Directories list the suppliers who paid for a listing. We screen the entire classified web — every shop, every certification page, every capability list — against your actual requirement.
Every supplier directory, marketplace, and sourcing platform shares one structural flaw: membership is voluntary. The companies you find there chose to be found — they paid for the listing, completed the profile, or courted the aggregator.
The specialist shop your requirement actually needs often did none of that. It runs a plain website, wins work by reputation, and has never needed marketing beyond its own customer base.
A directory is a sample biased toward suppliers who market themselves. For commodity parts that may not matter. For a critical-path component, sourcing from a self-selected subset is unmanaged risk.
RFI rounds spend weeks collecting facts most suppliers already publish. A certifications page stating AS9100D and ITAR registration answers the first screening question before anyone drafts an email.
Second-source strategies fail at the discovery stage. If you cannot enumerate who else can make the part — by process, material, and approval — you cannot price your dependence on the incumbent.
The uncomfortable arithmetic: in our industrial census work, a single equipment category held 367,478 domains worldwide. No directory carries a fraction of that — and the fraction it does carry is not a random sample.
The bias compounds quietly. Suppliers that market aggressively appear in every search; suppliers that run at capacity on repeat business appear in none. Your sourcing funnel systematically over-samples the shops that need work.
A census inverts that. When the pool is every operating company in the category, marketing budget stops deciding who you evaluate — stated capability does.
Prequalification evidence is not hidden. It sits on capability pages, quality pages, and equipment lists — published by the supplier, waiting to be read at scale.
"The subset of supplier-published claims — certifications, process lists, materials, approvals, facility data — that a sourcing engineer would otherwise collect through an RFI, captured verbatim with the page it came from."
None of this replaces the supplier's answers — it changes which suppliers you bother asking. An RFI sent to twelve evidenced candidates outperforms one blasted at two hundred directory rows.
And because every extracted field carries its source page, a skeptical quality engineer can verify any claim in one click before the first conversation happens.
A database field that says "certified: yes" is an interpretation. A quoted sentence — "certified to ISO/IEC 17025 by A2LA, scope: dimensional calibration" — is evidence a quality engineer can act on, and challenge.
The extraction is deliberately conservative. When a site claims a certification, we capture the exact claim text and source URL; when the language is ambiguous — "compliant with", "working toward" — the ambiguity is preserved, not resolved in the supplier's favor.
The same discipline runs through our acquisition screening. It is described in full on our method page, and the extraction machinery is identical whether the question is "can they make the part" or "do they fit the thesis".
We operate several AI platforms and provide large-scale specialized datasets — 300+ enterprise organisations run on our data, including one of Europe's largest telecom operators, adtech and cybersecurity platforms, and a leading airline metasearch.
The engagement runs the same two-pass architecture as our acquisition work: a cheap, broad triage pass over everything, then expensive deep extraction reserved for candidates that survive it.
Your sourcing spec becomes the screening template: processes, materials, tolerances, mandatory approvals, geography, and capacity markers. We stress-test it against real supplier sites before the run, so edge cases surface early.
Every domain in the matching categories is triaged — tens or hundreds of thousands of candidates reduced to live, operating suppliers. In one US-focused sample, 25,000 domains yielded roughly 17,300 operating companies.
Surviving candidates get a full read: certifications quoted, machine lists and process envelopes captured, facility data recorded, customer industries evidenced from case studies rather than guessed from keywords.
Suppliers ranked by evidenced match to your spec, with every inclusion carrying its quotes and every notable exclusion carrying its reason. The file drops into your SQE workflow as-is.
The full framework extracts fifteen signals per company; these five carry most of the weight in supplier discovery.
The prequalification layer, pre-collected. Explicit certifications only — AS9100, NADCAP, ISO/IEC 17025, ASME stamps — captured as exact claim text with source URLs.
Scope language matters: an accreditation limited to one lab or one process is recorded that way, because your auditors will read it that way.
The principal score, pointed at your spec instead of an acquisition thesis. Stated processes, materials, and tolerances are matched against what you actually need — so a shop that lists Swiss turning but not your alloy family ranks below one that evidences both.
Logistics and regionalization live here. Stated HQ, branch locations, and service areas are extracted with owned facilities distinguished from partner mentions — the difference between a supplier with a plant in your region and one with a sales rep there.
A careers page is a capacity disclosure. A shop hiring machinists for a second shift has load coming and room to take yours; a site whose last posting is years old tells a quieter story. We record what is posted, when, and for which functions.
Dormancy is a supplier-risk signal you can read from public evidence: the most recent dated content, the freshness of equipment lists, whether the news page stopped updating years ago.
We flag visibly inactive suppliers before they cost you an RFQ cycle — from published material only, never from speculation about a company's finances.
Named OEM authorizations, distributorships, and association memberships place a supplier in its ecosystem. An authorized service center for your installed equipment brand is a different candidate than an independent generalist — sometimes better, sometimes disqualifying, always worth knowing before the RFQ.
Our published precision-machining specimen — the same format a sourcing engagement receives — reads quality approvals directly off supplier sites and pairs them with capability and capacity evidence.
The keyword-missed fits deserve the emphasis. A fifth or more of confirmed fits in our industrial runs never use the category's obvious homepage vocabulary — they describe outcomes, not processes, and only a full read finds them.
The full report is on the specimen page; the vertical itself is profiled under precision machining.
Every field traces to a sentence. Every sentence traces to a page.
Each method answers a different question. The table shows which — and why census questions need census tools.
| Question | Paid directory | Marketplace RFQ | Full-web screen |
|---|---|---|---|
| Who is in the pool? | Suppliers who paid | Suppliers who bid | Every operating site in category |
| Certification data | Self-declared checkboxes | Claimed at quote time | Quoted from the source, with URL |
| Capacity signals | Not covered | Inferred from response speed | Hiring, facilities, activity trajectory |
| Second-source census | Partial by construction | Only respondents visible | Complete by construction |
| Cost structure | Annual subscription | Per-transaction fees | Project pricing, from €4,900 |
The part is critical-path, the approval stack is non-negotiable, or the strategic question is "who else can actually make this" — census questions that sampling methods cannot answer.
You need a commodity part fast and any of a hundred qualified shops will do. Directories are genuinely good at that job; we are not trying to replace it.
A discovery screen that oversells itself poisons the qualified list it delivers. Ours has four stated limits.
We replace the discovery and paper-prequalification stage, not the audit. What changes is that your auditors travel to genuinely qualified candidates instead of burning visits on mismatches.
Websites do not disclose balance sheets, and we do not estimate them. Credit checks and financial assurance stay with your existing process — our standards on this are published.
Evidence that a shop states a capability is not evidence of your part in tolerance. The screen shortens the path to a first article; it does not stand in for one.
The census covers the web-visible market. In industrial B2B that is nearly everyone, but a supplier with no website is invisible to us — we state that boundary rather than extrapolate across it.
The output is a working file, not a report to admire. Teams typically run it through four stations.
The ranked CSV loads into your SRM, PLM, or a plain spreadsheet — structured fields, one row per supplier, evidence columns included.
Match against your approved-vendor list. The interesting rows are the qualified suppliers your AVL has never heard of.
Filter on the quoted approval stack and capacity markers; your engineers review evidence snippets instead of browsing forty websites.
RFQs go to a defensible shortlist. When a stakeholder asks why a supplier made the cut, the answer is a quote, not a hunch.
The classic trigger. A single-source dependency gets priced only when the alternatives are enumerated — by process, material, and approval, not by directory membership.
A new part family needs suppliers the current AVL was never built for. The census finds the specialist base before the launch schedule starts compressing.
Nearshoring a category means rebuilding supplier knowledge in a geography where your team has no history. Evidence-first discovery substitutes for a decade of local relationships.
Two adjacent engagements extend the same machinery: channel & partner mapping when the question is distribution rather than supply, and custom taxonomy classification when your commodity codes deserve better than standard industry categories. For market-level counts, the census version is TAM mapping.
Your suite searches a licensed database — a subset of the market that opted in. We screen the classified web itself: every operating company in the matching categories, whether or not it ever registered anywhere.
The two are complementary. The gap between them — qualified suppliers your suite has never heard of — is usually the most interesting part of the deliverable, and it is measurable on your own requirement.
Narrow is where the method earns its price. A requirement like "hard-turning of case-hardened steels, with ISO/IEC 17025-accredited inspection in-house, within 500 miles" is exactly what LLM screening against stated capabilities handles and keyword search does not.
The spec is written with your engineers before the run, stress-tested against real sites, and re-runs on a refined spec are included in every engagement.
Each engagement re-reads supplier sites at run time — the extraction is fresh, not warehoused from an old crawl.
For standing programs, monitoring re-screens the qualified universe on schedule and delivers deltas: new certifications, lapsed language, ownership changes, hiring shifts. See universe monitoring for the mechanics.
No. The screen reads public websites only; no supplier learns it was evaluated, and no signal of sourcing intent exists until you create one.
That discretion matters when the requirement itself is competitively sensitive — a second-source program you would rather not announce, or a component that telegraphs a product launch.
A Proof project — one requirement, full universe — starts at €4,900. A full universe with deep shortlist starts at €9,900, and annual monitoring of a qualified supplier base from €18,000 per requirement family.
Custom re-runs with a revised spec are included, and the full schedule is public on the pricing page — no call required to learn a number.
The underlying classification is global — the 367,478-domain category example above is a worldwide count. Screens are usually scoped to the geography your logistics can serve, and regional scoping is part of the requirement spec rather than a technical constraint.
Send us one hard requirement. We will show you a specimen of what the full-web screen returns — approvals quoted, capacity evidenced, gaps in your current directory made visible.