Introduction: The Knowledge Bottleneck in SME Construction and Manufacturing
Small and medium enterprises (SMEs) in industries like construction and manufacturing consistently generate steady profits but are often undervalued, typically selling for just three to five times EBITDA. The key issue is what we call “subverbal IP”—the deep, experience-driven knowledge locked in the minds of owners and key employees. When these individuals leave, their expertise goes with them, leaving the business without a scalable or transferable foundation, which significantly reduces its market value.
An emerging solution is Retrieval-Augmented Generation (RAG), a specialized AI framework that transforms proprietary knowledge—project archives, emails, meeting notes, and subverbal IP—into a structured, living knowledge base. Unlike traditional data storage, RAG doesn’t just retrieve information; it learns from patterns embedded in the business, much like a seasoned apprentice learning from an expert. By capturing and systematizing this intuitive expertise, RAG makes it accessible, transferable, and actionable—enhancing daily decision-making while significantly increasing the business’s long-term valuation.
Why Traditional Software Falls Short
Most off-the-shelf software—CRMs, project management tools, and standard AI applications—fail to capture subverbal IP. They track invoices and schedules, but they don’t learn from experience the way a senior project manager or skilled tradesperson would.
Basic AI chatbots have been frustrating for years, and even advanced AI agents used in real estate—automating MLS data scraping, generating reports, and scheduling follow-ups—are limited because they rely on publicly available information. Within a year or two, AI models like OpenAI’s autonomous agents will generate these workflows on demand, making pre-built AI automations easily replicable and ultimately replaceable.
The real competitive advantage doesn’t come from automation alone—it comes from capturing and structuring domain-specific knowledge that LLMs like ChatGPT cannot access or replicate. When properly designed, an AI system doesn’t just retrieve information—it learns from patterns, recalls experience, and improves over time.
RAG: The AI Apprentice That Captures and Applies Subverbal IP
The key to unlocking this intelligence is Retrieval-Augmented Generation (RAG)—a system that enables AI to store, retrieve, and apply proprietary business knowledge in real time. Much like an apprentice learning on the job, RAG observes patterns, recalls past experiences, and applies them to new challenges.
Unlike traditional AI models that pull from public sources, a RAG system is built from your company’s internal knowledge—past project documents, emails, supplier negotiations, and site drawings. When a user asks a question, RAG retrieves the most relevant knowledge and augments it by injecting contextually relevant insights into the AI’s processing window. This allows a language model (LLM) like Cohere or ChatGPT to generate an expert-level response, uniquely tailored to your business.
How AI Captures Subverbal IP: The Power of Pattern Recognition
At its core, AI doesn’t “understand” information the way humans do—it recognizes patterns. This is where similarity search, a fundamental AI technique, comes in. When a contractor faces a new challenge, they don’t memorize every past job—they recognize familiar situations and apply what they’ve learned. AI does the same thing, but at scale.
RAG ensures that when a question is asked, AI retrieves insights grounded in real experience. Think of an apprentice on a job site—when faced with an unfamiliar problem, they don’t invent solutions; they recall past guidance from their mentor. Likewise, AI finds and reformulates the most relevant insights from your company’s archives. Over time, it becomes better at identifying patterns, surfacing expertise, and applying past knowledge to new challenges.
A Defensible AI Strategy for SME's
By embedding institutional knowledge directly into AI-driven decision-making, RAG turns AI for builders from a basic search tool into an invaluable domain expert. Beyond storing expertise, an AI-powered knowledge base unlocks a full ecosystem of AI-driven tools:
Service ticket chatbots that streamline workflows
Email assistants that manage communication
Training and onboarding chatbots that accelerate skill development
Workflow automation agents that handle complex, repetitive tasks
Perhaps the most compelling advantage? This AI strengthens over time. With every interaction, it refines its understanding, becoming more personalized and indispensable. A company that structures its AI correctly today may find that, within a year, the system is so deeply integrated that replacing it would mean losing a critical operational asset.
But Wait—Can We Trust AI for Builders?
For the RAG knowledge base to become a critical operational asset, it must be trusted and deliver consistent value. Concerns about AI's reliability—especially its tendency to hallucinate—are valid but often stem from how AI is trained. The principle of "garbage in, garbage out" highlights that the quality of input data directly affects AI performance.
A compelling example of trustworthy AI is OpenEvidence, an AI-powered medical search engine. It designed for point-of-care solutions where Doctors are with patients who may be in life-threatening situations.
Launched in early 2024, OpenEvidence has been rapidly adopted by over a quarter of U.S. doctors, with more than 35,000 verified physicians registering each month. Its success is attributed to its training on gold-standard medical content, including 35 million peer-reviewed publications and partnerships from the most trustworthy sources in the world like The New England Journal of Medicine.
This rigorous approach ensures that OpenEvidence provides reliable, evidence-based information, thereby earning the trust of the medical community. When David Nadler, the founder of OpenEvidence, was asked how he created such a reliable database, he replied: "Gold in, Gold out." This example underscores a fundamental truth: trustworthy AI is not a coincidence; it is a design choice. By training AI systems on high-quality, relevant data, businesses can develop reliable tools that stakeholders can trust.
Relevance Memory: Making RAG Feasible for SMEs
Implementing RAG systems can be particularly challenging for SMEs due to limited resources and technical expertise. However, the concept of Relevance Memory developed by Scelta in collaboration with the Supply Chain & Business Data Analytics department at the Odette School of Business, offers a practical solution. This approach involves developing AI's contextual understanding through conversational interactions, allowing subject matter experts within the company to iteratively refine and manage the AI's knowledge base without the need for specialized AI engineers.
By leveraging platforms like Cohere's LLMs and RAG capabilities, SMEs can:
Preprocess and Structure Data: Automatically filter, tag, and clean collaboration data before ingestion, minimizing manual preparation.
Reduce Noise Conversationally: Enable AI and SMEs to collaboratively refine relevant data in real-time, eliminating the need for complex evaluation sets.
Facilitate Continuous Improvement: Allow SMEs to manage and audit AI-driven knowledge systems without deep technical expertise, making AI adoption more accessible.
This approach not only democratizes AI implementation but also ensures that the AI system evolves in alignment with the company's unique operational knowledge and culture.
Conclusion & Call to Action
RAG isn't just another AI trend—it's a transformative framework for preserving, refining, and scaling the unique expertise that defines your business. For SMEs committed to staying competitive, the message is clear: now is the time to structure and preserve your proprietary knowledge.
Early adopters of AI technologies like RAG stand to gain significant advantages, as these systems become deeply integrated into daily operations, enhancing efficiency and decision-making. Implementing RAG through Relevance Memory offers a straightforward, accessible way to integrate advanced AI capabilities without the need for a full in-house AI development team.
Let's continue the conversation.
Follow John on LinkedIn
More Scelta
Scelta is the new industry standard for pre-construction sales, marketing, and client relationships.
Shop DIAL IN Merch
For those who set the standard.
Hats, sweaters, and tees available.
[Shop Now] www.DIAL-IN.ca
Comments