A month of manual checking, done in hours

Our own onboarding took up to a month, and one step accounted for most of it.

Before a single message could be sent for a new client, someone had to establish what they actually sold and to whom. That meant a conversation about target markets, industries, company profiles and the roles that decide. Then market research to understand how deep the market really was. Then building a raw database of companies matching that shape.

None of that was the bottleneck.

The bottleneck was checking the companies one at a time. A raw database is a list of names, not a market. Somebody had to open each company's website, read it for a few minutes, and decide whether this company genuinely fits what the client sells. Ten thousand companies in the raw list meant a month of work before the client received anything at all.

That is not a scaling problem. It is a wall. Every new client cost a month of a person's attention before the actual work began, and the number of clients we could take was set by how many websites a human could read.

A pipeline that starts from the client's own words and ends with a classified market — in hours rather than a month.

It starts in the meeting. The client conversation is transcribed, and the engine extracts the brief directly from what the client said — not from a form they filled in afterwards, and not from someone's notes. What they sell, who buys it, where, and what makes a company a good fit.

That becomes a specification. The engine turns those insights into a structured market definition: target countries, industries, NACE Rev. 2 codes and the supporting parameters that make a market searchable rather than describable.

The specification drives collection. Companies matching that definition are gathered automatically from a licensed commercial business database, country by country — a step that used to be assembled by hand.

Then every company's website is read. Each site is crawled and a model extracts what the company actually does: materials, services, products, certifications, ISO standards, and a summary written from the point of view of what our client is looking for. This is the step that used to consume the month.

And then the part that matters most.

We describe the ideal customer once. The system builds the archetypes and classifies the whole market against them.

The client tells us what a good company looks like — the types of business, what they must be able to offer, which certifications matter, what machining capability is required, whatever actually defines fit in their industry.

Rather than checking each company against that description one by one, the system first derives several distinct archetypes of target company from it — because "our ideal customer" is almost never one shape. It is three or four, and a human doing this by hand holds them loosely in their head and applies them inconsistently across ten thousand decisions.

Every company in the database is then classified against those archetypes automatically.

This single step turned a month into a couple of hours. It is the difference between a raw list of companies and a market a sales team can work, and it is the part of the pipeline we would not replace.

The reach numbers follow from that, rather than the other way round: a single run has covered 18 countries and roughly 40 industry codes, returning 7,925 companies for Austria alone and 21,938 companies qualified in one classification run.

Those figures are not the achievement. They are what becomes possible once nobody has to open ten thousand websites by hand.