Nicolai Baldin

He built a company to make data that isn't real.

Nicolai Baldin runs Synthesized, a London and New York company that produces production-like data for enterprises who cannot, for legal or logistical reasons, use their own. Banks want to test software against transactions that behave like transactions. Insurers want to share claims data with a vendor without sharing the claim. Baldin's platform manufactures the shape of the thing without the thing itself, and then hands it to a QA pipeline.

In September 2025 the company closed a $20 million Series A led by Redalpine Venture Partners, with IQ Capital, Mercia Ventures, UBS, Seedcamp and Deutsche Bank participating. Deutsche Bank was already a customer. The company is now around 48 people. Baldin has said he plans to roughly double that within a year.

The premise underneath all of this is that testing software has quietly become the interesting problem. Not because tests are new — they aren't — but because the software being tested has changed. Large language models write it. Coding agents suggest it. Non-engineers describe what they want and let a model translate. There is a lot more code, produced faster, by systems that are correct on average and wrong specifically. Somebody has to run it against data that looks enough like production to expose the specific wrongness. Baldin is trying to be that somebody.

His formal answer to why anyone should trust him with this is a PhD in machine learning and statistics from the University of Cambridge and a decade of adjacent research on statistical estimation. His informal answer, delivered to Fortune, was shorter: "You don't need to transfer code to us."

A convex body, and the gap between paper and product.

Baldin's Google Scholar page is not particularly the kind of thing you would expect a startup CEO to leave openly linked. One of his most-cited papers is "Unbiased estimation of the volume of a convex body," published in Stochastic Processes and their Applications in 2016. Another is "Statistical and computational rates in graph logistic regression," co-authored with Quentin Berthet, who now works at Google DeepMind in Paris. It is not obvious how you go from estimating the volume of a convex body to selling test data to Deutsche Bank. The connective tissue is the observation Baldin made while transitioning from Cambridge into work with UK public bodies: there was an enormous, largely unaddressed distance between the academic literature on data and the way large organizations actually handle it.

His summary of the founding thesis, when he told it later, was almost boring in its clarity. The theory was mature. The practice was not. Somebody had to sit in the middle and do the translation. Synthesized became that translation, in a specific direction: from statistics-of-datasets to production-ready synthetic datasets that a regulated enterprise could put through its own pipelines without triggering its own compliance team.

The company incorporated and started serious commercial work in the late 2010s, and closed a £2.2 million seed round two weeks before the UK's first pandemic lockdown in March 2020. That is not the kind of timing anyone plans for. Baldin closed and then, along with everyone else in the country, went home. The product hardened during the period when nobody could visit a data centre.

Two weeks before lockdown, and the loop with Deutsche Bank.

There are two stories worth telling about Synthesized that are not directly about synthetic data. The first is the seed round in early 2020. Baldin closed £2.2 million and, two weeks later, the country closed. Whatever plans the company had for in-person enterprise selling collapsed. The company survived. The product ended up being one that fit an entirely remote posture: an enterprise buyer could evaluate it without physically shipping data anywhere, which turned out to be a competitive advantage that was harder to see before the pandemic than after.

The second is Deutsche Bank. In September 2025, when Synthesized closed its Series A, the bank participated in the round. Deutsche Bank was already a customer. The dual role — investor and buyer — is a signal that most founders will happily accept, because it collapses two hard sales into one relationship. Baldin has used it, publicly and unsentimentally, as a validation credential.

A third detail, quieter: Baldin still keeps a Google Scholar page and still cites his co-authors, one of whom is now at Google DeepMind. The academic identity has not been retired. It sits alongside the CEO identity.