The next generation of finance companies will be data companies.

In the age of AI, the finance companies that win will be the ones that own their decision data. Models commoditise. Weights converge. What does not commoditise is the record of what a professional knew at the moment of decision, how they reasoned, and what happened next. You cannot scrape that record off the internet, and you cannot synthesise it credibly. It exists only where real institutional work happens. That is why we run a lending business to produce it.

Between roughly €1m and €40m of ticket size sits an annual pool of European credit demand worth low tens of billions. Institutional capital serves close to none of it. Above forty million, direct lending works and competes. Below one million, regional banks still write standardised products that fit risk-weighted templates. The band between has been structurally abandoned for fifteen years.

The banks left first. After Basel III, anything bespoke became too expensive for a bank balance sheet. If recovery depends on specific collateral or a specific workout path, the capital rules bite. The credit officer who knew the borrower by first name lost the authority to use judgment exactly when judgment was the right tool.

The funds left for a blunter reason. Underwriting cost barely varies with ticket size. A €5m deal needs roughly the same analyst-weeks, committee slot and legal bill as a €500m deal, while earning a fraction of the income. Point your analyst-weeks at five-million tickets and you underperform yourself out of a job. Funds that drifted into small tickets without changing the underlying economics died there. The floor was never about risk appetite. It was arithmetic.

The market's answer to this gap has always been more people. Hire more investment professionals, expand market by market, grow fixed cost alongside assets under management. Technology improved workflows inside that organisation without changing the organisation itself. Fifteen years and billions of euros of fund formation later, the band still pays mid-teens net returns on risk that is frequently senior to hard assets, statutory claims or long-dated receivables. Prices like that come from absence of competition, not presence of it.

The failure was never insight. Everyone could see the gap. It was organisational. Nobody rebuilt the firm around a different cost structure, so every attempt inherited the economics it set out to fix.

Step inside a traditional financing firm. Fixed costs everywhere: office floors, management layers, support functions sized for volume that never comes. A graduate spends their first five years writing memos, running compliance checks and feeding processes. Almost no investing. Almost no clients. Real deal-making waits for seniority, so the people closest to the information never develop judgment, and the people deciding last touched a live situation years ago. Careers stall because the promotion track rewards managing process, not owning outcomes.

Yet the one input that actually moves mid-market finance is a relationship. An originator who trusts you. An administrator who returns your call first. Relationships do not scale through headcount. They scale when the person holding them owns the outcome and shares in it. That, plus aligned capital, is the shape of the next financing firm.

We think AI's real leverage in finance is organisational. A copilot bolted onto a 1990s credit process makes the old firm slightly faster. Changing who does what changes the firm: an AI-native platform carries underwriting preparation, booking, monitoring, fund operations and investor reporting, while experienced credit professionals carry judgment and relationships.

We scale through joint ventures, not headcount. Dissei brings capital and the platform. Local professionals bring clients and specialist expertise, and are paid through promote economics rather than salaries. Origination becomes a variable-cost network with aligned incentives instead of a department with a payroll.

Venture capital's scout model showed how this works and why allocators want it. Scout programmes gave trusted operators small cheques and carried interest, letting firms like Sequoia reach founders years before a fund cycle could, without hiring a single partner. Large allocators backed them for exactly that reason: earlier access to deal flow, real incentive alignment, zero fixed cost. Our joint ventures run the same play with higher stakes. Our partners do not write small cheques; they underwrite complete transactions on shared infrastructure, with promote waterfalls tied to realised outcomes. For allocators, this opens local European deal networks that cannot be assembled from a desk in any capital city.

Origination stops being a cost centre and becomes a network.

Eligible activity produces a dated process record: the documents available, the analyses and decisions preserved in the transaction record, and the subsequent state changes and outcomes. Through anonymisation and bounded practitioner calibration, those records become training data, evaluations and environments.

Public financial data can teach facts, calculations and source grounding. It rarely contains the complete institutional chronology: what was knowable, which action was taken, how the decision changed and what subsequently occurred. That missing join is produced inside real financial work. Our JV partners produce the genuine article as a by-product of paid deal work: representative of real institutional workflows, rights-cleared through the lending process, impossible to scrape or fake. And this asset monetizes on its own, as licensable process data, evaluations and environments, independent of the lending book.

The loop closes. Lending creates the record. The same records train the models, and better models mean better underwriting. Better underwriting wins more of the abandoned band, which produces more records. The advantage compounds on two axes at once: cost per deal falls while judgment sharpens.

Models and synthetic tasks can be purchased or reproduced. Missing decision ground truth cannot be synthesised after the fact. Dissei’s advantage comes from participating in the activity that continually produces it.

On a first check, the system's job is speed: first mention to defensible preliminary decision in under an hour, most of it machine work, every claim tied to a grounded source. But a first check is not an investment. It is permission to spend real human hours. When capital commits, the machine prepares every structured input so the human can argue with it, then gets out of the way. A system that tries to make the commit decision itself is more dangerous, not better.

Human credit authority sits at every gate. Where that line falls between machine preparation and human decision is the most important architectural choice in this company. We hold strong opinions about it.

One company designed together from the start: the technology, the lending institution and the organisational model. We lend €1–40m senior secured to European mid-market finance companies, beginning in the Nordics and Germany, and license financial-reasoning environments to frontier labs. Licensing is the gate. Nothing deploys before it, and capital waits behind it.

None of this is secret. All of it is hard. The mispricing persists because describing an operation is nowhere close to running one. If you want this company to exist, as a lab buying environments, a lender putting capital to work, or a credit professional building your own book on our platform, talk to us.