How often do World Bank projects use geospatial data?

Evidence from 2,261 projects, read through 4,494 Project Appraisal and Project Information Documents

Count by
Out of

In 6 in 10 World Bank projects that use data (1,165 of 1,932, 60%), teams draw on geospatial data, mostly at the level of regions and districts. How reliable is this?

60 of every 100 projects that use data
  1. 2,261projects reviewed: every project with a PAD or PID in the collection
  2. 1,932use data in the passages read
  3. 1,165use geospatial data

Used in every sector and region

By approval year, the share runs from 59% (2021) to 62% (2023). 7 years of recent approvals are too few to show a trend.

Among the 111 countries with at least 5 projects, the share runs from 29% in Kyrgyz Republic to 100% in Guatemala.

So whatGeospatial data is already part of routine project preparation, not a specialist add-on.

Projects using geospatial data, by approval year

Under each year: the number of projects counted.

* 2026 counts approvals to 22 September.

Share of projects using geospatial data, by country

  • under 20%
  • 20–39%
  • 40–59%
  • 60–79%
  • 80% or more
  • fewer than 5 projects
  • none reviewed

Point at or tap a country to see its figures.

Not on the map: 117 regional or multi-country projects; 56 projects in 23 countries with fewer than 5 are grey. Region, country and approval year come from the World Bank Projects API.

Boundaries: World Bank Official Boundaries (Admin 0); small-state dot locations remain Natural Earth markers.

All 111 countries with at least 5 projects
Countryprojectsshare
Guatemala8100%
Brazil6392%
Micronesia, Federated States of786%
Afghanistan1385%
Pakistan4183%
West Bank and Gaza2882%
Iraq1182%
St. Vincent and the Grenadines580%
Niger1479%
Haiti1776%
Kenya2576%
Bolivia875%
Bosnia and Herzegovina1275%
Dominican Republic875%
Sudan1275%
Turkiye4374%
Sierra Leone1974%
Uganda1974%
Burundi1573%
Cameroon1173%
Cabo Verde771%
Central African Republic1471%
El Salvador1471%
Tonga771%
Armenia1070%
Kiribati1070%
Tanzania2268%
Kosovo1267%
Lebanon2167%
Myanmar967%
South Africa667%
Tajikistan2467%
Viet Nam1567%
Philippines2665%
Lao People's Democratic Republic2365%
Ghana1765%
Mali1464%
Solomon Islands1164%
Ukraine2264%
Mozambique3863%
Honduras1662%
Nepal2462%
North Macedonia862%
Peru1662%
South Sudan2162%
Yemen, Republic of2162%
Gambia, The1362%
India6261%
Chad1861%
Belize560%
Bhutan560%
Cambodia2560%
Ecuador1060%
Egypt, Arab Republic of1060%
Paraguay560%
St. Lucia560%
Timor-Leste560%
Senegal1759%
Comoros1258%
Somalia, Federal Republic of3158%
Liberia1958%
Indonesia2658%
Bangladesh4257%
Maldives1457%
Nicaragua757%
Angola1656%
Congo, Republic of1656%
Guinea1656%
Ethiopia2756%
Mongolia956%
Zambia2055%
Serbia1154%
Mauritania1354%
Madagascar2152%
Uzbekistan2352%
Argentina2050%
Benin1450%
Costa Rica850%
Eswatini650%
Fiji650%
Georgia1050%
Lesotho850%
Malawi2050%
Marshall Islands850%
Mexico650%
Moldova1450%
Nigeria2650%
China2348%
Rwanda2348%
Burkina Faso1947%
Tunisia1547%
Jordan1346%
Togo1146%
Colombia944%
Papua New Guinea944%
Guyana743%
Morocco1443%
Sri Lanka1443%
Kazakhstan540%
Montenegro540%
Tuvalu540%
Sao Tome and Principe1338%
Panama838%
Djibouti1937%
Congo, Democratic Republic of2236%
Albania933%
Cote d'Ivoire1833%
Croatia633%
Guinea-Bissau933%
St Maarten633%
Kyrgyz Republic2129%

By sector

From 38% in Financial Sector to 77% in Transportation.

Primary sectorprojectsshare
Transportation18177%
Water, Sanitation and Waste Management13172%
Education18065%
Industry, Trade and Services8865%
Agriculture, Fishing and Forestry22864%
Social Protection19863%
Digital Development6962%
Public Administration28161%
Energy and Extractives17152%
Health29745%
Financial Sector8138%

By region

A narrow spread, 57% to 67%.

Regionprojectsshare
Latin America and Caribbean28267%
Middle East, North Africa, Afghanistan, and Pakistan21066%
East Asia and Pacific24161%
Europe and Central Asia25960%
South Asia16359%
Eastern and Southern Africa42257%
Western and Central Africa34457%

Frequently asked questions

What period does this sample cover?

The sampling period is January 2020 through September 2026. The World Bank project-list filter for this period shows 3,557 projects; PAD/PID documents were found for roughly 2,000 of them and enter this analysis.

Out of which projects?

Those that use data (the default) counts projects with at least one data mention judged used. All projects adds the 329 projects that mention data only as plans, bare figures or sources not relied on. Every one of the 2,261 projects in the collection has at least one reviewed passage, so this is every project, not a sample of them.

Projects or documents?

A project usually has more than one PAD or PID. Projects count a project once: it uses geospatial data if any of its reviewed documents does. Documents count each PAD or PID on its own. The project view is the reporting unit, because region, sector and approval year belong to the project. The project share is higher because a project has more text in which geospatial data can appear. The same document sometimes arrives from two collections under two ids; it is counted once, by its World Bank document id, and a passage read twice counts once. 4,708 confirmed document ids are 4,679 distinct documents.

What counts as geospatial data?

Data the document ties to a place: an exact location (a school, a gauge, a village), a named area (a region, district, camp or an urban/rural split; admin level 1 or 2 is enough), or a figure derived from maps or satellite imagery. Coordinates are not required. A topic such as agriculture or displacement does not count on its own.

What counts as using data?

Each data mention was judged for use: does the passage rely on existing data, or only name it, plan to collect it, or cite a bare figure? Only mentions judged as used count.

Why not count datasets?

The same dataset is named many ways (“DHS”, “2016 UDHS”, “Demographic and Health Survey”), so counts of distinct datasets are unreliable. Whether a project or document uses geospatial data is a clear yes or no. Source names are merged across spellings for display only.

How reliable are these figures?

Every number is counted from the records; none is typed in. The figures rest on AI-extracted data-use mentions from a sample of passages: an AI model found each data mention, judged whether the data was used, and labelled its geography against a written rubric, quoting the passage words behind each place label.

Every share is a floor. Each project was judged from a few sampled passages, not its full documents. A passage is an excerpt of a PAD or PID, about a page long, taken where the text mentions data; it is not a data mention, and one passage can hold several. 519 of the 1,932 data-using projects have only one passage read. The more we read, the more we find: of the data-using projects with one passage read, 39% use geospatial data; with five or more, 87% do. Projects with two or more documents reviewed reach 68%, against 43% with one document. Larger, data-heavy projects also had more passages read, so part of the rise reflects the projects themselves. Read in full, more projects would likely count.

Data-using projects with geospatial data, by passages read

PADs show more than PIDs. Where a PAD was reviewed, 70% of data-using projects use geospatial data (850 of 1,217); counting PADs alone, 62% of data-using documents do (680 of 1,092). A PAD is the full appraisal; PIDs are short and often written at concept stage.

Stricter checks. Under stricter rules the share falls. For projects that use data it is 60% without uncertain labels and 51% when the quoted evidence must also match word for word; for documents, 46% without uncertain labels and 37% when the quoted evidence must also match word for word. The word-for-word rule is harsh, because quotes taken from tables often reorder the cells. A defensible statement is between 51% and 60% of projects.

Coverage and limits

2,261 of 2,261 projects had passages reviewed, through 4,494 of their 4,679 distinct PADs and PIDs. 1,932 of the 1,932 data-using projects were approved in 2020 or later, so this describes recent practice and does not show a trend. Sector and region come from the World Bank Projects API; sectors and regions with fewer than 40 projects or documents are not shown. Kinds of data are grouped from source names; their "tied to a place" share counts individual uses. Review sets used: PAD/PID review — original sample (5,946 passages); PAD/PID review — 500-passage addition (500 passages). A PAD/PID counts only when the World Bank documents catalogue confirms it: the document exists, has the stated type and belongs to the stated project. The same document arriving from two collections is counted once, by its World Bank document id.