For the latest installment of our ongoing Driving Innovation series, we got back in touch with Jordan Goure, chronic entrepreneur and, most recently, co-founder of the dynamic hiring platform Picsume. According to Goure, if your AI workflow doesn’t start with a knowledge base, you’re already falling behind.
The Big Idea: The knowledge base
In the context of consumer-facing AI-powered chat apps like ChatGPT, Gemini, Claude, or Perplexity, a knowledge base is a structured, centralized repository of specific information (think rules, documents, and baseline facts) that an AI-enabled system access to do its work: “reason,” answer questions, and make decisions.
While a “basic” AI tool (e.g., a new ChatGPT window) relies on a broad-spectrum information library assembled from years of scraping who-knows-what across the internet, an AI tool with a purpose-built knowledge base will prioritize the specific stuff that matter most to you and whatever work you do.
“A knowledge base provides the information layer that helps AI generate accurate, context-aware responses. It doesn’t make AI necessarily smarter, just more informed,” explains Goure. “Instead of re-explaining your business, your tone, or your processes every time you start a conversation, a knowledge base stores that information so the AI can reference it when needed. Think of it as the difference between working with a temp who needs everything explained from scratch and a trained employee who already has access to your company’s documentation and procedures.”
Alternatively, if we can borrow a different use of the key word, think of it as starting every query on third base: if a lot of the groundwork has been laid before your step up to the plate, the path to your goal becomes much shorter.
Connect the Bots: What your knowledge base can include
Any knowledge base will be populated by documents of various varieties; for example, custom GPTs (tailored versions of OpenAI’s ChatGPT tool built for specific purposes) permit direct file uploads. Popular static document formats include PDFs, Microsoft Word and Google docs, text-based training manuals, and standard operating procedures (SOPs).
The strongest possible knowledge base will also “talk to” several of the systems you already use in your work by integrating external software platforms. Goure has witnessed successful integrations of workplace-management, corporate communications, and productivity apps like Notion, Hubspot, Slack, Google Drive, and Airtable. Moreover, according to Goure, some platforms let you “deliver” live data, recorded in real time, directly into your AI workflows.
Keep It Simple: How to set up a knowledge base for yourself
Goure recommends a straightforward four-step process to construct a new knowledge base:
- Gather your documents: Comb through your company’s hard drives and cloud storage to identify existing SOPs, frequently asked questions (FAQs), product descriptions, brand guidelines, past emails, and policies that are still valid and relevant to your work.
- Upload the documents or approve direct connections to cloud apps to share them with your AI tool (e.g., ChatGPT, Claude, etc.).
- Test the AI by asking it a question your proprietary documents should answer; see if it responds correctly.
- Refine the process over time by cleaning up any outdated or conflicting information and re-uploading newer, better files.
Dos and Don’ts: How to take out the garbage
Are you familiar with the “GIGO” concept? Goure thinks you should be. GIGO, which stands for “garbage in, garbage out,” is a fundamental principle of computing. While the underlying idea—the quality of an agent’s input into a computer system determines the quality of that system’s output—has existed in some essential form for about 200 years, programmers, engineers, and mathematicians have been using the abbreviated version since at least the 1960s. In the context of machine learning and what we now call generative AI, savvy operators understand that poor-quality training data and prompt construction will invariably result in a poorly performing model.
Here are a few of Goure’s best tips for ensuring the next knowledge base you build sets your operation up for success:
Do:
✓ Use clean, clearly written documents.
✓ Keep information as current as possible.
✓ Use consistent formatting.
✓ Clarify context (who, what, and why) whenever possible.
Don’t:
✗ Upload “messy” files, or multiple files that contradict each other.
✗ Let outdated or obsolete documents “sit” in your knowledge base.
✗ Mix different naming conventions across your system.
✗ “Dump” heaps of raw data without adding structure.
Innovation at Work: How Goure puts knowledge into practice
According to Goure, Picsume’s platform is a knowledge base built around one central task: hiring. “Instead of recruiters re-reading the same PDFs over and over,” he says, “candidate information flows into what we call ‘structured profiles’ that AI tools can actually read, make sense of, and use.”
The philosophy is straightforward: structured data beats a stack of documents every time. Storing information matters, but the companies pulling ahead aren’t just digitally warehousing the data important to their work. Smart companies are building intelligently structured systems that process whatever is stored, keep it current, and, in Goure’s words, “put it to work.” He explains: “Picsume dynamic profiles aren’t static files; by design, the data updates as candidates evolve, which means the AI is always reasoning from an accurate picture. That is what separates a file cabinet from a knowledge base.”
The Last Word
“The companies ‘winning’ with AI aren’t necessarily using better tools,” concludes Goure. “They’re feeding those tools better information. That’s the real competitive edge.”
