How FinStatGlobe Works: Our Data Sources and Methods

FinStatGlobe publishes fintech and financial-inclusion statistics for every economy the World Bank tracks. This post explains, in plain terms, where the numbers come from and what happens between a source database and a page on this site. The full technical write-up lives on the methodology page; this is the short version.

Three sources, nothing else

Every statistic on this site traces back to one of three public datasets:

• World Bank, World Development Indicators (WDI) — annual macro-financial indicators such as remittance flows, private-sector credit, interest rates, and internet adoption.
• World Bank Global Findex — the demand-side survey of how adults save, borrow, and make payments, fielded in waves (2011, 2014, 2017, 2021–22, and 2024 in our extract).
• IMF Financial Access Survey (FAS) — supply-side data reported by regulators: bank branches, ATMs, mobile money accounts, agents, and transactions.

We do not license proprietary data, scrape vendor reports, or fill gaps with figures from press releases. If one of the three sources does not report a number for a country, the site simply does not show one.

Snapshot, then derive

Our build pipeline downloads the indicators we use from each source’s public API and stores them as snapshot files, each stamped with its retrieval date. As of the snapshot retrieved on June 10, 2026, those extracts contain 117 indicator series (40 from WDI, 59 from Findex, 18 from FAS).

A derivation script then turns the snapshots into one file per country. From that snapshot, the site covers 217 economies and 47 topics, for 6,075 country–topic pages built on 154,311 individual data points. Every derived number — year-over-year changes, growth rates, regional ranks, gender gaps — is computed from the source observations and nothing else. The build is deterministic: rerunning it on the same snapshots produces exactly the same site.

The honesty rules

A few rules are enforced in code, not just in policy:

1. No invented values. Pages are generated only for country–topic combinations with sufficient source data (at least two observations, the latest one recent enough to still be meaningful).
2. Observed values cite their actual year. A 2024 survey result is labelled 2024, even on a site called “Fintech Statistics 2026.”
3. Projections are labelled. The only 2026 figures we show are trend projections computed with a documented capped-growth method, and they are always marked as projections, never presented as observations.
4. Suspect data is dropped, and the drop is logged. For example, some Findex survey waves — and some WDI and FAS series — report a literal zero where a value was not collected for a country; we detect and exclude those sentinel zeros, and every excluded observation is written to an audit log with its reason.
5. Corrections never edit data files by hand. A QA overrides mechanism can suppress a flagged page or attach an editorial note — it cannot change a value.

What’s on every page

Each statistics page shows the latest observed value with its year, the full available time series as a chart and table, derived context (changes, ranks, regional averages, demographic splits where the survey provides them), the exact source indicator code, and the date we retrieved the snapshot — so you can verify any number against the original database yourself.

If you spot something that looks wrong, we want to know: contact us.