Paste your website address. See the ICP that comes out of it.

• Industries. What the search goes looking for. They aim the queries; they do not reject companies. No registry can confirm that a company counts as “e-commerce”, so dressing this up as a hard filter would be a lie about what the data can do.
• PKD codes. The Polish activity classification, written as a section, a division or a full class. Leave it empty if you do not think in those codes. They aim the search and help match a company to its registry record, but they reject nothing: a Polish company registers many codes at once and keeps them for years, so the code alone rarely settles anything.
• Company size. A headcount band. It cuts wherever size is visible and does nothing where it is not: no Polish registry publishes employee numbers, so the figure is estimated from the company's own site. A company clearly outside the band is held back, and unknown size is never a reason to reject one.
• Legal form. The Polish forms: sp. z o.o., S.A., sp. k., a foundation. It is checked wherever the company could be matched to its REGON record, and a company registered in a form you did not allow is held back. Leave it empty to accept any form.
• Service area. The voivodeships — Poland's regions — you can really serve. Empty means the whole country, which is also the default. Choosing regions changes how the search is planned: queries name towns inside your area instead of spreading across Poland.
• Who buys. The deciding roles on the other side — operations director, warehouse manager, the person who owns IT. Roles, never names: Pursiva describes companies and positions, never private individuals.
• Buying signals. The events that open a window: a new site, a change on the board, a specific job opening, an announced project. Companies showing them score higher. What counts as a signal, and what has to back one up, has its own guide.
• Exclusions. The most important field in the profile, and the one no ICP article teaches. Exclusions are the company types to skip even when they fit on every other axis: your current customers, your competitors, resellers and middlemen, public institutions, industries you will not work with, companies that already said no. Every guide tells you to describe who you want. Real lists are ruined from the other end — by rows that technically qualify and that you knew in advance were worthless. Write them down before the search runs, because otherwise you read them back off your own list.

A profile built in a meeting describes the customer you would like to have. A profile derived from your site describes the offer you actually have. The gap shows up on the first list: if the workshop concluded that you target large manufacturers, while your site describes three deployments at mid-sized distributors, then mid-sized distributors are your market and large manufacturing is an ambition.

The model reads specific sentences — what you do, which problems you solve, what your case descriptions say, how big the companies you name are, how you talk about price and implementation. Industries, the size band and the deciding roles follow from that material, which is the same material your buyers read before they ever reply to you.

The second reason is mundane: a whiteboard profile is never tested. It sits in a document, nobody opens it, and nothing depends on it. The profile in Pursiva becomes the input to the search itself — it sets the queries, the ordering and the scoring, and its exclusions are checked against every company that is scored.

The third is speed. You get a draft in minutes rather than after a meeting is scheduled, you can correct three fields, read how it sounds afterwards, and only then decide whether you want a search at all.

In most tools an ICP is a document. Here it is code. When you confirm the profile, the exclusions, legal forms, headcount band and service area are stored as conditions rather than as advice for a model to follow when it feels like it.

Before a list leaves, every scored company goes through a deterministic post-check — a separate step in the code, not one more instruction handed to a model. It does not guess and it has no opinion: it compares the facts stored about the company with your exclusions, with the legal form from the registry, with the headcount band and with the regions you chose. The same input always produces the same verdict, on every run.

That check is automatic and it runs before the results are sent. There is no queue and no stage where a list waits for somebody to look through it — the run delivers itself, and results arrive within an hour. The quality commitment is the test in the code, not a promise that somebody will get round to your list.

One rule is not yours to switch off, whatever the profile says: no private individuals appear on a list. Every company name is read for a person's first name and surname, and one that cannot be settled is held back. Pursiva delivers companies, not people — which is also why you will not get an e-mail address or a phone number for anyone here.

The profile finds nothing by itself; it is a description, not a list. The finding is done by the search: planning queries, reading company sites, looking for signals with a source, verifying records in the registry and scoring what survives. That is the part that costs money, and that is what you pay for.

One run is PLN 149 net plus 23% VAT, so PLN 183.27 is what you actually pay. Results arrive within an hour and the ranking holds up to 40 companies with a why-fit and a why-now line, each signal carrying the link it was read from.

What that looks like in practice is on the one published run — made 2 Sept 2026 on the offer of a Polish ERP and WMS vendor, our own submission, and that vendor is not a customer: 111 candidates from search, 40 companies scored and ranked, 35 minutes end to end. It is published in full, including the rows that did not hold up, so you can read it before you decide.

The order is deliberate and not the usual one: you see the profile built from your site, you correct it, and you pay only once you agree with it.

First run one-off

PLN 149 + 23% VAT (PLN 183.27)

Weekly watch a month

PLN 299 + 23% VAT (PLN 367.77)

Prices are net; 23% VAT is added at checkout. Results arrive within an hour of payment.

Frequently asked questions

Do I have to pay to see my ICP?

No. Reading your site and building the profile is free, with no account and no card. Payment appears only when you confirm the profile and start the search — the part that genuinely costs: market research, reading company sites, registry checks and scoring. If the draft profile does not convince you, you can stop there and pay nothing.

Can I change the profile I am shown?

Yes, and that is what the screen is for. Every field is editable: add industries, narrow the headcount band, pick regions, add exclusions and deciding roles, delete whatever the model read off your site incorrectly. You edit before you pay. Once the search starts the profile is locked, because it is what sets the queries and the scoring — otherwise the list would answer a different profile than the one confirmed.

What if my website says very little about the offer?

Then the draft profile comes out generic and you will see that immediately — which is useful information about the site. You have two ways forward: paste the address of a page describing one service or a case study instead of the home page, or type into the profile what the site does not say. You are the one who confirms it, so your correction always beats the model's reading.

Read next

• Buying signals
• What a list of companies to call really costs
• Software houses
• A run published whole, with the rows that did not work
• Start a run