AI supplier discovery: how to build a verified shortlist

István Olasz, oiSystem · 7 November 2026

To find suppliers with AI, start with a clear purchasing brief, search traceable sources, and build a shortlist that separates evidence from assumptions. Verify each company's identity, documents and product capability before requesting comparable quotations. AI can organise the research; a person must approve the supplier and the purchase.

In a small B2B company, supplier research happens between customer calls and urgent orders. AI can structure that work, but a plausible list is not a procurement decision when technical requirements and customer commitments are at stake.

1. define the purchase before searching the market

Write down what a suitable supplier must actually deliver. Include the product specification, material, dimensions, tolerances and intended use. If a drawing is involved, identify its revision. Separate mandatory requirements from preferences: a required material grade is a gate; a nearby warehouse might simply be a useful advantage.

State the initial quantity and expected repeat demand. Set a workable minimum order quantity, or MOQ, rather than discovering late in the process that a promising manufacturer only accepts orders beyond your budget. Distinguish the quotation deadline, sample deadline and required arrival date. “Delivery in six weeks” means little unless everyone agrees when the clock starts.

Define the search region and destination market. Distance affects freight, customs, communication and the practicality of a site visit. Specify the certificates and product documentation you genuinely need, including their scope. A certificate covering a company's management system does not automatically demonstrate that your particular product meets the relevant requirements.

a purchasing brief you can copy

Keep the first research brief free of confidential drawings, customer identities and unnecessary personal information. Only share sensitive material through tools and processes approved by your business. You can usually describe a manufacturing requirement well enough for an initial search without uploading the full design.

2. ask AI for a research method, not an answer to trust

Use AI to expand the search vocabulary. Ask for product synonyms, manufacturing processes, relevant industry terms and local-language expressions. A distributor and a specialist manufacturer may describe the same item differently. Searching only your internal purchasing name can miss useful candidates.

A practical prompt is: “Using this brief, suggest search terms and potential supplier sources. For each candidate, provide an accessible source URL, the date checked and a short supporting extract. Mark anything you cannot verify as unknown. Do not infer certifications, capacity or pricing.” If the tool cannot browse, ask it for a search plan rather than a current supplier list.

Look across manufacturers' websites, trade association directories, exhibition catalogues and specialist marketplaces. Directories are discovery tools, not endorsements. Open the underlying pages yourself: an AI answer may combine different companies, quote an outdated catalogue or mistake a reseller for a manufacturer. A search snippet is not a substitute for the original source.

Capture the source at the point you find the claim. A product page may support a capability statement, but not a current lead time. Those should remain separate fields. This small discipline prevents a polished summary from quietly becoming stronger than the evidence behind it.

3. build a shortlist you can challenge

A useful shortlist records more than names and contact details. Include the legal company name, country, manufacturer or distributor status, relevant capability, evidence links and verification status. Keep MOQ, price and lead time as unknown until a reliable source or a direct quotation establishes them.

The table below is illustrative only. Supplier A, B and C are placeholders, not real businesses. The scenarios are invented to demonstrate the workflow; they are not recommendations, verified findings or market offers.

Use explicit stages such as discovered, screened, verification pending, ready for quotation and excluded. An unknown requirement is not a passed requirement. If you use scoring, apply mandatory gates first: a low price should not compensate for an unproven safety requirement or an impossible arrival date.

4. verify identity, capability and evidence separately

check who you would be buying from

Use the relevant official company registry to check the legal name, registration status, address and identifiers. For Hungarian businesses, the official company register is a starting point. The European Commission's VIES service can check an EU VAT number for cross-border transactions. A valid number is not proof of financial strength or product quality.

Compare the registry details with the website's legal information and the entity issuing the quotation. Establish who manufactures, who invoices and who handles warranty claims. If payment details change, confirm them through an independently established contact channel, not the new number included in the change request.

read the certificate, not just the logo

Check the certificate holder, covered site, scope, validity and issuing body. Where available, verify the record through the issuer's official channel. A PDF attached to an email is evidence to investigate, not automatic proof. Product compliance depends on the item and destination market; seek qualified advice when the requirements are unclear.

Ask for references relevant to your application and order size. A page of recognisable customer logos does not show that the supplier can make your component. Request a reference contact with permission to approach them, or a suitably anonymised example of comparable work.

test the product and notice warning signs

Agree sample acceptance criteria before the sample arrives. Check dimensions, material, function, packaging and accompanying documentation as applicable. A successful sample does not establish consistent production quality. Depending on the risk, a controlled trial order or a manufacturing review may still be necessary.

Pause when legal names do not match, certificate claims cannot be checked, technical answers contradict each other or delivery promises look implausible. Pressure for immediate full prepayment also deserves scrutiny. One warning sign does not prove fraud, but it should trigger a documented check rather than a reassuring AI explanation.

5. send a comparable request for quotation

A request for quotation, or RFQ, should give every candidate the same specification and quantity breakdown. Ask suppliers to separate unit prices, tooling, samples, packaging and transport. Specify the currency, treatment of taxes, quotation validity and payment schedule. Request an explicit list of deviations rather than assuming silence means compliance.

Clarify the delivery terms, including the named place and applicable Incoterms edition where relevant. Ask when production begins, what approvals are required and what could change the lead time. Include the process for defects, replacements and warranty claims. The lowest unit price is not necessarily the lowest cost of receiving usable goods.

AI can extract quotations into a comparison table and draft questions about missing fields. Check every material number against the original documents before deciding. Currency, pack size, net versus gross pricing and excluded freight can change the comparison. Keep the original files attached to the decision record.

6. keep ownership of the decision

AI cannot independently inspect a factory, guarantee document authenticity or know a supplier's live production schedule. It can also confidently attribute one company's information to another. Assign a person to approve technical suitability, commercial terms and the first order. Research support should not become permission for an automated system to commit your business.

Finish with a decision record: what passed, what remains open, who owns each unresolved question and why the preferred supplier was chosen. Save source dates and quotation versions. When requirements change or you reorder later, that record shows which assumptions need to be checked again instead of starting from an unexplained ranking.

frequently asked questions

which AI tool should a small purchasing team use?

Prioritise accessible citations, clear handling of unknowns and acceptable data controls. Test the workflow on a product you already understand. A tool that produces an attractive table without traceable evidence is not enough.

is a supplier website enough to verify a company?

No. Combine first-party claims with official records, checkable documents and direct contact. Product suitability may also require samples, testing or a specialist review. Different checks answer different questions.

how many suppliers belong on a shortlist?

Keep enough viable options for a meaningful comparison, but no more than you can properly evaluate. In a specialised market, a small evidence-backed list is more useful than padding it with weak matches to reach an arbitrary target.

can AI contact suppliers automatically?

It can be part of an approved workflow, but a person should first review the recipients, message and information being shared. Start with draft RFQs and follow-up questions before enabling any outbound automation.

turn the search into a repeatable process

The goal is not a longer list. It is a shortlist your team can explain and act on. If you want help shaping an evidence-led AI workflow around your own purchasing needs, talk to oiSystem. A concrete requirement and the brief above are a useful starting point.