AI Sourcing, 7 to 170 suppliers in 4 to 16 minutes
Describe what you need in plain language. The pipeline plans a search strategy across 26 languages, reads company pages the way a buyer would, scores every supplier against your brief, and records the contact details and certificates a supplier publishes itself. Eight published runs returned between 7 and 170 suppliers in 4 to 16 minutes, median 39: how many you get depends on how many suppliers the category holds at all.
Manual sourcing is broken: and it is costing your team weeks on every project.
A typical procurement team spends 30+ hours on a single category search: Googling in one or two languages, scraping lists from association websites, cross-checking half a dozen directories, pasting results into a sprawling Excel that immediately starts going stale. The real vendors you need: niche, regional, certified: hide behind local-language sites, on page 7 of Google, or inside specialized directories you have never heard of. You cannot read Turkish, Italian or Polish supplier catalogs at scale, so you default to the same 20 names everyone in your industry already contacts.
Legacy tools do not help. Traditional supplier databases were built a decade ago: they charge $40k+ per seat, cover maybe one region well, and their data ages out in months. B2B marketplaces show only vendors who pay to list. Your own ERP supplier master only reflects who you have already worked with. None of these find the new, better, cheaper vendor who just entered the market last quarter.
Meanwhile your CFO is asking why supply costs are up 8%, your category managers are drowning in RFQ admin, and every time the board pushes a nearshoring initiative you restart the research from scratch. Sourcing should not be the bottleneck of modern procurement: but until now, there was no tool built for how AI actually works in 2026.
Step by step.
Describe what you need: in one sentence
Type a prompt the way you would brief a colleague: "Injection-molding vendors in the EU for automotive interior parts, IATF 16949 required, 50k-500k units/year, lead time under 8 weeks." No boolean operators, no tag pickers, no search syntax. The prompt can be in any of 26 supported languages: the pipeline understands Polish, German, Turkish, Italian, French, Spanish, English and more natively.
Powered by Gemini 2.0 Flash: the same model behind the AI interpretation of your brief is used to plan search queries across every supported market.
The Strategy agent plans a multi-market search
Before a single query is sent, the Strategy agent decomposes your brief into sub-categories, selects the 5–15 most relevant countries, and generates 20–40 targeted search queries per country: each written in the local language with local taxonomy. It picks the right directories (Wer liefert was, Kompass, GoldenLine, Europages, national chambers of commerce) and the right long-tail web queries, calibrated to your certification and capacity constraints.
Strategy output includes a query budget plan: you see exactly how many searches will run and can adjust scope before the campaign starts.
The Scanning agent crawls the web in parallel
Up to 20 worker threads run simultaneously, combining Google search (via Serper.dev), direct crawls of supplier directories, national company registries, and trade association listings. Rate limiting, retry logic and a disk-backed cache keep cost under $2.50 per campaign while pulling from sources legacy databases never touch. Every URL found is scraped and fed to the next stage: no human paging through results.
90-day cache TTL means a follow-up campaign in the same category costs roughly half as much and runs in minutes.
The Analysis agent scores every supplier for fit
Each candidate is evaluated against your brief by an LLM reviewer: does the website evidence the capability you asked for? Are the certifications you need mentioned or implied? Is the company size in range? Are there trust signals: clients listed, case studies, ISO marks: or red flags (empty about page, dormant social, broken contact forms)? Suppliers are scored 0–100 and the obvious misfits are dropped before enrichment to keep the shortlist tight.
Scoring rubric is consistent across campaigns: so vendors from your last 5 searches are directly comparable.
The Enrichment agent fills in the missing data
For every qualified supplier, the enrichment pass pulls the company website and the contact details it can find on that site (email, phone, LinkedIn), plus the certificates the supplier publishes about itself. What lands in your dashboard is a shortlist you export to a spreadsheet and take wherever you work: there is no connector and no API, so the file is the handover.
An Auditor pass runs last to flag inconsistencies: e.g. a capability claim the LLM could not verify on the site, or missing contact information.
Key capabilities.
- 26-language native research: Polish, German, French, Italian, Spanish, Turkish, Dutch, Czech, and 18 more
- Multi-source coverage: national registries, trade directories, association listings, long-tail web
- Regional search strategies calibrated per category (EU-focused, global, near-shoring, reshoring)
- Centralized Supplier Database: AI match scores, contact details, certifications, and campaign history per vendor
- Decision-maker contacts: purchasing lead, procurement manager, CEO: with deliverability signals
- Campaign history per supplier, so a list you ran last quarter is still there next quarter
- AI capability scoring: suppliers ranked 0–100 on how well they match your brief, with visible reasoning
- Certificates recorded where the supplier publishes them: ISO 9001, IATF 16949, FSC, FDA, CE, OEKO-TEX and the rest. Recorded, not confirmed with the issuer
- One-click Excel: full supplier list with all enriched fields, audit-ready format
- The same ten columns every run, so a file always lands the same way in whatever you load it into
What it looks like in your industry.
Finding 8 alternative injection-molding vendors across the EU after a Tier-1 supply chain disruption.
6 weeks of desk research, 200+ cold emails, 12 qualified vendors after one procurement manager works the category full time.
The published bearings run: 39 suppliers in 12.3 minutes, each one scored against the brief, with the rejections and their reasons in the same file. Certificates appear where a supplier publishes them, and we do not confirm them with the issuer.
Sourcing catering, AV and scenography vendors in Berlin for a product launch two weeks away.
4 vendors per category found via English-only Google, no local-language coverage, premium agency pricing, nothing booked 10 days out.
The published catering run, a 400 guest launch in Chicago: 77 caterers in 11.7 minutes, read in the local language rather than through English keywords. Collecting the quotes is still your job, by email.
Subcontractor sourcing for HVAC on a 200-unit residential development tender in Mazowieckie.
Phone calls to the industry rolodex, 8–12 qualified bidders after two weeks: mostly the same names everyone invites.
The published scaffolding run: 67 contractors in 10.2 minutes, each classified as a maker or a reseller from what its own site says. References are something you check on the call.
Private-label cosmetics producer, migrating volume from China to EU/Turkey near-shore manufacturing.
Alibaba searches, three months to shortlist, MOQs too high for a private-label brand scaling 0 → 1.
The published folding carton run: 170 converters in 6.2 minutes across 19 countries, the largest list of the eight. Minimum order quantities are not something the engine can see, so they stay a question for the first email.
More than a search: a data layer for the rest of your procurement stack
AI Sourcing is not a standalone tool. Every campaign feeds your Supplier Database: AI-scored vendor profiles, multi-language supplier records, campaign history per supplier, and a spreadsheet export with a fixed column layout. You get supplier research that compounds across projects instead of restarting from zero every quarter.
- Supplier Database included on every campaign: AI scores, contacts, certifications, campaign history: no separate data management needed
- Native 26-language search keeps regional vendors in the shortlist instead of filtered out by your English keywords
- Campaign history per supplier, so you can see every campaign a name has appeared in
- Excel export on every plan, with the columns fixed so a run always lands the same way in whatever you load it into. There is no API and no connector: the file is the interface
Questions buyers ask about this.
- How accurate is the AI capability scoring?
- We have not run a precision study, so any percentage here would be invented. What we can show is eight published runs with every score and every rejection in the file, so you can judge the matching on real output rather than on our number. Top candidates should still be verified manually: AI Sourcing gets you to a shortlist of 20–30 real contenders in minutes; the final decision stays with your team.
- Which languages and regions does it cover?
- Full support for 26 languages covering the EU plus major global markets. Highest precision today: Polish, German, French, Italian, Spanish, English: thanks to very strong directory and registry coverage. Turkish, Czech, Dutch, Portuguese and Nordic languages are fully supported. Non-Latin scripts (Arabic, Chinese, Japanese) are on the roadmap.
- What does one sourcing campaign actually get me?
- A scored list you export to a spreadsheet, with the rejections and their reasons alongside it. The eight published runs returned between 7 and 170 suppliers, median 39, in 4 to 16 minutes. The number depends on how many suppliers the category holds: the folding carton run found 170, the gastro dealer run found 7 and is published as the failure it was.
- Can I ask for a certification in the brief?
- You can name one, and the Screener weighs what the supplier's own pages say about it, which is also what the score explains. What we do not do is confirm a certificate with the body that issued it, so treat a certificate on a row as a claim to check rather than a verification. Certificates you hold yourself are a different thing: those you upload, and the system watches the expiry date.
- What do the plans differ on?
- Volume, and nothing else. Starter is $69 a month for 10 campaigns, Growth $159 for 50, Scale $299 for 150, and every plan runs the same pipeline with the same export. There is no Enterprise tier, no API and no SSO: above 150 campaigns a month the answer is a conversation, not a fourth column on the pricing page.
- Can I see the output before I buy?
- Yes. Book a call and we will run a live campaign on one of your own categories, so you see a real shortlist built from your brief before committing. Packages start at Starter, $69/month for 10 campaigns.
See it run on your own category
Book a call and we will run a live campaign on a category you actually buy, so you judge the shortlist on your own data before picking a package.