Should Brands Build a Knowledge Graph?
Or am I about to make the same mistake I made twenty years ago?
This week I found myself reading Bill Hunt’s excellent article on knowledge graphs and the future of AI discovery.
I’ve followed Bill’s work for years because he has a habit of spotting structural shifts before most of the industry. While many of us have spent our careers optimizing pages, Bill has consistently pushed the conversation toward the systems underneath them. His latest article is another example.
The central argument is compelling…
We’ve spent decades building websites for humans. Increasingly, our customers are interacting with AI systems that don’t browse websites the way people do. They need structured knowledge, not beautifully designed pages. Bill argues that organizations should stop thinking in terms of webpages and instead build a governed knowledge layer that can power websites, AI agents, search engines, and whatever comes next.
I think he’s right.
But as I was reading it, I realized I wasn’t asking the same question Bill was.
Bill was asking:
How should organizations build this capability?
The question running through my head was different.
Should my organization invest in building one at all?
That probably says more about my job than it does about Bill’s article.
I don’t spend my days designing knowledge graphs.
I spend them leading digital growth teams, building business cases, and helping executive leaders decide which capabilities are worth investing in.
Those are very different conversations.
I’ve been here before
About twenty years ago, I convinced an engineering team to invest in supporting RDFa.
At the time, Yahoo was promoting it heavily.
There were conferences.
Engineering events.
SearchMonkey beer mugs.
Big promises.
The thinking seemed obvious. If search engines could understand our content more deeply, they’d create richer search experiences. Richer experiences would improve click-through rates. More clicks would drive more traffic.
I believed it enough that I made it one of my OKRs. Engineering invested months implementing it. Then Yahoo quietly abandoned it. The capability we built never became strategically important. I’ve never forgotten that lesson.
So when someone tells me, “This is the future,” my first instinct isn’t excitement.
It’s due diligence.
The second conversation
That’s why I think Bill’s article raises another conversation we need to have.
Not...
Should organizations build knowledge graphs?
But...
How should executive teams decide whether they’re worth investing in?
Because projects like this rarely fail because the technology is difficult. They fail because organizations never become convinced the investment is worthwhile. Every successful transformation depends on change management. Every successful change management initiative begins with a compelling business case. And that business case almost never starts with technology. Over the years I’ve noticed something.
Executives rarely approve technology.
They approve capabilities.
No executive wakes up wanting to buy:
Schema
JSON-LD
RDF triples
Graph databases
Entity models
They invest in things like:
Better customer experiences
Faster decision-making
Lower operating costs
Reusable AI infrastructure
Competitive advantage
The technology is simply the mechanism.
The capability is what gets funded.
The Capability Lens
After enough hype cycles, I realized I needed a better way to evaluate emerging technologies. Not whether they sounded exciting. Whether they were worth asking an organization to invest in.
I’ve started thinking about that process as The Capability Lens.
Whenever I evaluate an emerging technology, I ask four questions.
What capability does this create?
Why does that capability matter?
What evidence supports it?
How would I explain the business case to leadership?
If I can’t answer those questions, I’m probably not ready to recommend the investment.
That’s where I’m still wrestling
I genuinely don’t know yet whether building an enterprise knowledge graph is the right investment for my organization. But over the past week I’ve started digging into the evidence. What I found actually increased my confidence. Just not for the reasons I expected. I think there are three different questions that often get blended together.
The first is:
Do leading AI companies believe knowledge graphs are valuable?
The answer appears to be yes.
OpenAI, Anthropic, Google, Microsoft, Neo4j, and others are all investing heavily in graph-based architectures to improve retrieval, reasoning, and long-context understanding.
The second question is:
Are modern LLM systems actually using knowledge graphs internally?
Again, the answer appears to be yes.
Architectures like GraphRAG, Anthropic’s work around Model Context Protocol, and Google’s GraphRAG infrastructure all point toward knowledge graphs becoming an important part of enterprise AI systems. But that’s not actually the question I’m trying to answer. The question I care about is different.
Will building my company’s external knowledge graph improve how LLMs understand and recommend my business?
That’s where the evidence becomes much thinner.
OpenAI using GraphRAG internally does not automatically prove ChatGPT is consuming my company’s public knowledge graph. Google’s investment in graph technology does not automatically mean Gemini will understand my business better if I build one.
Those are adjacent truths. Not the same claim.
As executives, we have to be careful not to confuse them.
So what evidence would convince me?
This is probably the question I’m spending the most time thinking about.
I’d love to see:
Independent case studies showing measurable improvements in AI recommendations.
Evidence that information unique to an enterprise knowledge graph is actually being surfaced in LLM responses.
Crawl behavior or documentation showing how these graphs are discovered and refreshed.
Examples where graph-based understanding produces meaningfully better recommendations than page-level content alone.
I’m not looking for certainty. I’m looking for enough evidence to responsibly recommend a significant engineering investment.
Those are very different standards.
Then I realized something else
The more I thought about Bill’s article, the more I wondered if AI discoverability is actually the wrong business case.
What if discoverability is simply the first dividend?
Even if external LLMs never consume my company’s graph directly, a governed knowledge layer could still become reusable infrastructure for internal copilots, customer support, APIs, search, personalization, analytics, and future AI initiatives.
Maybe the real capability isn’t AI discoverability at all.
Maybe it’s organizational memory.
Maybe it’s better reasoning.
Maybe it’s consistency.
Maybe it’s creating a single source of truth that every future AI initiative can build upon.
That’s a much more compelling investment thesis than “better SEO for AI.”
The question I’m taking into my next leadership meeting
I still haven’t decided whether my organization should build a knowledge graph. Bill’s article moved me closer. My RDFa experience reminds me to stay cautious. Somewhere between those two experiences is probably the right answer.
One of the reasons I created The Room is because I realized I couldn’t answer questions like this alone. The most valuable conversations I’ve had this year haven’t been about prompts or the latest AI model. They’ve been with executive leaders trying to answer the same question:
Is this a capability worth investing in today, or are we still too early?
Every month I get to compare notes with leaders across financial services, software, retail, and technology who are wrestling with these same decisions.
Some are investing aggressively.
Some are waiting.
Some are running pilots.
Some remain unconvinced.
I learn something from every one of those conversations. Because the job of an executive isn’t to predict the future. It’s to make the best possible decision before the future is obvious.
If you’re wrestling with questions like these, I’d love to continue the conversation with you inside The Room.