ROG AB
Consulting, Technology, AI: Engineering Context, Delivering AI Certainty.
Unlock High-Revenue AI Engagements in Microsoft 365 through Semantic Context Engineering.
The Problem: Information Overload & Suboptimal AI ROI
In today's enterprise, organizations are drowning in information: too many Teams channels, unmanaged documents, proliferating sites, and disparate tools. This unchecked expansion exacerbates existing pain points, creating a paradox where information is abundant yet effectively scarce due to its disorganization.
The stark reality:
- Information workers spend 2.5 hours per day searching for information.
- Poor data quality costs the U.S. economy $3.1 trillion annually.
- Over 2 million new SharePoint Sites are created daily, contributing to data sprawl.
- Organizations hesitate to renew expensive AI licenses (e.g., Microsoft Copilot at $30/user/month) when AI cannot effectively navigate their unstructured data.
Many organizations attempt AI with a "crawl everything and hope for the best" strategy. This is inefficient, costly, and rarely delivers the promised ROI because AI lacks the semantic context to understand the 'why' and 'what' behind the data.
What We Do: Bridge the Semantic Gap
ROG Refocus on Goals AB provides the proprietary methodology, framework, training, and ongoing support to management consulting firms. We empower them to deliver AI certainty by bridging the critical semantic gap in Microsoft 365 environments.
Establish Semantic Layer
We build a dynamic, self-updating semantic layer that makes your data relevant to organizational goals.
Transform M365
M365 becomes a Goal-Oriented Document Management Hub, leveraging social media incentives for engagement.
Empower AI
We ensure AI models are fed with highly relevant, authoritative, and validated data for precise results.
Drive Measurable ROI
Our approach leads to quantifiable improvements in efficiency, compliance, and productivity.
Seamless Integration
Solutions are deployed directly into your existing Microsoft 365 environments, including Microsoft Search and Copilot.
Train & Certify Talent
We train and certify your existing consulting workforce in our specialized framework.
ROG's Semantic Context Engineering: Our Proven Methodology
Unlike approaches that simply "crawl and hope," ROG employs a rigorous, structured methodology to ensure your AI understands your business at a profound level. This is how we deliver the AI promise, operating across two main, simultaneously applied dimensions for maximum efficiency and impact.
What is Semantic Context Engineering?
Semantic Context Engineering is defined as: *"Intelligently selecting, structuring, and injecting the most relevant and meaningful information into a language model's context window and managing primary sources to avoid context pollution."*
Why is Semantic Context Engineering necessary?
- Avoids Context Pollution: Simple Retrieval-Augmented Generation (RAG) with keyword and semantic search (via Microsoft Graph for example) often pulls information from non-authoritative and outdated sources. This pollutes the context with duplicate, misleading, and contradicting data, leading to inaccurate AI responses.
- Addresses Complex Context Needs: A large fraction of the necessary context is impossible to retrieve in a single RAG turn from documents and primary sources. It must be collected, structured, reviewed, and updated with human feedback, ensuring a comprehensive and accurate understanding for the AI.
- Reduces Variability and Ensures Deterministic Behavior: AI search results must be replicable. Without consistent and high-quality context, AI agents using enterprise search will not behave in a deterministic way, which in turn will make human oversight impossible. Semantic Context Engineering ensures predictable and reliable AI outputs.
Dimension 1: Technical Creation of the Semantic Layer
This dimension focuses on building the underlying data infrastructure that enables intelligent information management and AI readiness.
-
Incrementally Crawl Your Data Using a Ring-Based Model: Systematically ingest and index all disparate data sources. Instead of a monolithic "crawl everything," we employ a ring-based model of relevance for crawling. This approach delivers faster project results and efficiently extracts the relevant semantic data needed for making sense of subsequent rings (definitions, acronyms, terminology, authority).
- Why a Ring-Based Model? It addresses the difficulty organizations have in identifying critical document locations, delivering incremental value and quick wins.
- Example Ring Structure:
- **First (most relevant) ring:** Org chart, high-level management system, core terminology, values, goals.
- **Second ring:** Business units, staff functions, responsibilities, processes, frameworks.
- **Third ring:** Departments, governance, regulations, policies, published documents/landing pages.
- **Fourth ring:** Projects, initiatives, articles, FAQs, how-tos, videos.
- Re-organize and Tag Sources: Restructure and logically tag information. For mutable data, sources are reorganized in M365 (e.g., SharePoint libraries, Teams files). For read-only/external data, relevant tags/metadata are linked to original sources.
- Build a Self-Updating Semantic Middle Layer: Create a dynamic, intelligent layer that continuously learns and knows everything about your organization's information, ensuring ongoing relevance and accuracy through automatic maintenance.
- Feed AI with Context + Primary Data: Supply AI models (e.g., Copilot, ChatGPT Enterprise) with enriched context *before* accessing primary data, ensuring highly accurate, pertinent, and trustworthy responses.
Dimension 2: Management Consulting for Goal-Oriented Context Engineering with Business Centers
This dimension leverages human intelligence and strategic understanding to imbue data with organizational meaning, creating trusted sources.
- Extract Semantic Context (Human Intelligence): Guides consultants to identify and extract critical metadata (primary sources, authority, validity, ownership, etc.), translating raw data into meaningful business context.
- Measure and Continuously Improve AI Results: Implement exhaustive benchmarks and continuous monitoring to track AI performance against business objectives, driving iterative improvements and ensuring sustained value.
Key Technical Techniques in Detail:
- **Before Crawling:** Prepare org charts with identifiers/tags, implement **Ring Crawling**.
- **Initial Crawling:** Perform **Semantic Context Extraction**, **Acronym/Keyword Collection**, create **Source Tag Tables**, implement **Sensitive Content Detection and Exclusion**.
- **Vectorization:** Apply **Semantic Segmentation** for modular vector stores, perform **Metadata Extraction and Ingestion**, **Information Restructuring**, create **Textual Snippets**.
Key Management Consulting Techniques in Detail:
- **Prompting:** Perform **Prompt/Conversation Analysis**, **Injection of Relevant Context**, develop **Prompt Shields**.
- **Searching:** Enable **Primary and Secondary Source Identification**, apply **Automatic Metadata Filtering**, implement **Query Rewrite** mechanisms, utilize **Semantic Ranking**.
- **After Answer Generation:** Conduct **Evaluation against Provided Context and Instructions**, perform **Contextual Rewrite**.
Direct Connection to Microsoft Search via Custom Microsoft Graph Connector
A critical technical capability of our solution is the direct connection of the semantic layer to Microsoft Search. This is achieved through the ingestion of our structured and contextualized information by a **custom Microsoft Graph Connector**.
Why is this crucial?
- This integration makes the semantically enriched information directly discoverable by Microsoft Copilot and other Microsoft 365 search functionalities.
- It establishes a robust link between the vast array of enterprise documents and their vital business context, significantly enhancing the quality and relevance of AI-generated answers.
- This directly complements the substantial investments CTOs and organizations have been making in Copilot and M365 licenses by ensuring they yield tangible, accurate results.
Example Scenario: Enhancing Workation Policy Search with Copilot
Without Semantic Layer:
A user asks Copilot, "What is our workation policy?" Copilot might pull generic information from various unvalidated sources, leading to a broad but potentially misleading or incomplete answer.
With Semantic Layer:
Our semantic layer, connected via a custom Microsoft Graph Connector, has ingested and contextualized relevant HR policies. It also includes an AI-generated document (created by crawling dozens of individual department sites, reviewed, and approved by HR) that lists all departments that *do not* offer workations. Copilot, leveraging this enriched context, notices that the user's specific department is on that exclusion list. Consequently, Copilot provides a complete and correct answer, stating the general policy and explicitly noting that the user's department is an exception, thus preventing misinformation and improving user experience.
Key Facts & Figures:
- 95% accuracy in AI-driven information retrieval.
- 3x faster deployment of new AI-powered applications.
- Significantly reduced compliance risks through automated policy adherence checks.
- Measurable ROI within 12 months of implementation.
- Improved AI Efficiency: By providing AI with relevant and structured context, the risk of "hallucinations" (incorrect answers) is reduced by up to 90%.
- Increased Consistency: Standardized context management ensures consistent performance for AI systems across different scenarios and users.
- Scalability: Semantic Context Engineering is crucial for building robust, enterprise-grade AI systems that can handle complex use cases and large datasets.
- Enhanced Reasoning Capabilities: AI agents can better remember and prioritize information over long interactions, adapting as goals and context shift, leading to more sophisticated and relevant results.
AI Quality Measurement: The AI Judge on the Bench
The Unspoken Crisis in Enterprise AI
Executives are making huge strategic bets on AI, but a lack of reliable measurement is leading to a crisis of confidence. A staggering 42% of AI initiatives are being cancelled, and 72% of companies struggle with data quality. This isn't just about technical issues; it’s about a fundamental liability where unreliable AI can be worse than no AI at all.
The problem is that many AI initiatives rely on a flawed method of self-evaluation known as "LLM as a judge." Can you truly trust an AI to reliably judge another AI? Our research shows you can, but only with a crucial caveat: it requires a precise, business-driven framework. We've identified that the most common evaluation methods are dangerously flawed, creating a false sense of security.
From Black Box to Business Instrument: Reliable Evaluation
To transform AI evaluation from a black box to a reliable business instrument, we need to manage the inherent variability of AI responses. Our research, based on a calibrated dataset of 60 Question-Answer pairs, reveals that a simple approach is the most effective.
- The Problem with Rubrics: The most common method—a complex, rubric-based prompt—only correctly identified the right answers 42% of the time. This is a shocking failure rate caused by over-engineering instructions, making it inconsistent and unscalable.
- The Winning Method: A simple, direct scoring prompt that compares a model's output to a reference answer and scores it from 0 to 5 is the clear winner. This method focuses on factual correctness and relevance, achieving over 80% reproducibility. This is the foundation of a scalable AI quality assurance process.
The Strategic Imperative: Trust, but Measure
Our methodology has been successfully deployed in a production enterprise RAG system, proving that precise AI evaluation is not only possible but scalable. We also found critical factors that can make or break an AI deployment:
- The Model is the Judge: Different AI models have different evaluation capabilities. For example, we found that OpenAI's GPT-4o consistently delivered more stable and accurate results than GPT-4.1. This highlights the importance of a calibrated framework, especially as models like Microsoft's Copilot are updated.
- Context is King: Including the original question in the evaluation prompt significantly increased accuracy. This shows that small, disciplined changes can have a profound business impact.
For enterprise leaders, the path forward is clear: the era of blind trust in AI is over. Your AI evaluation tools must be treated like any other critical business instrument—calibrated, verified, and documented. We help you move beyond the hype and unlock the full, trustworthy potential of your AI investments by building a foundation of confidence and measurable performance.
The Benefits / A Use Case
Transforming Information Management: A Large Utility Company
A major utility company faced significant challenges managing vast technical documentation and compliance data. Information silos led to delayed decisions and inefficient knowledge transfer.
ROG implemented its Semantic Context Engineering approach, leading to significant improvements:
- Time Savings: 40% reduction in time spent searching for critical information.
- Cost Efficiency: 25% decrease in operational overhead related to data management.
- Collaboration: Enhanced cross-departmental knowledge sharing.
- Productivity: Improved employee satisfaction and access to accurate data.
Who We Are / Our Philosophy
ROG is a team passionate about unlocking the true potential of information and AI. We believe AI's power lies in the quality and context of the data it processes.
Our philosophy is rooted in precision, measurable outcomes, and a deep understanding of your business challenges. We engineer solutions that deliver tangible, transformative results.
Our commitment: empower your organization to thrive in an information-rich world, making data work for you.
Our Team
Meet the dedicated professionals behind ROG, committed to engineering context and delivering AI certainty for your organization.
Jean-Francois Wipf
Founder & Principal Consultant
Jean-Francois brings extensive experience in strategic consulting and digital transformation, guiding clients to achieve their most ambitious goals with AI.
LinkedIn ProfileKarsten Held
Certified Azure AI Engineer & M365 Specialist
Karsten is a highly skilled AI consultant with over 20 years of experience in software development, SharePoint, Power Platform, and M365 solutions, specializing in GenAI.
LinkedIn ProfileStephan Kiöbge
Senior Consultant, Change Management & Implementation
Stephan specializes in driving successful technology adoption through expert consulting, effective change management strategies, and seamless implementation processes.
LinkedIn ProfileContact Us
We'd love to hear from you. Fill out the form below, and our team will get back to you shortly. Please complete the human verification to send your message.