Why we started Release Proofing

We are a group of senior operators and entrepreneurs. Between us, we have advised companies through their IPOs, listed companies of our own, run MiFID II regulated entities, sat on the boards and management teams of listed companies, and led their investor relations functions. We have seen the reporting process from every chair in the room.

From every one of those chairs the view is the same: preparing an annual report, or any other financial release, is a nightmare.

Too many authors, one document

Think about how an annual report actually comes together. Tens of people write sections. Finance writes the detailed notes. The CEO writes his or her own review. Risk management writes its section, legal writes its own, sustainability adds another. Each part is written by people who know their area well, in their own voice, against their own deadline.

Then it finally lands on the desk of the team that is ultimately responsible for tying it all together into one consistent story, often the IR team. It is a thankless position: they own the whole document but did not write large parts of it, and every change they make touches someone else's text. That is where the real trouble starts.

Smoother can mean riskier

In a listed company, making a text read better is never a neutral act. Rephrase a sentence about the market outlook and it may start to read as guidance. Move a statement out of the section where it belonged and it may say something the company never decided to say. Round a figure for readability and it no longer matches the note it came from. Tidy up the wording around a KPI and its definition may drift from what the market was told last quarter.

The disclosure rules (MAR, ESMA's guidelines on alternative performance measures, IFRS, the exchange rules) do not care how polished the prose is. They care whether every statement is accurate, consistent and in line with what the company has already published. A well-meant edit can turn a compliant draft into a release that attracts the attention of the financial supervisor.

So the people who sign off are left holding two things that pull in opposite directions: precise detail and a coherent narrative. Nobody really owns the reconciliation of the two. Quality is checked in a meeting that has time for a fraction of the hundreds of sentences and dozens of figures. The errors that slip through are not carelessness. They are a property of the structure.

Then add languages, jurisdictions and exchanges

Now multiply it. A Finnish version and an English version. A Swedish version for a dual listing. Different exchanges, different supervisors, different practices. Every translation is another chance for a claim to shift meaning, for a figure to change, or for a nuance to disappear. Every additional jurisdiction adds another set of rules to check against. The carnival grows every year, and the reporting calendar does not move.

We did not like this, so we built something

Release Proofing is our answer. It is a system of AI agents, each with one job. One recomputes every figure from its components. One compares every claim against the company's full disclosure history. One compares the language versions claim by claim. One tracks KPI and APM definitions from period to period. One checks the text against the applicable rules. The output is a list of findings, each one a concrete point to verify, delivered before the release goes out.

Two principles matter to us above everything else.

First, humans decide. Release Proofing never sends or publishes anything. It shows the people who carry the responsibility exactly where a second look is needed, and leaves the judgement with them.

Second, every run also reports what passed and what was not checked, not only what was flagged. A clean report that hides its blind spots is worse than no report at all.

What we have seen so far

We ran the first real review on publication day, 19 August 2026, using public material only. Since then we have run the system on the published reports of several Nordic and European issuers, and fetched the entire disclosure history of Nasdaq Helsinki since 2019 (1,168 ESEF packages and 7,394 releases with attachments) in a single day. In October we reviewed one group's complete publication base, across three countries and three languages, in six days.

The lesson from those runs is consistent: the issues are rarely dramatic on their own, and they are almost never anyone's fault. They are the natural result of many capable people editing one document under time pressure. That is exactly the kind of problem machines are good at.

Where we are going

Checking is the first step. The next is to help while the report is being written, generating most of the text from the company's previous reports and the decisions management has recorded. Our long-term view is simple: the people set the intent and make the decisions, and the report is produced from that intent every time. When something is corrected, it is corrected in the system, so the same class of error does not come back next quarter.

A good night's sleep

Every team has its own Maija. Out of the tens of people who write and review the release, she is the one who has actually read all of it, from the CEO's review to the last note. If Maija has read the publication and she is happy with the result, the CFO can relax.

Release Proofing is the Maija in every team. It reads every line, every figure and every language version, every time. Once Release Proofing has read the material, the people responsible for the release can relax too.

CFOs, boards and IR teams deserve to go to bed the night before a release knowing that every figure has been recomputed, every claim has been checked against what the company has said before, and every language version says the same thing. Not because a machine took over their responsibility, but because the people who carry it finally have a tool built to help them carry it well.

Welcome to the new era. Welcome to Release Proofing.