MMOOZZGG.com

AI against corruption

Open data in public procurement: how transparency works against corruption

7 min read3 sources

Open data in public procurement lets you check every tender, not a sample: if notices, records, and contracts are published in machine-readable form, a program, a journalist, and an audit body can all read them. The World Bank platform ProACT gathered such data for 110 countries for 2006–2020 and calculated risk indicators from it.

On this page · 6 sections

In short

Open data in public procurement lets a machine check every procedure and show a person only the risky ones. ProACT holds data from 110 countries, and the OCP method has 73 red flags calculated from such data. A flag means risk, not proof of a violation.

Why openness lowers risk

Openness changes behavior before any check: the buyer knows that any of its tenders can be recalculated. The data becomes shared by the audit body, market players, researchers, and journalists, and one procedure comes under several eyes at once.

An example from an Open Contracting Partnership (OCP) report: the procurement agency of the Dominican Republic tracks all tenders in real time by 21 red flags. Together with other anti-corruption measures, this brought clear changes: the share of open procedures grew from 94% in 2020 to 96%, and the share of tenders with a single bid fell from 63% to 60% in 2023.

In the same country in 2023, more than 60 suppliers were barred from tenders for violations, and there were fewer unresolved complaints and canceled tenders. According to OCP, trust in procurement grew: more than 20,000 new suppliers registered.

The Brazilian robot ALICE rests on the same basis. Since 2015, it has collected more than 500 procurement notices every day and checked about 40 types of risk. Without electronic publication of notices, the robot would have nothing to collect.

What data the analysis needs

The minimum set of fields is visible in the ProACT transparency indicators. The platform checks whether the buyer, the contract title, the supplier, the supply type, the contract price, the location of the buyer or of delivery, the procurement method, the number of bids, and the award date are published.

Red flags need more: the dates of the notice and of bid submission, all bidders and their prices, the reasons for rejecting bids, contract changes, acceptance, and payments. OCP recommends publishing them in the Open Contracting Data Standard (OCDS), where each stage of procurement has its own fields.

Data on acceptance and payments is needed separately. Without it, the signs of the delivery stage are not visible: contract price growth, delivery delays, payments above the contract amount. This is the stage where ProACT counts delays and price growth.

The analysis gets stronger when procurement is linked to other sources: a company register, financial statements, and ownership data. OCP directly advises adding beneficial ownership data to procurement publications.

Each flag needs its own fields. The “single bid” flag needs the procurement method and the number of bids, and the “shared owner among bidders” flag needs an ownership register. If a field is missing, the flag cannot be calculated, and this is visible before the project starts. The full list of signs is covered on the red flags page.

ProACT: 110 countries in one database

ProACT is a World Bank platform with open data on public procurement in 110 countries for 2006–2020. The database rests on national data from 45 countries and on data about contracts financed by the World Bank.

Risk indicators in ProACT: non-competitive procurement methods, unpublished notices, contracts with a single bidder, deadlines for bids and decisions, a price check by Benford’s law, the supplier’s share of the buyer’s spending, supplier registration offshore, delivery delays, and contract price growth.

Benford’s law here checks prices. In large natural sets of numbers, the first digits follow a known distribution, and adjusted prices often deviate from it. A deviation does not prove manipulation, but it shows where to look.

The platform is built for staff of procurement agencies, civil society organizations, corruption researchers, and international organizations. Shared indicators let you compare countries and buyers by one method.

How different countries use such data in oversight is collected in the review of world experience.

Data quality: gaps and errors

An analysis is only as good as the data is complete. OCP gives an example: a field may exist but be filled in too rarely; say, bidder data exists for only 10% of open tenders. With such coverage, a flag about bidders will show nothing.

Typical problems of open procurement data: empty fields, different spellings of one buyer or supplier, prices in different units, contracts with no link to a notice, and contract changes that are not published.

A standard helps find them. According to OCP, once data is converted to OCDS, errors can be compared and fixed across datasets from different countries and regions, and OCP has its own tools to check the quality of such publications.

AI is useful here too. A language model matches records where one company is written in different ways and fills empty fields from attached documents. Each such fill is marked, so a person knows that the field was restored, not published by the buyer.

A quality check is the first step of any project. A report on field completeness for each year and each buyer is quick to prepare, but it shows at once which flags can be calculated now and which cannot yet.

Bad data gives false flags. If a sign fires on almost every procedure, check the data and the formula first, and only then the tenders.

Public oversight and AI

Open data works not only for the audit body. Journalists, civil society organizations, and the bidders themselves see the same procedures and report a problem before a scheduled audit reaches it.

In Indonesia, the organization Indonesia Corruption Watch uses procurement data and red flags so that public monitors can choose procedures to check and prepare well-founded complaints. According to OCP, the share of rejected complaints fell from 78% in 2021 to 39% in 2022, and complaints began to be reviewed 76 days faster on average.

According to OCP, the same tool, opentender.net, gives auditors a more objective way to choose procedures for audit than the discretion of local authorities.

AI lowers the entry barrier: a monitor does not need to write database queries to find single-bid tenders at their buyer. But a public monitor’s flag remains the same sign of risk as an audit body’s flag, and it does not become proof.

For a government body, public oversight takes off part of the load. A well-founded complaint with data and links to documents saves the auditor time on collecting materials, and flags help separate such complaints from the rest of the flow.

Openness often scares government bodies themselves: everyone will see the data, and the errors too. How to answer this fear calmly is covered on the page about why people fear AI.

What a buyer should do

A buyer does not need to wait for a general reform to make its procurement verifiable. Five steps are enough:

1. Publish all stages of procurement: the plan, the notice, the bids with prices, the record, the contract, the changes, acceptance, and payment.

2. Use common identifiers for the buyer, the supplier, and the tender, so that records link to each other.

3. Publish data in machine-readable form, not only as scans of documents.

4. Check regularly that the fields needed for red flags are filled in, and fix the gaps.

5. Calculate the flags on your own data first: this way, errors in the data are found before others find them.

You can start small: export last year’s tenders and calculate three or four signs from the ProACT list, such as the share of single-bidder tenders and contract price growth. Even this slice will show where the data is incomplete and which procedures to open first.

Openness also protects the buyer itself. When each tender is visible and recalculated, an honest employee can more easily show that a decision was justified. By the OCP method, a flag means risk, not proof of a violation, and the auditor makes the decision on it.

Sources and fact-checking

Checked by the editors on 16 October 2026.

  1. World Bank Blogs, 30 June 2021 (blogs.worldbank.org)
  2. OCP, Red flags in public procurement, 2024, 2024 (open-contracting.org)
  3. OECD OPSI, Robot Alice, 24 June 2024 (oecd-opsi.org)

Found a mistake? Tell us

Publications

For government bodies

We build such systems for government bodies and state companies, from data analysis to a pilot on one scheme. The data stays on the organization's own servers.

Contact us