Method sheet

How a target list is made, and how far to trust it

This page describes the method behind every campaign: where the facts come from, how they are checked, and how sites and personal data are treated.

Sources

Search: Brave Search API, in the target country's language and region. Queries are product groups, industries and municipalities, not company names.
Sites: Only the companies' own websites. Directories, marketplaces, job boards and news sites are skipped before any page is loaded.
Pages per company: At most eight: home, products, contact, about, dealers, management, careers and, where needed, the Finnish office page.
Not used: LinkedIn, bought lists, guessed email addresses, or anything behind a login.

The language model classifies and extracts. Code decides what is accepted.

Claims: Accepted only as a short quote found word for word on a fetched page. Otherwise the field stays empty.
Names: A person's name must appear on the page as whole words. A title is accepted only next to the name.
Contacts: Emails and phone numbers are taken from the page text and links as they are, never from the model. A person gets only the ones next to their name. Foreign dealers' numbers are kept apart.
Located in Finland: A Finnish address, a +358 number, a Business ID with a valid check digit, or an FI VAT number. Domain or language alone is not enough.
Rejections: Pure subcontracting shops, dealers of other brands, directories and companies without their own products are rejected, with the reason recorded.

Priority is computed in code from the evidence. Size does not count.

A: Own products, a dealer network or exports, and a named decision-maker or salesperson with their own contact details.
B: Own products and at least one sales signal: exports, dealers, sales hiring or a named decision-maker.
C: Own products with no other signal, or subcontracting as the main business.
Export level: Exporter: an export statement, or two or more foreign countries or three foreign numbers on dealer and contact pages. Likely: one foreign country or the site's language versions. Domestic: Finnish places only.
Structure: Parent, subsidiary or division, stock listing, and whether the unit runs itself: its own domain, brand, sales or management.

robots.txt: Obeyed on every site (RFC 9309), including Crawl-delay up to 10 seconds.
Pace: At most two requests at a time per site, at least 250 milliseconds apart.
Identification: The agent identifies itself with its own user agent and contact address.

Language model: Runs on our own hardware. Page text is not sent to a foreign AI service.
Personal data: Only role contacts the companies published themselves, with their source pages. Processed on legitimate interest for B2B sales.
Removals: Removal requests and objections are honoured and passed on for data already delivered.
Messages: We send no messages to target companies. The customer decides on any contact.

Visibility: A company without its own website, or whose pages are script with no text, is missed or yields less.
Coverage: Search is not a register of every company in a field. A run continues until no new companies turn up, and reports where it stopped.
Freshness: Facts are what the pages said at run time. The monthly refresh picks up changes.