← Guides
Guide

AI for deal sourcing: what a screen of public company registers finds

In October 2026 we had an AI screen every company on Nasdaq Stockholm for a question corporate-finance advisers ask: which of them may need to raise money? For the 248 that were the right size, it looked up the 233 with a Swedish organisation number in the company register, then read each company's latest annual or year-end report. The reading took one day. Matching names to the register would have picked the wrong company for 28 of them; the register itself had no revenue at all for 61, and called three too small that weren't.

Most pages about AI for deal sourcing describe a tool. This one shows a screen: the count at each step, where the register data goes wrong, and what a score from the reports can and can't tell you.

The question

Which listed companies worth 100 MSEK to 1 bn SEK, with more than 30 MSEK in revenue, show signs that they may need new capital? It is the kind of list an adviser builds before calling anyone.

The screen, step by step

Step Where it comes from Companies
Listed on Nasdaq Stockholm, main market and First North Nasdaq's own share lists 693
Worth 100 MSEK to 1 bn SEK, share classes added up Nasdaq's page for each share, 22 October 2026 248
More than 30 MSEK in group revenue Each company's annual or year-end report 199 in, 45 out, 4 to look at again
Scored for signs of capital need The same reports 28 of the 199 scored 4 or 5 out of 5

Along the way, 233 of the 248 were joined to the Swedish company register by organisation number, through our data layer. The other 15 are foreign companies with no Swedish number; their revenue came from their reports alone.

Where the register alone gets it wrong

The register has no revenue for many listed companies. It had a revenue band for 172 of the 233 we could join. For the other 61, nothing.

Where it has revenue, it is the parent company's. For a listed group the register's figure comes from the parent company's own accounts, and a holding company's own revenue can be a fraction of the group's. Three companies the register put under 30 MSEK had between about 50 and 170 MSEK in their group accounts. Take the revenue from the group accounts in the report.

A name match finds the wrong company. Joining to the register by name matched 28 of the 248 to the wrong one: a subsidiary such as "… Operations AB" instead of the listed parent, or an unrelated company with a similar name. Join by organisation number only.

And last year's revenue can describe a business the company no longer has. In a handful of cases the latest full year cleared 30 MSEK only because of a business the company had since sold, or a single one-off sale. Reading the report shows this. A filter on the revenue figure doesn't.

The score

For each of the 199 we scored signs of capital need from the latest annual or year-end report, one point each:

  1. a loss in the latest year;
  2. cash at year-end below one year's loss;
  3. a going-concern, financing-need or reconstruction note;
  4. a recent or open share issue, warrants or bridge loan;
  5. more interest-bearing debt than cash.

7 companies scored 5, 21 scored 4, 28 scored 3, 40 scored 2, 61 scored 1 and 42 scored 0. Every point has its reason written beside it, with the report it came from.

A score from last year's report is where the reading starts, not the list to call.

What a person still has to do

The screen gives the company, not the call. Who on your team knows someone at the company is where the call starts. Our guide Claude for M&A and corporate-finance advisers shows how to check a list of companies against your team's LinkedIn connections.

And someone should check a sample of the figures against the PDF of the report. Where the AI couldn't find a figure in a document, our sheet leaves the cell empty rather than guessed.

How to run one yourself

  1. Start from the exchange's own list. The register doesn't say which companies are listed.
  2. Join to the register by organisation number, never by name.
  3. Take revenue from the group accounts in the annual report.
  4. Score the need with a reason for every point, and the report it came from.
  5. Say what you left out. Ours read Nasdaq Stockholm only, not NGM or Spotlight, which list further companies of the same size.

We built it with The Revenue Assistant on our own computer, from public sources only, with the data layer for the register look-ups. Each row of the sheet carries the address of the report it came from.

Questions

Is this investment advice?

No. It is a method and its counts. No company is named, and nothing here says whether to buy or sell anything.

Why don't you name the companies?

The page is about the method. Which companies came out is working material, and we don't publish it.

Does it work outside Sweden?

The register look-ups work in the countries whose registers our data layer holds: over 53 million active companies in all. The reading works wherever companies publish their reports.

Where does my data go?

Your own AI does the work on your computer. Your deals never reach our servers.

What does it cost?

The app is free for 30 days, with no card, and then €49 a month for a team of three. The data layer is €199 a month for the firm, billed from the day you buy it, with a 30-day money-back guarantee.

Counts as read 22 and 23 October 2026, from Nasdaq's share lists, the Swedish company register through our data layer, and the companies' own annual and year-end reports. Nasdaq, NGM and Spotlight are trademarks of their owners. Not investment advice.

Something here unclear or wrong? Write to us — it reaches the person who wrote it.
Augmented Mind AB · PB 13, 00561 Helsinki, Finland