Done 100's of Prototypes Taken 10 applications to Production ranging from simple RAG to more complex Agentic systems Specialize in AI in the SDLD or TDLC

This slide deck itself is living proof of Copilot's capabilities

Sources: Quote 1: Widely circulated in articles and LinkedIn posts (mid-to-late 2025) as businesses moved from piloting AI to large-scale deployment Quote 2: Sam Altman's forward-looking statements throughout 2024 and 2025, emphasizing the rapid, transformative role AI agents will play in daily business operations These quotes frame AI adoption as both inevitable and fundamentally about human change, not just technology

Typical ordering inside the context window System prompt – The hidden instructions from the platform or developer. Example: “You are ChatGPT, a large language model trained by OpenAI. Follow these guidelines…” Can also include special behavior rules, safety policies, or formatting requirements. Developer or application-specific instructions – Additional hidden setup from the app integrating the model. Example: “Always respond in JSON unless otherwise specified.” Conversation history – Your past messages + the model’s past responses. May be direct text or summaries if the history is long. Injected knowledge or retrieved content – Snippets from web search, databases, documents, or memory. Your latest prompt – The most recent user message.

System Prompts can vastly ajust the behavior, some issues in the past have been Google, Implicit prompt bias, with image generation Grok, Spreading extremist narratives OpenAI Sycophantic behavior

M365 Copilot has access to a vast array of enterprise data sources All data access respects existing permissions - you only see what you're authorized to see This comprehensive access is what makes Copilot so powerful for enterprise use Data stays within your Microsoft 365 tenant boundary Copilot uses Microsoft Graph API to access these sources securely

M365 Copilot works best when you give it context Start simple and build up complexity Works across the entire Microsoft 365 ecosystem

PACET helps structure effective prompts Not all elements needed every time, but more detail = better results Exemplars are especially powerful for consistent formatting

Look at my Teams chats, meetings and emails I've sent in the last two weeks. Based on the above analysis, what Star Wars character would represent my style of work best, and why? What would my abilities, strengths and personality flaws be? Can you format that as a Star Wars character dossier for me? Also give me a Star Wars character friend group based on my recent interactions with my current top [five] collaborators. Provide them with different types of Star Wars characters, include their real name, star wars type, home planet, strengths and weaknesses, and why they are my friends and how they help me in my work here at Accenture. Also identify my rival based on recent interactions with my colleagues and a short story of how we became nemeses but ended up being friends. List my real name, my most similar existing Star Wars Character, and include the same attributes listed above for friends and the rest of the information formatted and edited in a dossier style. ## **Breaking It Down** - **Persona**: Executive assistant - **Ask**: Create summary of last week's work - **Context**: Emails, Teams, OneNote, calendar - client deliverables & meetings - **Exemplars**: Bulleted status report format - **Tone**: Concise, achievement-focused This example showcases M365 Copilot's ability to pull from multiple data sources Demonstrates cross-application integration (Outlook, Teams, OneNote, Calendar) Shows how Copilot can save hours of manual status report compilation

## **Why It Works** - Produces more accurate, reliable results - Reveals the logic behind recommendations - Helps you verify the analysis process - Great for complex decisions Chain-of-thought prompting dramatically improves reasoning quality Forces the model to work through problems methodically Particularly effective for analysis, planning, and problem-solving tasks Makes AI reasoning transparent and verifiable Prompt: "Analyze the data to provide a recommendation. Ensure the recommendation accounts for outliers and trends."

## **Why It Works** - Ensures consistent formatting - Captures your specific style - Reduces need for detailed instructions Few-shot learning is incredibly powerful for consistent outputs 2-3 examples are usually enough to establish the pattern Great for formatting tasks, style matching, and template generation Works better than lengthy explanations of what you want

## **Why It Works** - Activates relevant knowledge domains - Adjusts tone and expertise level - Provides domain-specific insights - Contextualizes responses appropriately Role-based prompting leverages the model's training across different domains The persona helps frame the response with appropriate expertise Works well for technical advice, strategic planning, creative tasks Can combine roles: "You are both a CFO and a marketing strategist..."

## **Why It Works** - Evaluates multiple solutions - Compares trade-offs systematically - Leads to more balanced decisions - Reduces confirmation bias Tree-of-thought extends chain-of-thought to explore multiple reasoning paths Particularly powerful for strategic decisions and complex problems Forces consideration of alternatives before committing to a solution Great for scenarios where there's no obvious "right" answer Helps surface options you might not have considered

## **Why It Works** - Faster than crafting perfect prompts upfront - Allows course correction as you learn - Builds context naturally - More conversational and efficient Iterative refinement is often more efficient than trying to write the perfect prompt Leverage Copilot's ability to maintain conversation context Each refinement narrows focus and improves precision Great for exploratory work where you're not sure exactly what you need Mirrors natural human problem-solving: broad → specific

## **Why It Matters** - Confirmation bias leads to flawed analysis - Objective data reveals real insights - Better decisions require honest assessment Leading questions can cause Copilot to rationalize your assumptions rather than analyze objectively The model will try to satisfy what it perceives as your expectation This is particularly dangerous in business analysis, competitive research, and strategic planning Ask open-ended questions and let the data speak If you want critical analysis, explicitly ask for it: "What are potential flaws in this approach?" Be especially careful when you have a preferred outcome - that's when leading is most tempting Neutral prompts lead to more reliable, actionable insights

## **Why It Works** - Copilot applies concepts to YOUR context - Adapts ideas rather than copying templates - Encourages contextual intelligence "Like" is a powerful word for getting Copilot to understand intent without over-constraining It signals: "Use this as inspiration, but make it fit my situation" Particularly useful when you have a general idea but want Copilot to contextualize it Compare: "Do it EXACTLY like X" vs "Do something LIKE X but for my context" Works great for: writing styles, document formats, analysis approaches The model can access your actual work data and apply the concept appropriately Gives Copilot room to be intelligent rather than just following a rigid template Use "like" when you want conceptual guidance with contextual adaptation

## **Perfect For** - Consolidating team updates for leadership - Creating consistent project dashboards - Standardizing cross-functional reports - Preparing board-level summaries Normalization is a powerful use case for M365 Copilot in reporting workflows Instead of manually reformatting different styles, Copilot can standardize them Maintains information integrity while ensuring consistency Particularly valuable for managers consolidating team inputs Saves hours of manual reformatting and rewriting Can establish organizational standards while respecting individual work styles Great for creating executive dashboards from varied sources Helps ensure leadership gets information at the right level of abstraction

Live demos with AI are always an adventure The demo gremlin is real - embrace the uncertainty Great opportunity to show real-world behavior and troubleshooting

Prompt Library eliminates reinventing the wheel for common tasks Creates organizational knowledge around effective AI use Enables less experienced users to leverage expert-level prompts Drives consistency in outputs across teams Acts as training tool - learn from successful prompts Great for onboarding new Copilot users

Scheduled prompts transform Copilot from reactive to proactive Automates repetitive information gathering tasks Ensures you never forget routine status checks or updates Delivers insights when you need them, not when you remember to ask Great for managers who need consistent reporting Reduces cognitive load - one less thing to remember

Context pollution is a real issue - keeping old conversations active degrades quality Starting fresh helps Copilot focus on what matters now Think of it like clearing your workspace before starting a new task This is especially important in M365 Copilot where you're pulling from so many sources

Work mode keeps everything within your organization's security boundary Web mode gives you broader knowledge but no access to your company data Many users don't realize they can toggle between these modes This is a key feature that maintains data security while providing flexibility

Temporary chat ensures your conversation won't be saved to chat history Especially useful when testing prompts with real client names or sensitive info Great for "what if" scenarios you don't want documented Think of it as incognito mode for Copilot Important Caveats Enterprise Governance: In Microsoft Copilot, even Temporary Chats may be subject to retention, auditing, or eDiscovery under organizational policies. [marcotran.com.au] Not Fully Anonymous: While not used for personalization or training, platforms may temporarily store data for abuse prevention or compliance. [help.openai.com]

The shield/protected indicator is critical for enterprise users It confirms that Microsoft's commercial data protection is in effect Without the shield, data may not have the same enterprise protections Users should always verify the shield is present when handling sensitive info This is especially important in organizations with strict compliance requirements

Memory feature allows Copilot to learn your preferences over time Unlike context window which is temporary, memory persists across chats Great for storing long-term preferences that apply to all interactions Can be managed - add new memories or delete outdated ones Helps Copilot provide more personalized, relevant responses Particularly useful for consistent tone, format preferences, and role context Be mindful of what you store - review periodically for accuracy

Custom instructions are explicit rules you define upfront Memory is implicit learning from your actual usage patterns Together they create a highly personalized Copilot experience Custom instructions are great for consistent requirements (tone, format, role) Memory captures nuances that emerge from how you actually work Example: Custom instruction says "be concise", Memory learns you prefer bullet points Both persist across chats but can be updated independently This combination makes Copilot feel increasingly tailored to your work style

GPT-5.2 represents a significant leap in AI capabilities Available to M365 Copilot users with appropriate licensing Great for complex reasoning, analysis, and creative tasks May have different performance characteristics - test with your use cases

Quick response is the workhorse - handles 90% of everyday tasks efficiently Think deeper is for when you need the extra reasoning power Auto mode learns when to use which approach Consider cost/performance tradeoffs in your organization Test different modes to understand which works best for your specific use cases

All Chats is your personal knowledge repository of AI interactions Powerful for finding that one perfect response from weeks ago Search functionality makes it easy to rediscover insights Helps you learn what prompts work best over time Great for maintaining continuity across projects Think of it as your searchable AI conversation archive

M365 Copilot Search transforms how you find information across the enterprise Traditional search relies on exact keyword matching - Copilot understands what you mean Aggregates information from emails, Teams, SharePoint, OneDrive, and more Provides synthesized answers with citations instead of forcing you to click through links Saves significant time in knowledge work by reducing the search-to-answer cycle

M365 Copilot Analyst is designed for data-driven decision making Democratizes data analysis - no advanced Excel or analytics skills needed Can handle large datasets and complex calculations Great bridge between business users and data teams Works seamlessly with Microsoft's data ecosystem https://catalog.data.gov/dataset?q=&sort=views_recent+desc

M365 Copilot Researcher transforms how you gather and synthesize information Saves hours of manual research by aggregating and summarizing from multiple sources Provides proper citations so you can verify and trace back to original sources Combines web search with your organizational knowledge base Ideal for consultants, analysts, and anyone doing knowledge work

Copilot Pages bridge the gap between ephemeral chat and persistent documents Unlike chat which is sequential, Pages give you a canvas to develop ideas spatially You can create a page from any chat response, then continue working on it Pages support real-time collaboration - multiple people plus Copilot working together Great for when chat conversations evolve into something worth preserving and sharing Think of it as "chat graduate to document" - start conversationally, end with artifacts Can reference the page in future chats or continue editing with AI inline Combines the best of both: conversational AI + structured documentation

Create functionality extends Copilot beyond analysis to content generation Image generation uses DALL-E integration for professional visuals Great for presentations, reports, and marketing materials Can specify style (photographic, illustration, minimalist, etc.) Include details like perspective, lighting, colors, and composition Useful when you need custom visuals but don't have design resources Remember: Generated images should align with your organization's brand guidelines Can create multiple variations and refine based on feedback

Brand Kit ensures all Copilot-generated content aligns with organizational branding Particularly important for images, presentations, and documents shared externally Upload once, apply automatically to all future creations Helps maintain professional, consistent brand identity Great for marketing teams, client-facing roles, and communications Reduces back-and-forth with design teams for brand compliance Can include multiple brand kits for different sub-brands or campaigns Works especially well with Create functionality for images and presentations Administrators can set up organization-wide brand kits for consistency

Notebooks solve the problem of losing valuable chat history and context Perfect for long-term projects where you need to maintain continuity Acts as a collaborative workspace where teams can build on AI insights Great for research projects, planning documents, and knowledge management Can be shared across teams to democratize AI-generated insights

Agents are the next evolution of M365 Copilot - specialized AI assistants Built using Copilot Studio with no-code/low-code tools Can be configured with specific knowledge bases, tone, and workflows Reduce repetitive questions and standardize processes across teams Enable subject matter experts to scale their knowledge organization-wide