How the data is produced
Sourcing, extraction, deduplication, coverage, and corrections — the full production pipeline behind every row in FundingData.
01Sourcing
Funding rounds are extracted from public press articles, wire services, and official registries. Older records (mostly before 2016) come from the tech.eu deal database, an editorial database of European rounds.
Every row links its source: the published article it was extracted from or, for deal-database records, the database entry. When several outlets report a round, the row shows the most authoritative one (wire release or tier-1 press before regional press, aggregators and database pages) and lists the others. A round without a citable source is not published. What each year contains is on /coverage.
02Extraction & gating
Structured fields (company, amount, stage, investors, geography, date) are extracted from source text by an LLM pipeline with evidence enforcement: amounts and geography must literally appear in the article text — the extractor cannot invent or infer figures that are not written in the source.
- Each extraction receives a confidence score; rows below the threshold are not published.
- Low-confidence extractions go to a review queue instead of the dataset.
- Closed, not sought. A round is published only when the article states the money is committed — a close, or an announced definitive deal. Reports that a company "aims to raise", "is in talks", "targets" or "is set to raise" an amount are parked as rumors (with the quoted sentence) and are promoted only when a later source reports the close.
- Rows extracted from articles pass the evidence and confidence gates before they are published. Deal-database records are imported as recorded by tech.eu and carry month-precision dates.
03Deduplication
The same round is typically reported by multiple outlets. Duplicates are removed in multiple layers:
- URL identity — the same source article is never ingested twice.
- Semantic key — company + announcement date + amount identify one round across different outlets.
- Fuzzy matching — near-duplicates (name variants, currency conversions, date drift) are caught by a similarity pass.
04Coverage & freshness
Coverage spans 146 countries read from 381 active sources. The corpus is not uniform over time: coverage by year shows, per year, how many rounds cite a published article, how many are corroborated and how many carry an exact day.
| Region | Funding rounds |
|---|---|
| Europe | 50,432 |
| Americas | 35,686 |
| Asia | 14,061 |
| MENA | 5,256 |
| Africa | 1,593 |
| Türkiye | 1,451 |
As of 2026-10-30 18:50 UTC
Sources are read every ~10 minutes. Publisher archives are backfilled through the same extraction and gating as new reporting; deal-database records are imported as described above.
Coverage change, 2026: press-wire and regional feeds were added to the live pipeline in June 2026. Monthly round counts before and after that point are not directly comparable, most visibly for the United States (January–May 2026 counts are lower than June onward).
04bInstrument & scope
- Instrument — every round is equity, debt, grant or secondary, as stated in the source. Debt facilities and grants are listed but kept out of equity totals.
- Scope — rounds of $50M or more are classified by who raised the money: startup, public company, corporate unit, infrastructure project, government body or fund. Public-offering wording is classified by rule; the rest by a referee model, applied only at high confidence. Unclassified rounds count as startup.
- Startup equity — the default totals: equity rounds, excluding public companies, corporate units, infrastructure, government and funds. "All instruments" includes everything.
- Lead investor — recorded only when the source says an investor led or co-led the round.
05Measured accuracy
Audit of 300 seeded rounds, 2026-09-08. 227 had a readable source article; each field was judged against the cited article.
| Field | Correct |
|---|---|
| Company | 100% |
| Amount | 98.6% |
| Date | 99.1% |
| Lead investor listed | 99.0% |
| Geography | 97.9% |
| Stage | 96.3% |
Stage is inferred when the article does not state it. Lead investor listed means the lead appears in the round's investor list.
06Corrections & revisions
A quality pipeline continuously re-checks published figures against their sources. When better sourcing becomes available — a more precise amount, a corrected date, a disclosed valuation — the row is revised rather than duplicated, and keeps its citation.
07Contact
Questions about the data, a specific row, or a correction: