The 30-hour problem of manual supplier sourcing
Your boss asks for "5-10 options by Friday" for a new component category. You open Google. Seven hours later, you have a browser with 34 tabs, a messy spreadsheet with six suppliers, and four inboxes to check for responses.
That is not bad luck. That is the baseline. Industry benchmarks put manual sourcing for a single category at 25-35 hours, split across research, contact discovery, outreach, follow-up, and comparison. A typical Google search for "plastic injection manufacturers" returns roughly 12 million results. A trained purchasing manager realistically clicks through about 50 of them before giving up and going with the first three that replied.
The math is not kind. Thirty hours times a loaded €50-60/hour rate means each sourcing campaign quietly costs your company €1,500-€1,800 in internal time, before you have received a single quote. Multiply that by 15-25 campaigns per year per buyer and you are looking at €30k-€45k per head spent on research that produces a 5-supplier shortlist.
This post walks through the 4-step funnel that gets you from "one search query" to a scored shortlist with email addresses on it, in under an hour rather than under a month. How long that shortlist runs depends on how many suppliers the category holds: the eight published runs returned between 7 and 170 companies, median 39.
The hidden tasks nobody budgets for
When procurement teams estimate sourcing time, they usually count "research." They forget four tasks that eat more hours than research itself: chasing non-responsive suppliers, translating foreign-language websites, re-keying data from PDFs into a spreadsheet, and formatting the comparison report for the person who will actually make the decision.
Research is maybe 30% of a sourcing campaign. The other 70% is coordination work. That is why software that "helps you search Google faster" barely moves the needle, the search was never the bottleneck.
Why Excel breaks at 20 suppliers
Every spreadsheet-based sourcing project looks fine until supplier 18, when someone replies with a PDF in Italian with four price tiers, a different MOQ per tier, and payment terms expressed as "30% pre / 70% BL." You copy it into the spreadsheet, it breaks three formulas, and you spend 40 minutes normalizing the row. At supplier 25, you stop comparing and just pick someone.
That is how 70% of mid-market sourcing projects end, not with a decision, but with fatigue.
What counts as a "verified" supplier, and what does not
Directory listings are not verification. A profile on Europages, Kompass, or ThomasNet confirms one thing: someone filled in a form and probably paid for the listing. It does not confirm the company still trades, its certifications are current, or the contact person is reachable.
Real verification in 2026 means four things at minimum:
- Trade registry presence, the company exists in an official registry (KRS in Poland, Handelsregister in Germany, Companies House in the UK, Registre du Commerce in France). No registry, no supplier.
- Valid VAT number, cross-checked in VIES for EU suppliers. "VAT pending" means you do not send them a deposit.
- Working website and working email, the website loads in the last 30 days, and at least one email domain MX record resolves. Dead websites with a contact form are red flags for dormant companies.
- Certification match, if your brief says "ISO 9001 required," the certificate number must be valid and listed on the certifying body's public registry (IAF CertSearch covers most accredited bodies).
Anything weaker than this and you are just sending RFQs into the void.
A missing certificate never drops a good maker. It lowers a score, and the score is a number you can argue with.
Where to find suppliers online, and why directories alone are not enough
Most procurement teams default to one of three tools: Alibaba for Asia, Europages for Europe, ThomasNet for North America. Each is useful. None is sufficient.
General directories (Alibaba, Europages, Thomasnet, Kompass) give you breadth but poor signal. The suppliers who pay for premium listings are not always the best, they are just the ones with marketing budget. Directory data also ages badly: a 2019 listing for a plastics factory tells you nothing about whether that factory still exists in 2026.
Regional specialty directories (wlw.de for German industry, panorama-firm.pl for Poland, infobel.com for France) are much better for depth in one country. But you need one per country, each in a different language, each with a different search UX. Running a five-country search manually means learning five interfaces.
Trade shows (Hannover Messe, Fespa, Anuga) still matter for strategic relationships but are a terrible research tool for fast sourcing. The cycle, plan, attend, follow up, is six months at best.
AI-assisted search reads all of the above (plus raw web content and press releases) and does the language translation in the background. You type one brief in English; the system searches in German, Polish, Turkish, Portuguese, and Italian simultaneously. Coverage multiplies, language barrier disappears.
The key point is not "replace directories." It is "stop relying on any single source." Serious sourcing in 2026 combines at least three of the above with automated cross-referencing.
The 4-step AI-assisted sourcing funnel
The workflow below is what Procurea runs when you click "Start campaign." It is not novel, procurement consultants have taught versions of this for years. What is new is compressing all four steps into a single automated pass.
Step 1. Discover. Convert your brief (category, geography, certifications, volume) into 25-40 targeted search queries per country, in the local language. Run them in parallel against the open web, which is where the regional directories and the company pages both sit. Output: a queue of candidate URLs, of which the published runs went on to read 91 to 1,205 company pages.
Step 2. Screen. Scrape the candidate pages. Apply a relevance filter: is this actually a manufacturer (not a reseller, not a dormant page, not a news article)? Does the company serve your volume range? Does it claim the certifications you need? Output: the candidates that survive screening, each rejection kept next to the reason it was dropped.
Step 3. Enrich. For each surviving candidate, find the decision-maker email (procurement contact if listed, otherwise plant manager or sales director), the VAT number where the company publishes one, and a short capability summary. Output: 7 to 170 enriched supplier records in the published runs, median 39, depending on how many suppliers the category holds.
Step 4. Score. Re-read each supplier against its own pages and score the claims. Where an EU supplier publishes a VAT number it goes to VIES, the one independent source in the chain. Certificates are recorded as stated, with the page they came from, so an ISO or IATF claim is checkable in IAF CertSearch by you rather than asserted by us. Output: a scored shortlist that says where every field came from.
Done manually, this is four to six weeks. Done in Procurea, the eight published runs took 4 to 16 minutes of compute from click to finished list, with the buyer doing about 90 seconds of input at the start.
Discover
25-40 targeted search queries per country, in the local language, run in parallel against the open web. Output: the 91 to 1,205 company pages each published run went on to read.
Screen
Scrape + relevance filter. Is this a real manufacturer, right size, right certifications? Output: the candidates that survive screening, each rejection kept next to the reason it was dropped.
Enrich
Find the decision-maker email, the VAT number where one is published, and a capability summary for each survivor. Output: 7 to 170 enriched records in the published runs, median 39.
Score
Re-read each supplier against its own pages and score the claims. VAT numbers go to VIES; certificates are recorded as stated, with the page they came from. Output: a scored shortlist that says where every field came from.
The funnel, summed across eight published runs
Step 1, Define a brief AI can actually use
"Find me cheap plastic suppliers" is not a brief. A usable brief has five fields: (a) product category with HS code if available, (b) target countries or regions, (c) MOQ and annual volume, (d) mandatory certifications, (e) deal-breakers (e.g. "no suppliers smaller than 20 employees"). Five fields, three minutes. This is the only step that requires you.
Step 2, Multilingual search across 5-10 countries
"Plastic injection molder" is the English query. The German version is "Kunststoff-Spritzguss Hersteller." Polish is "producent wtryskarek tworzyw." Italian is "stampaggio a iniezione plastica." Each returns different top-ranked sites. AI sourcing runs all of them in parallel, a typical campaign hits 25-40 localized queries per country, 5-10 countries, 200-400 searches total in under two minutes.
Step 3, AI screening on relevance, size, and certification
For each candidate URL, the system scrapes the homepage and key pages (products, certifications, about), extracts structured data, and scores against your brief. A plastics reseller with no production gets screened out. So does a manufacturer that only does automotive Tier 1 when you asked for consumer goods. The output is filtered, not just re-ranked, a real "no" goes in the discard pile.
Step 4, Enrichment and scoring in one pass
For each surviving candidate, enrichment finds the right contact, validates the VAT (VIES), and records certification claims with the page they were stated on, so you can check them in IAF CertSearch yourself. Suppliers with invalid data get flagged, not silently dropped, you see why each failed so you can override if needed. Final output in the published runs: 7 to 170 scored suppliers, median 39, most of them with a contact address.
How to verify a supplier in under 5 minutes
Even with automation, some shortlisted suppliers will need a fast human sanity check, usually the top 5-10 you are about to send an RFQ to. Use this three-minute checklist:
- Open the trade registry entry. Confirm the legal name matches. Confirm paid-up capital matches a company large enough to serve your volume (red flag: €10k capital claiming to manufacture for Fortune 500). Check if the managing director has been in role less than 12 months, rapid turnover is a signal.
- Run VIES on the VAT number. Takes 30 seconds. Valid = the company is tax-registered for cross-border trade. Invalid = ask why before you quote.
- Open their website. Is the copyright year current? Is the news or blog updated within the last 12 months? A website that was last touched in 2021 is a dormant company 80% of the time.
- Reverse-image search one product photo. If their "factory" photo also appears on four competitor websites, you are dealing with a trader, not a manufacturer.
- Check certifications on the issuer's registry. IAF CertSearch for ISO/IATF, notified-body websites for CE. An expired ISO 9001 is worse than no claim, it means the supplier let their quality system lapse.
Three minutes per supplier, times five top candidates, is 15 minutes of manual verification. This checklist is the verification step, and it is yours: step 4 narrows the field and tells you where each claim came from, it does not confirm the claim with the body that issued it.
Common mistakes when sourcing manufacturers
Five mistakes eat 80% of the bad outcomes in supplier sourcing:
Relying only on paid directory listings. The suppliers who pay for premium placement are self-selected for marketing spend, not operational quality. Healthy manufacturers often do not maintain directory profiles at all.
Ignoring language barriers. Sending an English-language RFQ to a Turkish or Polish manufacturer cuts response rate by about 60%. You are not reaching decision-makers, you are reaching whoever in the office can handle English. That person is rarely empowered to quote.
Over-weighting price on the first pass. A 15% lower quote from an unverified supplier routinely turns into a 40% overrun by the time you factor rework, expediting, and legal. Price compare after you have verified capability.
Skipping the sample order. A €500-€2000 sample run pays for itself the first time it catches a quality-system gap before a €50k bulk order.
Picking "the one that replied fast." Fast response often correlates with desperate sales, which often correlates with capacity issues. The shop that takes four days to respond and asks three sharp questions is usually the better partner.
None of this is new to experienced buyers. All of it still happens to experienced buyers who are rushed and under-resourced. Widening the funnel in step 1, going from 20 candidates to 200, is the cheapest way to buy yourself room to avoid these mistakes.
Worked example: one published run, 1,205 URLs to 85 suppliers
Brief: medical disposables, FDA registered manufacturers, single use, sterile. Worldwide, China excluded. Every number below is read out of the campaign record, because this run is published in full. The previous version of this section described a cosmetics campaign that never ran.
Discovery pass. The engine put the addressable market at about 1,458 companies and worked through 1,205 URLs.
Screening pass. 245 candidates were dropped as too weak a match for the brief, three were skipped as sample pages, and one turned out to be a government procurement portal. Each rejection keeps its reason, which is the part you can argue with.
Enrichment pass. 85 suppliers survived: 58 primary, 27 reserve. 28 of them came with a contact address and 28 with a certificate stated on the page. The gap between "found" and "reachable" is the honest part of this business and no amount of engine tuning closes it.
Scoring pass. Every kept supplier is re-read against its own pages and scored: median 85 out of 100, best 95. Where an EU supplier publishes a VAT number it goes to VIES, which is the one independent source in the chain. Certificates are recorded as the supplier states them, with the page they came from, so you can check IAF CertSearch yourself.
Result. 85 suppliers in 11.6 minutes of compute. The first landed 88 seconds in and the tenth at 118 seconds, so the list is readable long before the run ends. What happens next is yours: the run ends in a spreadsheet, and sending the enquiries is a separate job we do not do for you today.
472 suppliers across 8 briefs, 1053 rejections still on the record with the reason each one was given.