So, I know that what I am about to say is going to come as somewhat of a shock. And so, it is with a sense of shame that tells you this: I wrote some of this series of articles using AI. Generative AI, nonetheless.
In fact, I used Retrieval Augmented Generation. I did. Hi, I’m Seth, and I am a RAGhead.
Uh, hm…. Maybe that joke didn’t come out quite right.
Annnnneeeeyyyyhow. I mention this because of the following 3 reasons:
I needed an introduction to this article, a “cold open” if you will, insofar as a business article about business technology can have one.
I wanted to show you 100% genuine results from a RAG process
I wanted to show you that low-rent RAG can be a highly useful, simple process. I’ll show you the “simple” later, but for now, take it on faith. The table below shows you some basic, useful market research that you can take to your boss or your team and with which, you can make some information backed decisions. Oh, the RAG process and analysis it can be far more complex and powerful, but I generated useful results by using the bog-standard Claude souped up with the ability to talk to Perplexity.ai. Writing all the “humor” (you can see the air quotes, yes?) took substantially more time than the substance below. Writing this bullet literally (in the literal sense of literally, not the whatever it’s become nowadays) took more time.
Just think about that for a moment when you look at the table. I’m not exaggerating at all. Of course, it may have taken me a long time to write that last paragraph, but I promise that I’m not cheating like that.
Before we venture into the table, I just want to take a second to re-set the stage, in case you’re just joining other viewers here in this episode, and also because it frames the applications below in terms of what we’re thinking for mid-market enterprises, not just sorta “B2B.”
Here we go!
Why Does Traditional AI Fall Short for Mid-Market Companies?
Using genAI can feel like bringing a Swiss Army knife to build IKEA furniture. Sure, there’s a tool for everything, but none of them quite fit the job at hand. Not only is there no hex bit, but there’s a distinct lack of marriage counselling.
What represents the metaphorical lacking hex bit for genAI/LLMs in our imaginary flat-pack IKEA furniture?
The Hallucination Problem: LLMs often generate confident-sounding but incorrect information when they lack proper context about your business. Sometimes colloquially known as the “WTF?” problem.
Corporate Blindness: No, this isn’t what happens when you’re promoted into the C-Suite. LLM’s have zero knowledge about your unique products, processes, and tribal knowledge that makes your business special. Of course, one might say that about the C-Suite as well. (Hey now – I can make those jokes; I have friends who are in the C-Suite.)
Source Trust Issues: Generic LLM responses lack an automatic connection to your authoritative business sources, leaving you wondering, “Where did that come from?”
ROI Question Mark: The cost-to-value ratio can be difficult to justify when generic AI solutions don’t address your specific business challenges. This, I think, is an enormous problem that lurks right beneath the surface of genAI right now–it’s stalling out some, people are struggling with how and where to deploy it to solve actual problems for them.
Also, I bet you blew right past this without noting it, but isn’t it amazing that I can say “traditional generative AI approaches” with a straight face? The industry is moving quickly right now. Unsustainably so, IMHO, but we’ll talk about that more another time.
The RAG “Sweet Spot” Advantage
For mid-market companies, RAG helps hit that perfect balance–like finding business casual attire that’s both comfortable AND impressive in meetings. (Take the Berluti Cashmere Stretch Travel Pants as an example. They are $1650 of yum.) Share this article with your friends so we can get the “put Seth in cashmere sweats” fund going.
The description (not the cashmere sweats analogy, but the whole “comfortable AND impressive” phrase, is an apt description to keep in mind when considering RAG. The comfortable comes from the LLM, and the impressive comes from your data–the context–that you bring to the party.
No Data Left Behind: It leverages your existing data investments without requiring expensive model training.
Fact-Checking Built In: It produces accurate responses grounded in your verified information, not creative fiction. If you are generating fiction, the RAG system can help ground your fiction in fact.
Grows With You: It incorporates new information as your knowledge base expands.
IT-Friendly: It requires less technical expertise to implement than custom genAI solutions.
RAG combines the best of both worlds—the factual accuracy of traditional information systems with the conversational, human-like responses of generative AI. It’s like having an expert employee who’s read every document in your company and does a pretty job of remembering the right documents at the right time.
I like to look at it like this. GenAI has a circle of problems that it can solve. You have a circle of problems. RAG enlarges the GenAI circle, engulfing more of the “your problems” circle. Perhaps it kinda skooches the two of them closer together. All of this ensures that the overlap is larger, making for a diagram that would tingle Dr. Venn’s toes.
Because I was curious when I was writing this article, I read a bit about the eponymous John Venn. (And I was looking for a good Venn diagram joke – sadly, the set of good Venn diagram jokes and the set of bad Venn diagram jokes overlap and make a geometrically perfect circle)
Did you know the Venn diagram has been around for over 150 years? Also, this stained-glass window is in the university’s chapel he worked in. There were also 3 other beautiful panes, but that’s a different story for a different time. One I’m not sure I want to touch any more than this passing mention.
RAG Applications/Quantifiable Benefits Table
Ok, here’s the promised table that I made you work for. a bunch of different applications for RAG, along with references and quantifiable benefits (or benefits that are one step away from being quantifiable, where you’d need to pick an aspect to focus on.)
| Category | Application | Description | Benefits | Sources |
|---|---|---|---|---|
| Internal Enterprise – HR | Leave Management | Cross-references employment histories with policies to generate personalized responses | Accurate calculation of leave entitlements, automated compliance | AWS, Northern Light |
| Recruitment | Analyzes job descriptions against candidate resumes while flagging compliance issues | Better candidate matching, reduced legal risks | IBM, TheBlue.ai | |
| Training Programs | Creates personalized development paths aligned with roles, performance metrics, and skill gaps | Dynamic content generation, targeted skills development | Beyond Chatbots, Reworked | |
| Internal Enterprise – Operations | Manufacturing Troubleshooting | Connects IoT sensor data with maintenance logs to generate repair guides | Real-time visibility, faster resolution of equipment issues | AWS, Denser AI |
| Vendor Risk Assessment | Compiles financial stability, delivery performance, and compliance histories | Comprehensive risk profiles, automated due diligence | Forbes, TheBlue.ai | |
| Supply Chain Management | Processes weather data, shipping manifests, supplier communications to predict disruptions | Proactive risk mitigation, resilient supply chains | Beyond Chatbots, Orases | |
| Internal Enterprise – Knowledge | Enterprise Search | Retrieves relevant passages from meeting transcripts, documentation, emails | 40% faster contract reviews, improved information discovery | AWS, Northern Light |
| R&D Innovation | Synthesizes internal data with patent filings and academic papers | Accelerated innovation, competitive awareness | LinkedIn, WEKA | |
| External Enterprise – Customer | E-commerce Personalization | Retrieves product specs, reviews, inventory data for tailored recommendations | Hyper-personalized shopping experiences | CustomGPT, Glean |
| B2B Communications | Drafts client communications enriched with project histories and market intelligence | Better prepared sales teams, contextual client interactions | Salesforce, Orases | |
| External Enterprise – Compliance | Regulatory Monitoring | Monitors regulatory updates across jurisdictions to generate guidance documents | Real-time compliance, reduced regulatory risk | IBM, Northern Light |
| Risk Assessment | Correlates applications with geospatial data, claims histories, IoT device outputs | 60% reduction in manual research, improved coverage accuracy | Forbes, Denser AI | |
| External Enterprise – Marketing | Global Brand Consistency | Retrieves style guidelines, performance metrics, cultural norms for campaign copy | Culturally appropriate messaging, brand consistency | Salesforce, LinkedIn |
| Crisis Response | Uses real-time news/social media monitoring to draft AI-generated statements | Fast, empathetic, compliant crisis communications | Glean, WEKA | |
| Industry-Specific – Healthcare | Diagnostic Assistance | Correlates patient symptoms with medical research for differential diagnoses | Evidence-based diagnosis support, improved clinical outcomes | LinkedIn, WEKA |
| Drug Discovery | Analyzes clinical trial data against compound libraries and adverse event reports | Accelerated research timeline, better drug candidates | Northern Light, Reworked | |
| Industry-Specific – Legal | Contract Management | Compares draft agreements against regulations and precedent clauses | Faster contract cycles, reduced legal risk | AWS, IBM |
| Audit Enhancement | Verifies financial statements against tax code updates and SEC requirements | Higher audit quality, automated compliance checks | Forbes, Orases | |
| Industry-Specific – Manufacturing | Predictive Maintenance | Integrates equipment manuals with sensor telemetry for repair prioritization | Reduced downtime, optimized maintenance scheduling | Denser AI, Beyond Chatbots |
| Logistics Optimization | Factors weather patterns, customs clearance times, carrier performance histories | Up to 18% fuel cost reduction, optimized routing | Databricks, Orases |
Hey! A Conclusion!
There’s not really much to conclude about here. RAG good. RAG applicable. RAG make money.
You read next article: Article 4/7: RAG: Starting the Implementation Process