Microsoft has bet big on Copilot. It's integrated across Microsoft 365, embedded in Windows, and positioned as the AI assistant that will transform how you work. At $30/user/month on top of your Microsoft 365 license, it's also a significant investment.
Copilot can draft emails. It can summarize documents. It can generate slide decks.
What it can't do is remember your business.
This isn't a bug—it's a fundamental architectural limitation that affects what Copilot can accomplish for your organization. Understanding this limitation is essential before you invest.
What Copilot Actually Does
Document-Level AI
Copilot operates at the document level. Open a Word document, and Copilot can help you write within that document. Open an Excel sheet, and Copilot can analyze that data. Open a PowerPoint, and Copilot can generate slides from that content.
The pattern: Copilot works with what's in front of you. Its context window is the current document, email, or conversation.
Microsoft's documentation describes this clearly: Copilot "grounds" itself in your Microsoft Graph data—your calendar, emails, chats, documents. It can pull information from these sources when prompted.
The limitation: Pulling information when prompted is different from understanding organizational context. You still need to ask the right questions. Copilot still needs to search and retrieve. The burden of context remains on you.
Task Assistance vs. Strategic Understanding
What Copilot does well:
- Draft an email (you provide the context)
- Summarize a meeting (that just happened in Teams)
- Generate a slide deck (from content you specify)
- Analyze data (in the spreadsheet you're looking at)
What Copilot can't do:
- Understand your strategic priorities
- Remember decisions made across multiple projects
- Connect today's work to organizational goals
- Learn from your organization's history
This is task assistance. It makes individual tasks faster. It doesn't make your organization smarter.
The Context Gap
The Question Test
Ask Copilot: "Based on our strategic priorities and current project portfolio, which initiative should we prioritize this quarter?"
Copilot can't answer this meaningfully. It doesn't know your strategic priorities unless they're in a document it can find. It doesn't understand your project portfolio holistically. It doesn't know what "should" means in your organizational context.
What you'll get: A generic response about how to prioritize, or a request for you to provide the context it needs.
What you need: An AI that already knows your strategy, understands your projects, and can reason about priorities based on organizational reality.
This is the context engineering problem. Copilot optimizes for better prompts on limited context. Context engineering solves for building comprehensive organizational memory.
The Memory Test
Ask Copilot: "Why did we decide not to pursue the enterprise market last year?"
Copilot will search your documents and emails. If someone wrote that decision down and you can find it, Copilot might retrieve it. If the decision was discussed in meetings, Slack, or hallway conversations—if it was understood but not documented—Copilot has nothing to search.
The limitation: Copilot's memory is your documented content. Organizational knowledge that exists in people's heads, in undocumented decisions, in cultural understanding—all invisible to Copilot.
Research shows that 80% of organizational knowledge is tacit—in people's heads rather than systems. Copilot can only work with the 20%.
The Connection Test
Ask Copilot: "How does the marketing campaign we're planning connect to our product roadmap priorities?"
Copilot can't synthesize across domains unless you've documented the connections. Marketing data lives in marketing tools. Product data lives in product tools. Even if both are accessible through Microsoft Graph, Copilot doesn't understand the strategic relationships between them.
What you need: AI that understands how different parts of the organization connect, that can trace from strategic goals to departmental initiatives to individual projects.
This requires organizational memory architecture—not just document retrieval.
The Architecture Problem
Retrieval vs. Understanding
Copilot uses RAG (Retrieval Augmented Generation): when you ask a question, it retrieves relevant documents and generates a response based on what it found.
Retrieval limitations:
- Must find the right documents (search quality matters)
- Must interpret correctly (context within documents matters)
- Must synthesize across documents (often does poorly)
- Can't use knowledge that isn't documented
Understanding requirements:
- Know how your organization works
- Remember decisions and their rationale
- Connect information across silos
- Learn from patterns over time
RAG is a technique for augmenting AI with data. It's not organizational memory.
Session-Based Context
Copilot sessions are stateless. What you discussed with Copilot yesterday isn't remembered today. Each interaction starts fresh, with you providing (or prompting for retrieval of) the necessary context.
The implication: Every strategic conversation starts from zero. You can't build on previous AI interactions. There's no cumulative organizational learning through AI.
What organizational memory requires: Persistent understanding that grows over time. AI that remembers not just documents but interactions, decisions, patterns.
Siloed Intelligence
Copilot in Word knows what's in your Word document. Copilot in Teams knows what's in your Teams conversation. Copilot in Outlook knows what's in your email.
The silos persist: Even within Microsoft 365, AI intelligence doesn't flow naturally across applications. The same fragmentation that affects human work affects AI work.
What's needed: AI that understands the full organizational picture—not AI fragmented by application boundaries.
What Real AI Memory Looks Like
The Four Layers of Organizational Memory
The Context Compass framework describes what comprehensive AI memory requires:
Working Memory: Real-time awareness of what's happening now—current projects, active conversations, today's priorities.
Episodic Memory: Historical knowledge of what happened—decisions made, projects completed, lessons learned.
Semantic Memory: Structural knowledge of how things work—processes, policies, organizational structure.
Procedural Memory: Operational knowledge of how to do things—workflows, best practices, established procedures.
Copilot touches Working Memory (current context) and can retrieve from Semantic Memory (documented policies). It lacks true Episodic Memory (organizational history) and Procedural Memory (learned workflows).
From Task Assistant to Strategic Partner
Task assistant (Copilot level):
- "Draft an email to the client"
- "Summarize this document"
- "Create a slide from this data"
Strategic partner (organizational memory level):
- "Based on our strategic priorities and this client's history, what should we propose?"
- "Given our past project experiences, what risks should we watch for here?"
- "How does this initiative align with our quarterly goals and resource constraints?"
The gap between these isn't marginal improvement—it's fundamental capability difference.
The Cost Calculation
What You're Paying For
Copilot pricing: $30/user/month for Microsoft 365 Copilot
100-person company: $36,000/year for Copilot licenses (on top of Microsoft 365 base licenses)
What you get: Document-level AI assistance. Email drafting. Meeting summarization. Slide generation.
What You're Not Getting
- Strategic AI that understands organizational context
- Historical memory that learns from your decisions
- Cross-functional intelligence that connects silos
- Cumulative learning that grows over time
The ROI Question
Is Copilot worth $30/user/month? That depends on how you measure.
If you measure task efficiency: Maybe. Drafting emails faster has value. Summarizing meetings saves time.
If you measure strategic capability: No. Copilot doesn't provide strategic intelligence. That capability remains on humans.
The opportunity cost: $36,000/year invested in Copilot is $36,000 not invested in AI with actual organizational memory. The question isn't just "is Copilot valuable?" but "is Copilot the best use of AI investment?"
The Alternative Path
Building Organizational Memory
Instead of accepting document-level AI as the ceiling, some organizations are investing in true organizational memory:
Unified work platforms: Where documents, projects, decisions, and communication exist in one system—making AI context comprehensive by default.
Context engineering systems: Where organizational memory is built deliberately across the four layers, not retrieved piecemeal from documents.
Learning AI systems: Where AI capabilities grow with organizational history, becoming more valuable over time.
This is what Waymaker is building—AI that remembers your organization, not just your documents.
The Integration Question
"But we're invested in Microsoft. Can organizational memory work with M365?"
Yes and no.
What's possible: AI systems that augment Microsoft 365 by providing the organizational memory layer M365 lacks.
What's limited: Microsoft's architecture silos data in ways that make comprehensive memory difficult. True unification may require different foundations.
The strategic choice: Optimize within M365's constraints, or invest in platforms designed for organizational intelligence.
Experience Real AI Memory
Want to see what AI looks like when it actually remembers your business? Waymaker Commander brings organizational memory to work management—AI that knows your strategy, remembers your decisions, and connects context across your work.
The result: Ask strategic questions and get strategic answers. Build on previous interactions. Watch AI capability grow with your organization.
Register for the beta and experience the difference between task assistance and organizational intelligence.
Copilot is document-level AI marketed as workplace transformation. It does document tasks well. It doesn't provide organizational memory, strategic understanding, or cumulative learning. Understanding this gap is essential for making informed AI investments. Learn more about our Context Compass framework and explore how context engineering builds the organizational memory AI actually needs.
Stuart Leo has evaluated enterprise AI solutions for Fortune 500 companies. He's the author of Resolute and founder of Waymaker, designed for organizations that need AI with actual organizational intelligence.
About the Author
Stuart Leo
Stuart Leo founded Waymaker to solve a problem he kept seeing: businesses losing critical knowledge as they grow. He wrote Resolute to help leaders navigate change, lead with purpose, and build indestructible organizations. When he's not building software, he's enjoying the sand, surf, and open spaces of Australia.