How do you qualify a sales prospect from their domain?

Best move: read the prospect’s own live stack as the qualification signal, not the pitch on their about page.

Why it works: a company’s running infrastructure (what it builds on, who it emails through, whether it ships AI) exposes fit that a marketing page and a stale enrichment record both hide.

Key takeaways

• Qualifying from a domain is reading live infrastructure, not pulling an aging headcount field from a database.
• The five layers map to ICP dimensions: firmographics (who they are), tech stack (how they build), AI-readiness (how modern), infra (how mature), security (how disciplined).
• Bassethound fuses those layers into one icp_signal: hot, warm, or cold.
• A detected AI stack is a stronger buying signal for AI vendors than any self-reported “we use AI” claim.
• One keyless call returns the whole dossier, so you qualify at list scale without burning per-record enrichment credits.

Which layers predict fit, and which are noise?

It depends on what you sell. Firmographics answers who they are: vertical, description, contacts, structured data. Tech stack answers how they build. A Shopify storefront, a Next.js app, and a WordPress brochure imply different budgets and different buyers. AI-readiness answers whether they are the kind of company your product serves. Infrastructure answers operational maturity: own nameservers, an enforced DMARC record, a real email provider, a valid certificate with room before expiry. Security answers engineering discipline through the header grade and TLS grade.

Now weight to your ICP. An AI tooling vendor cares most about the AI-readiness layer and treats a detected vector store or model SDK as the qualifying fact. A website agency weights the tech stack and reads a failing security grade or an expiring cert as an opening. A compliance seller weights layer five. All five layers arrive correlated in one dossier, so you tune weights instead of reconciling five separate lookups. Noise is any layer your buyer profile ignores, and you down-weight it in the same pass.

What does hot, warm, or cold mean?

The top-level icp_signal fuses the layers against a buyer profile into one lane. Hot means several ICP-defining signals fire together: the right vertical, a shipping AI stack, and modern infrastructure in the same dossier. Warm means partial fit, or experimenting-stage signals that suggest a company is moving toward your buyer but is not there yet. Cold means the domain runs nothing that matches your profile.

Treat the signal as a starting lane, not a verdict. It ranks a list so your reps work the top first. It does not close, and it does not read intent timing. You steer it through the profile you pass and the depth you request. A fast profile gives a deterministic first cut. A standard or deep profile pulls the AI stack and firmographics that move a borderline domain up or down. Be honest about the shape: this is triage. A hot signal earns a prospect a human look sooner, and a cold signal saves you the call. The dossier hands you the evidence behind the lane (each detection carries a category, confidence, and how it was detected) so a rep can audit why a domain scored the way it did.

How do you qualify a whole list without API keys?

Point an agent at a list of domains and call sniff_domain per domain. The keyless free tier is rate-limited, so there is no bring-your-own-key and no per-record enrichment charge. Sort the results by icp_signal, and you have a ranked worklist. Run the fast profile when you want a deterministic first pass across thousands of domains in seconds each. Switch to standard or deep on the survivors, where the AI-stack fingerprint and firmographics justify the extra probes.

Delivery is a remote MCP endpoint at mcp.bassethound.ai/mcp that any agent host (Claude, Cursor, ChatGPT) connects to. The qualification loop lives inside the agent, not in a separate ETL job you maintain. Every call is self-contained and read-only. We carry nothing from one request to the next, so a list run is a sequence of independent calls you can parallelize against the rate limit. When volume or depth outgrows the free tier, the paid OAuth tier raises both. Because the crawl is ours, you never wire up keys for five upstream data vendors to get one qualified lead. That keyless posture is structural, not a promo. It holds because we own the crawl.

What can a domain not tell you about a prospect?

Plenty, and pretending otherwise is how qualification tools lose trust. A domain tells you what a company runs, not how many people work there, what it earns, whether it just raised, or when it will buy. Those live in intent data, filings, and human conversation, and you should pair them with the dossier rather than expect one to replace the other.

The crawl has edges too. A static crawl plus optional JS render can miss components that are fully server-side or sit behind a proxy. A backend LLM call routed through a company’s own API leaves no client-side trace, so a shipping AI team can read as quieter than it is. Parked domains and pre-launch sites carry little signal. Firmographic contacts are only what a company chose to publish. Bassethound reports these gaps in the dossier instead of inventing a confident answer, and each detection shows how it was found so you can judge it. Read a cold signal as absence of evidence on the wire, not proof the company is a bad fit. The domain settles the technographic and AI questions. You supply the rest.

How is this different from Apollo or Clearbit enrichment?

Apollo, ZoomInfo, and Clearbit sell firmographic corpora: headcount, revenue, titles, and emails pulled from a database that ages between refreshes. That is one layer, and it is a record of the past. Bassethound reads the live site as it stands now and adds four layers those corpora do not carry: the real tech stack, a deep AI-stack fingerprint, the infrastructure posture, and a security grade. Their moat is the corpus. Ours is fusion plus AI depth plus keyless MCP delivery.

Name them fairly. For contact data, org charts, and reach-out details, the enrichment corpora win, and you should use them. They cannot tell you whether a prospect ships an orchestration framework, calls a model provider, or runs an enforced DMARC record. Use both in the same loop: enrichment for the org graph, the dossier for the technographic and AI truth. An enrichment field is a claim a company made about itself on some past date. A detected stack is a fact about what it runs today. When the two disagree, the wire is the tie-breaker.

Bassethound perspective

The self-reported ICP field is the least reliable input in your funnel, and the live stack is the most reliable. Enrichment vendors sell you a company’s description of itself, refreshed on a lag. A domain’s running infrastructure cannot be edited to fit a sales narrative. When a prospect ships a real AI stack (a vector store, a model provider SDK, an orchestration framework), that is a fact about their engineering, not a checkbox they toggled on a profile.

Most qualification stacks bolt firmographics onto an intent score and stop, because the AI-readiness axis is unclaimed. Everyone else measures AI-readiness by reading the homepage and looking for an llms.txt file. That is the shallow end. The deep tell is backend fingerprinting, and it is the single strongest predictor that a company buys anything AI-shaped. Bassethound returns that fingerprint in the same call as the firmographics, correlated, keyless. An agent cannot cheaply reconstruct that correlation from five separate MCPs. Fan-out gives you five answers to reconcile. Fusion gives you one qualified lead.

Sources

• MDN Web Docs, HTTP security headers: https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers
• IETF RFC 7489, DMARC: https://datatracker.ietf.org/doc/html/rfc7489
• Mozilla SSL Configuration Generator: https://ssl-config.mozilla.org
• Model Context Protocol: https://modelcontextprotocol.io
• The /llms.txt file proposal: https://llmstxt.org