Methodology

European startup funding — 38,360 companies and 51,644 funding rounds across 79 countries, with 30,339 named investors. About 98.4% of companies are European; the rest are non-European entries with little or no round activity.

…and 55 more countries.

Rounds in this snapshot span 1994–2026; coverage is comprehensive from roughly 2010 onward and sparse before. The most recent round in this snapshot is dated 2026-07-17.

Every disclosed amount is converted to euros at the European Central Bank reference rate for the round's date — the nearest published rate on or before that day (the ECB does not quote weekends or holidays). We load the full ECB daily series back to 1999; the original amount and currency are always kept alongside.

Of 51,644 rounds, 44,540 carry a disclosed amount: 21,998 were converted from another currency at the at-date rate and 22,542 were already in euros. The remaining 7,104 are undisclosed and shown without an amount — never an estimate. Recomputing every priced round this way totals €701.6B; re-pricing only the foreign-currency rounds and comparing to the stored figures reconciles to within 0.71%.

Each figure carries a confidence level from two independent checks. First, conversion quality — how close the ECB rate is to the round's date (within a week counts as high). On this measure the euro figure itself is rarely the uncertain part. Second, the source — whether a round links to a recognised funding-news publisher. We treat every reputable outlet equally, with no preference for Tech.eu over Sifted, EU-Startups, Webrazzi, Reuters or others.

31,575 rounds trace to a recognised publisher, 8,126 carry only an unverifiable link, and 11,943 have no source link recorded (77% carry some source). Combining both checks over priced rounds: 27,734 high, 7,081 low, 9,721 unscored — where confidence is lower it is almost always the citation that is missing, not the amount.

Most-cited sources: tech.eu · finsmes.com · deutsche-startups.de · siliconcanals.com · eu-startups.com · startupticker.ch · elreferente.es · uktech.news.

Investor names are recorded inconsistently ("Accel" vs "Accel Partners"). We normalize each name and strip corporate suffixes to merge obvious variants, resolving 33,947 distinct names into 30,529 investor entities (30,339 named, after setting aside generic and undisclosed placeholders — the figure shown in the directory). The merge is deliberately conservative: no fuzzy guessing, short or generic names ("angel", "seed") are never merged, and large clusters are flagged for review rather than force-joined. We would rather leave two records separate than wrongly fuse two different firms.

Companies list free-text industries; we map each to one of 31 canonical sectors aligned to a Dealroom-style taxonomy. Vague business tags fold to Software, vague consumer tags to Consumer, and stray non-sector tokens are dropped rather than mis-filed. Funding stages are likewise mapped from the announced label to 13 consistent stages, with the original label kept.

The signals (Heat, likely-to-raise-next, exit candidates) are an explainable base-rate model — not machine learning and not a unicorn predictor. We compute cohort statistics from this dataset — typical time between rounds per stage, how often a stage is followed by another, round size relative to peers that year, and an investor's historical re-investment rate — and combine them as transparent, labelled factors ("in typical raise window", "up-round momentum", "high-conviction backers"). Every score shows its base rate and the factors that moved it.

The signal has measurable predictive lift. In a back-test (cut-off 30 June 2023, 18-month horizon), companies the model placed in their typical raise window went on to raise again about 1.8× as often as the average tracked company (~14% vs ~8%), while companies flagged dormant raised under 1% of the time. These scores describe patterns in past funding behaviour; they are not predictions and not investment advice.

A deterministic scan flags anomalies — empty records, missing, implausible or future-dated rounds, and likely duplicates. When a fix is researched, an AI step may only propose a correction strictly from cited web sources: it must abstain rather than guess, the company and amount must match, and a person reviews every change before it is applied. The AI never writes to the database directly, and every applied change is recorded and reversible.

Some narrative summaries on the site are written by an AI model and labelled as such. They are grounded in figures we already computed from this database (the model is only shown our own data), figures are checked against that data, house style is enforced deterministically, and every draft is reviewed by an independent second model before it is published — anything unverifiable is held back for human review. All core data, charts and rankings are computed without any AI. This labelling is designed to meet the transparency obligation of EU AI Act Article 50 (which applies from 2 August 2026).

• This database is not exhaustive. It focuses on European companies and is comprehensive from roughly 2010 onward; earlier years and very early or unannounced rounds are under-represented.
• Undisclosed rounds appear without an amount rather than an estimate — we do not fabricate figures.
• Some rounds lack a verifiable source link; where a figure's confidence is low, it is the citation that is missing, not the euro amount.
• Sector and stage labels are normalized and can occasionally mis-bucket an edge case.
• Signals describe historical patterns, not the future, and are not investment advice.
• Founder gender, where shown, is confidence-gated and never asserted as fact or inferred.

We process business and publicly-reported funding information on the basis of legitimate interest under the GDPR. To request a correction or removal, contact Tech.eu.