Start-ups are surprisingly hard to study. Official statistics register them as companies, but say little about what makes them start-ups: funding rounds, patents, founding teams, acquisitions. To fill that gap, the OECD has assembled a global database, presented in a working paper published in March 2026.
Nearly 4.5 million companies
The OECD Start-ups Database covers nearly 4.5 million start-ups founded between 2000 and 2025, with information on funding, intellectual property, founders, financial indicators and exits. The paper describes the results as preliminary.
The authors explain why the task is difficult. Start-ups are small, privately owned and often short-lived, which makes it hard to collect consistent data and compare countries.
Several sources combined
There is no single source that covers start-ups well, so the OECD combined several. The basis is two commercial databases: Crunchbase, created in 2007 with a focus on the United States, and Dealroom, created in 2013 with better coverage of Europe. To these the OECD added data on patents and trademarks, company accounts, public and corporate venture capital and other transactions.
The result reaches beyond the venture-capital-backed companies that earlier databases mainly captured. That matters, because only around one in ten companies in the database has any recorded external funding at all. Venture capital remains the dominant source of funding, but other forms of financing also play a significant role, particularly in Europe.
The authors tested the database against venture capital statistics from national and European industry associations. It consistently records more deals and larger amounts than those sources. They have also deliberately avoided a single definition of what counts as a start-up, and state that the database does not aim to be statistically representative.
Companies appear three to four years late
The most revealing passage concerns what the database cannot see. Neither Crunchbase nor Dealroom has access to official business registers. Whereas registers record a company when it is founded, the commercial databases tend to pick it up only when something makes it visible, such as a funding round or a product launch. On average, it takes three to four years for a company to appear in the data.
The authors therefore warn that delayed registration, the risk that mainly surviving companies are captured, and uneven coverage between countries must be considered when the data are used to describe trends. These limitations, they note, apply to all sources that are not built on official registers.
At the same time, the paper is clear that register-based data have their own weaknesses: they usually cover fewer countries and contain fewer variables. Each type of source sees things the other does not.
Who sees companies in their first years
The OECD’s observation raises a question about where knowledge of companies’ earliest years actually exists. One answer lies with the organisations that work with companies from the start. In a separate report on incubation, the OECD describes how incubators scout for promising entrepreneurs and select and prepare new ventures. Much of that work is likely to take place during the years when, according to the database paper, global databases do not yet see the companies.
What such organisations know about their companies, including when they started and what support they received, is usually kept in their own records, separate from both official registers and international databases.
Combining sources locally
The OECD’s own method shows what can be gained by combining sources. Its database became more complete by linking commercial data with information on patents, trademarks and company accounts. A similar approach is conceivable at regional level, where a support organisation’s records could be linked to official business registers through company registration numbers, making it possible to follow companies from their first contact with the support system.
There are already examples of data being collected across support organisations. In the United Kingdom, incubators that receive public funding are required to share data, and in Portugal the national incubator network collects data on its members’ activities.
Every source has blind spots
The OECD Start-ups Database is a valuable new resource, and its authors are unusually open about its limits. That openness is perhaps its most useful lesson: every data source has blind spots, and a fuller picture comes from combining sources whose strengths and weaknesses complement each other.
