-4o)Native image generation and vision in the same conversation — strong for diagrams and mockups
Grok (Aurora)High-quality photorealistic image generation via Aurora model
Claude (vision input)Cannot generate images but excels at reading and analysing screenshots and diagrams you paste in
Air-Gapped / On-Premises Deployment
Regulated data that cannot leave your network (HIPAA, CJIS)
Copilot Cannot
✗ Copilot Cannot

All three Copilots (M365, Studio, GitHub) are cloud-only. Data is processed on Microsoft's infrastructure. For clients with HIPAA, CJIS, or air-gapped requirements where data cannot touch a commercial cloud endpoint, Copilot is not a compliant option.

→ Use Instead
Llama 3.x (self-hosted)Open weights — run completely on-prem on your own GPU hardware. No data leaves the network.
DeepSeek R1 (self-hosted)Frontier-level reasoning, open weights, self-hostable — strong for regulated environments
Claude (Azure deployment)Claude on Azure has data residency options and BAA support for HIPAA workloads
At-a-Glance Comparison
Capability M365 Copilot ChatGPT Claude Grok
Reads your M365 tenant data
Real-time web search+$+$
Deep open-ended reasoning
Long-form creative writing
General code generation
Image generation
Custom API / tool calling
Model selection / fine-tuning
Self-host / air-gapped deployAzure
Included in M365 subscription+$30/user
The Right Answer
01
Copilot + Claude = Full Coverage

Use M365 Copilot for anything that lives in your tenant — emails, meetings, SharePoint, Teams. Use Claude for everything else — reasoning, writing, code, and analysis that doesn't touch M365.

02
Copilot Is Not a Standalone AI

Clients who buy M365 Copilot expecting a ChatGPT replacement will be disappointed. It's a productivity layer on top of M365 data — not a general-purpose assistant. Set expectations before rollout.

03
Grok for News, Claude for Docs

If a client needs real-time intel (vendor outages, CVEs, market news), Grok beats everything in currency. If they need polished deliverables (reports, proposals), Claude Sonnet is the better pick.

04
ChatGPT for Code Interpreter

ChatGPT's built-in Python code interpreter runs and tests code inline — no IDE needed. For quick data analysis, CSV processing, or one-off scripts, it's faster than any IDE-based tool.

05
Three Copilots ≠ One Product

Remind clients: M365 Copilot, Copilot Studio, and GitHub Copilot are three separate products with separate licenses, billing, and capabilities. Having one does not mean you have the others.

06
Regulated Clients: Use Open Models

For HIPAA, CJIS, or air-gapped environments — none of the Copilots qualify. Recommend self-hosted Llama 3.x or DeepSeek R1 on client-owned hardware, or Claude via Azure with a signed BAA.

Cross-Copilot Tips
01
Know Which Copilot You're In

They share a name but are completely different engines. M365 Copilot reads your tenant data. GitHub Copilot reads your code. Studio builds agents. Confusing them leads to wrong tool, wrong budget.

02
Studio Generative Msgs = Budget Risk

Each GPT-4 generative answer costs 3 messages. A bot handling 500 users asking open-ended questions will burn through the 25K/mo allowance fast. Set fallback topics for common queries to avoid generative calls.

03
M365 Copilot Needs Clean Data

Copilot is only as good as what's in your tenant. Messy SharePoint, untagged files, and sprawling mailboxes reduce accuracy. Run a data hygiene pass before rolling out to clients.

04
GitHub Free Tier is Worth It

Even for MSPs doing occasional scripting, the free tier's 2,000 completions and 50 chats/mo is enough for most PowerShell and Python work. Upgrade to Individual ($10) for heavy use.

05
Verify All Outputs

All three Copilots can hallucinate. M365 Copilot can misread meeting context. GitHub Copilot can suggest deprecated functions. Studio can return out-of-date info from stale knowledge sources. Always review.

06
Licensing Stacks — Plan Ahead

M365 Copilot + Copilot Studio + GitHub Copilot Business for one developer could cost $30 + $19 = $49/mo minimum, plus the base M365 seat. Model the full stack before committing clients.