Nepal open data: what's actually usable for business

Some of Nepal's open data is good enough to run a business decision on, but less than you'd hope. We checked 11 public sources in September 2026. Nepal Rastra Bank's monthly Excel files, its exchange-rate API, the World Bank API and OpenStreetMap are fresh and machine-readable. Labour data is from 2017/18, trade data sits behind an expired certificate, and data.gov.np doesn't resolve.

That's the short version. The longer one matters if you're sizing a market, picking a shop location, setting prices or forecasting demand, because the gap between "the data exists" and "the data is usable" is where most Nepali analytics projects lose their first month.

How did we score each source?

We used a simple rubric with four tests, each scored 0 to 2, for a total out of 8. It favours what an analyst actually hits on day one, not what a portal promises on its homepage.

• Format: 0 if PDF only, 1 if Excel or mixed, 2 if CSV, JSON or a working API.
• Freshness: 0 if the latest period is more than two years old, 1 if within about a year, 2 if within two months.
• Access: 0 if the site was down or the domain didn't resolve, 1 if it loaded with problems (certificate errors, listings that stop years ago), 2 if it just worked.
• Licence: 0 if we found no stated reuse terms on the pages we checked, 1 if terms vary by dataset, 2 if there's a clear open licence.

Everything below is as checked in September 2026. We only list sources we actually opened. NEPSE market data and the Economic Census 2018 are left out because we didn't verify them this round, and the dedicated 2021 census results site didn't load for us, so we can't score it fairly.

What's genuinely usable right now?

Nepal Rastra Bank is the standout. Its Monthly Statistics for mid-August 2026 went up on 18 September 2026, in both PDF and Excel, which is roughly a one-month lag. That's good by any regional standard. The Excel file matters more than it sounds: you can pull a remittance or private-sector credit series straight into a model without retyping anything.

NRB's exchange-rate API is even easier. It returns clean JSON by date range, and when we called it, it gave rates for 20 September 2026 with the US dollar at Rs 153.10 buying. If you price imported goods, that feed belongs in your dashboard. We used it the same way in our Power BI vs Data Studio vs Metabase comparison.

The international sources fill gaps well. The World Bank API had Nepal's 2025 GDP at about $45.5 billion, last updated in July 2026, under a CC BY licence. The Humanitarian Data Exchange lists 360 Nepal datasets, many at district level with administrative boundaries attached. OpenStreetMap is rebuilt daily and is the best free source for where roads, banks and competing shops actually are.

Where does it break down?

At the national portal, first. data.gov.np returned a non-existent-domain error from Google's public DNS. The NSO's own data portal refused connections on two separate tries, while its main site stayed up. If you were planning to build on a government open data catalogue, there isn't a working one to build on.

Labour data is the biggest content gap. The last full Labour Force Survey is from 2017/18, when unemployment was 11.4 percent. The fourth round started in March 2026 and runs until March 2027, with results expected in August 2027, according to the Kathmandu Post. So any wage or hiring analysis today leans on data that's eight years old, or on the 2022/23 living standards survey.

Then there's the PDF problem. The Economic Survey 2082/83, released in May 2026, is a PDF. NTA's monthly telecom reports are PDFs, and the MIS listing page we loaded only showed years back to 2080. Trade statistics are the most frustrating: TEPC had published 11 months of FY 2082/83, but its site served an expired security certificate when we checked, and the annual book we found on the Customs site was for 2079/80.

Which sources fit which business decision?

• Market sizing: start top-down with World Bank population and income series, then split by district using HDX population tables. Cross-check with NRB remittance figures if your buyers depend on remittance income.
• Site selection: OpenStreetMap for competitor and footfall-proxy locations (banks, schools, bus parks), plus HDX boundaries to draw catchments. Check OSM coverage on the ground; it's thinner outside the big cities.
• Pricing: the NRB forex API for landed cost, and the inflation figures in NRB's monthly releases. Pass exchange-rate moves into your price list on a rule, not a gut feeling.
• Demand forecasting: NRB monthly series for credit, deposits and remittance as leading indicators. They're monthly and about a month behind, which is enough for a quarterly forecast.

None of this replaces your own sales data. Public data tells you the size of the pond; your POS and accounting systems tell you how many fish you're catching. The value comes from putting the two side by side, which is the point of a business dashboard that pays for itself.

How do you turn PDFs and Excel files into a usable dataset?

The work is less glamorous than the dashboard, and it's most of the job. A working pipeline has five stages. Each exists because of a specific Nepali data problem, not because it looks tidy on a diagram.

• Extract: pull JSON from APIs on a schedule, read Excel sheets directly, and run table extraction on PDFs. PDF extraction always needs a human spot-check, because merged header cells and Devanagari numerals trip up every tool.
• Clean: convert Nepali digits to Arabic ones, strip thousand separators in the lakh/crore style (12,34,567), and check the unit line on every sheet, because tables in the same release can use different scales.
• Reconcile dates: map Bikram Sambat months and fiscal years to Gregorian dates with a lookup table. The fiscal year runs Shrawan to Asar, roughly mid-July to mid-July, and BS months run 29 to 32 days, so there is no formula that works every year.
• Model: load everything into one warehouse with a shared date table and shared district codes, so a remittance series and a store's sales can sit on the same axis.
• Visualise: publish to a dashboard with the source and 'as of' date printed on every chart, so nobody mistakes 2018 labour data for last month's.

The date table deserves a special mention. NRB labels its latest release "2083-04 (Mid Aug, 2026)", meaning the fourth BS month of 2083, Shrawan, which ends in mid-August. Get that mapping wrong by a fortnight and your seasonal demand curve shifts with it. We build the same kind of table for government service data in our e-government citizen services assistant blueprint.

What should you do with this?

• Automate the two NRB feeds first. They're fresh, structured and cover pricing and demand.
• Treat any PDF source as a monthly manual job until it's proven stable, and keep the original file.
• Print the 'as of' date on every chart. It's the cheapest protection against deciding on stale numbers.
• Don't plan around data.gov.np or the NSO data portal coming back. If they do, it's a bonus.

Public data in Nepal is patchy, but it isn't useless. A handful of sources carry most of the value, and the rest can be made usable with a few days of careful extraction work.

If you want public figures sitting next to your own sales on one screen, our data visualisation team builds these pipelines, date tables included. Tell us what decision you're trying to make and we'll tell you which sources can actually support it.