Latest / Future of Work Tech with Fexingo: Remote Tools, AI Productivity, and Workplace Software / How AI Is Automating Your Internal Knowledge Graph
Transcript
- Lucas: Luna, I want to talk about something that sounds like a Wikipedia back-end but is actually one of the most under-hyped productivity shifts happening inside companies right now. Luna: Okay, I'm intrigued. What is it? Lucas: The internal knowledge graph. Not a wiki, not a shared drive, not an AI chatbot that searches documents. I'm talking about a structured map of who knows what, which documents are related to which projects, and how expertise flows across teams. Luna: So instead of folder structures, you get a web of connections. I've seen those diagrams. They look beautiful. But do they actually work? Lucas: They do, but only when done right. Let me give you a concrete example. I talked to a mid-sized software company—about 800 employees—that rolled out a graph-based knowledge tool earlier this year. Their goal was to cut new-hire ramp time. Before, a new engineer would spend four to six weeks just figuring out who to ask for what. After three months with the knowledge graph, that ramp time dropped by about 30 percent. Luna: Thirty percent in three months is significant. How did they do it? Lucas: They started by ingesting every document, every Slack message, every Jira ticket, and every meeting transcript. Then the AI built a graph: nodes for people, documents, projects, and skills. Edges connecting them based on who contributed to what, who commented on which document, who was tagged in which ticket. So when a new hire asks, 'Who knows about our authentication API?', the system doesn't return a list of keywords. It returns a ranked list of people with a confidence score, plus the five most relevant documents, plus related projects. Luna: That sounds powerful. But I'm guessing the hardest part wasn't the technology. Lucas: You're exactly right. The hardest part was getting employees to contribute their tacit knowledge. The graph is only as good as the data you feed it. If people don't tag their documents, don't update their profiles, don't mark which projects they're working on, the graph stays sparse. This company actually had to incentivize contributions—they made it part of quarterly goals to add at least five connections per month. Luna: That feels like a recipe for gaming the system. People will just link random things to hit a number. Lucas: They did at first. But the graph has a self-correcting mechanism. If you link yourself to a project you never worked on, the system can detect it because your name never appears in the associated Slack channels or documents. So the AI can flag low-confidence connections. Over time, the graph actually becomes more accurate as the AI learns which connections are real. Luna: Let me push on the privacy angle. If this graph maps who knows what, and who talks to whom, isn't that essentially surveillance? Managers could use it to see who's not connected enough, who's a bottleneck. Lucas: That's the biggest concern, and it's valid. The company I spoke with addressed it by giving employees control over their own nodes. You can mark certain connections as private—for example, if you're mentoring someone but don't want that publicly visible. You can also opt out of the graph entirely, though then you lose the benefits. The key is transparency: everyone knows what data is collected, and they can see their own graph. Luna: So it's opt-in, not mandatory. That seems fair. But does the graph actually reduce the time people spend on internal communication, or does it just shift where they look? Lucas: In their case, it reduced the number of 'who knows about X' messages in Slack by about 40 percent. People just typed their query into the graph and got an answer. That saved maybe 15 minutes per person per week. Across 800 employees, that's 200 hours a week. Not negligible. Luna: I wonder how this scales. A startup with 50 people probably doesn't need a graph—they can just shout across the room. At what employee count does this become worth it? Lucas: General consensus among the people I've talked to is around 200 to 300 employees. That's when the informal knowledge network breaks down. You can't know everyone. You start getting duplicate work because two teams independently solve the same problem. That's exactly when a knowledge graph starts paying off. Luna: So for our listeners who are at companies approaching that size, what's the first step? Do you need a dedicated knowledge graph platform, or can you build it with existing tools? Lucas: You can start small. Some companies build a lightweight graph using a tool like Notion with backlinks, or even a custom Google Sheets that maps people to skills. The AI piece becomes valuable when you have thousands of documents and hundreds of people. But the conceptual shift—thinking in terms of connections instead of folders—that's free. You can start today by asking your team to tag documents with related people and projects. Luna: I like that. It's a mindset shift, not just a tech purchase. Lucas: Exactly. And honestly, if today's conversation gave you something usable—maybe you're thinking about your own company's knowledge sprawl—that's the kind of thing that makes this show worth doing. If it was worth a coffee to you, there's a link: buy me a coffee dot com slash fexingo. No pressure, just a way to keep the show ad-free and focused on stuff like this. Luna: Yeah, I think a lot of people get value from these deep dives. So thanks for considering that. Lucas: Alright, back to the graph. One more thing I want to mention: the next frontier is making these graphs dynamic in real-time. Imagine a graph that updates as you type a document, automatically linking it to related projects. Some startups are already doing that. It's going to make knowledge management feel less like a chore and more like a natural byproduct of work. Luna: So the ultimate goal is a knowledge graph that builds itself. As long as people do their jobs, the connections form. Lucas: Exactly. And that's when it becomes truly powerful—when the system helps you discover what you didn't know you needed to know. Luna: I'm still a little skeptical about adoption. But the 30 percent ramp-time reduction is hard to ignore. Lucas: It's early days, but the trajectory is clear. Knowledge graphs are going to become as standard as shared drives within the next five years. Luna: Alright, Lucas. I'll start tagging my documents. You've convinced me. Lucas: That's all I can ask. Until next time.