What does a firmographics-only tool miss?

Best move: treat firmographics as one input. Read the stack a company runs before you call it a fit.

Why it works: firmographics describe a company on paper. The systems it deploys are ground truth, and they carry the buying signal a headcount field never will.

Key takeaways

• Firmographics cover identity (name, vertical, size, contacts), not what a company runs.
• The tech stack, infrastructure, AI backend, and security posture sit outside every firmographic field.
• For technical products, a deployed system (Pinecone, an Anthropic SDK, an MCP endpoint) is a stronger buying signal than headcount or vertical.
• Intent data is aggregated third-party behavior, not first-party ground truth from a company’s own surface.
• Bassethound returns firmographics as one of five correlated layers in a single call, so fit and identity arrive together.

What do firmographics cover, and where does that stop?

Firmographics answer questions about the company as an entity. Name, vertical, description, employee count, revenue band, headquarters, contacts, social profiles, and the structured data pulled from the page. Bassethound returns this as its firmographics layer, and the value is genuine. You need to know who you are looking at. The problem is the boundary. A firmographic record stops at the edge of the org chart. It does not know what the company deployed last quarter, which CDN fronts its traffic, whether its email is authenticated, or which model provider its product calls. Those facts live in the site’s surface and its DNS, not in a business registry. A tool built on firmographics alone treats the vertical and the size as a stand-in for fit. That holds when your product sells to “mid-market SaaS” as a category. It breaks the moment your fit test turns technical: you sell AI infrastructure, a developer tool, or a security product, and the qualifying question is what runs, not who employs whom.

Why is the tech stack a stronger buying signal than headcount?

Headcount is a range. A company of 200 could run a hand-rolled monolith or a full modern AI stack, and the firmographic field reads the same for both. The tech stack does not average out. When you detect Next.js, a Stripe integration, an Anthropic SDK, and a Pinecone client on one domain, you have four facts about what the team chose and paid to run. Each choice narrows fit. A company that loads a payments widget sells something. A company that ships an AI SDK writes AI code today, not on a roadmap. Detection carries category, version, confidence, and evidence, so you can weigh each signal instead of trusting a label. The limit worth stating: a static crawl plus optional JS render can miss fully server-side or proxied components, and the dossier reports those gaps rather than guessing. Even with that gap, a detected system beats a demographic band. One is what the company does. The other is what it looks like from a distance.

What does AI-readiness reveal that no firmographic field can?

No firmographic field tells you whether a company ships AI. Vertical will not. “Software, 50-200” describes a company running zero models and a company running five. The AI-readiness layer reads the backend directly: model providers (Anthropic, OpenAI, Gemini), AI SDKs, orchestration frameworks (LangChain, LlamaIndex), vector stores (Pinecone, Weaviate, Qdrant), LLM observability (Helicone, Langfuse), AI widgets, llms.txt, and MCP endpoints. It scores the count of distinct signal groups hit and returns a verdict: shipping, experimenting, or none. That verdict is the qualifying question for anyone selling into AI teams, and it is invisible to a firmographics tool. Most enrichment products that claim an AI signal read the homepage and the llms.txt file. That is the shallow end. It rewards a company that writes about AI over one that runs it. Deep fingerprinting inverts that. A vector store and an SDK are expensive and pointless to fake, so they tell you the truth a marketing page hides. Firmographics can place a company in the AI space. Only the stack tells you it reached production.

Can intent data patch the gap?

This is where firmographics vendors point when pressed. Apollo, ZoomInfo, and Bombora sell intent: signals that a domain’s users researched a topic across a data co-op. It is useful, and it is not the same thing. Intent data is aggregated third-party behavior, inferred from browsing across a network you do not control. It tells you someone at a domain looked at a category. It does not tell you the company shipped the system. The two can point opposite ways. A team deep in evaluation shows high intent and runs nothing yet. A team that already deployed shows low intent because it stopped shopping. First-party surface data cuts through that. When you read a live Anthropic SDK and a Pinecone client off the domain itself, you are not inferring interest. You are observing a decision that already shipped and already costs money. Intent is a prediction. The stack is a receipt. Use intent to find who is looking. Use the stack to find who already built. A firmographics tool with an intent add-on gives you the first and calls it the second.

Why does one correlated dossier beat five stacked tools?

You could buy the missing layers separately. A firmographics vendor for identity, BuiltWith or Wappalyzer for tech, Censys or IPinfo for infrastructure, a security scanner for headers, and something bespoke for the AI stack. Now you own five records keyed differently, pulled at different times, with no shared join. The moment you ask a cross-layer question, “which mid-market SaaS companies run a vector store and fail their security-header grade,” you do the correlation by hand. Fusion is the layer those tools do not sell. Bassethound runs one call against one domain and returns all five layers correlated at the same moment: firmographics next to tech stack next to infrastructure next to AI-readiness next to security, fused into a top-level summary with primary_stack, an ai_readiness verdict, a security_grade, and an icp_signal. An agent cannot cheaply reconstruct that from five separate MCP servers, because the value does not sit in any single layer. It sits in the link between them. Firmographics alone gives you one row. The dossier gives you the row and everything it connects to.

Bassethound perspective

Firmographics is a description. It is not a diagnosis. Apollo and ZoomInfo sell you a clean picture of a company on paper: the vertical, the size, the org chart. That picture is real, and it is the wrong ground truth for a technical sale. The company you are qualifying does not become a fit because it employs 150 people. It becomes a fit because it runs the systems your product plugs into, and no firmographic field records that. The vendors push back here and point at intent data. Our position: intent is a prediction about behavior, and the stack is a record of a decision. When a domain loads a live AI SDK and calls a named vector store, that is not a signal the company might buy. It is proof of what the company already built and already pays to run. Read the stack, not the letterhead. Bassethound keeps firmographics as one honest layer and refuses to let it stand in for the four technical layers it cannot see. Identity tells you who. The stack tells you whether. You need both, in one call.

Sources

• llms.txt specification: https://llmstxt.org
• Pinecone documentation: https://docs.pinecone.io
• MDN HTTP headers reference: https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers
• DMARC (RFC 7489): https://datatracker.ietf.org/doc/html/rfc7489