Check every tender while it is open, not after the contract is signed.

Weak competition is spotted too late

World Bank economists estimated public procurement at about $11 trillion in 2018, out of global GDP of nearly $90 trillion. Health is the largest category, at about a third of OECD procurement spending.

In the EU, competition is thinning. The European Court of Auditors found that 42% of EU contracts in 2021 were awarded when only one company bid, and that the number of bidders per procedure fell from roughly six to three over the decade to 2021. An audit that finds a single-bid contract after award comes too late to change its price.

What we would model, and with which data

Many governments already publish procurement data, so a first model can start without a data-sharing agreement.

• Contracting data in the Open Contracting Data Standard, or OCDS, a common data model for every stage of a contract. The Open Contracting Partnership's registry lists data from over 100 publishers.
• Notice portals, such as Tenders Electronic Daily, the EU's supplement to its Official Journal for public tenders.
• Tender documents, bids and contracts, which hold terms that structured data misses.
• Company registers, to link suppliers that share owners or addresses.
• The World Bank's Global Public Procurement Database, which compares procurement laws, e-procurement systems and procurement cycle metrics by country.

How the model is built and checked

A red flag is a published risk indicator, such as a tender that drew one bid or a bidding window far shorter than usual, and the model computes each flag for every tender. Price benchmarks compare unit prices for the same item across buyers and over time. Document reading converts tender PDFs to text, and a language model compares eligibility terms with similar tenders to find unusually narrow ones. Record linkage connects bidders that share a director or an address. Auditors review a sample of flagged and unflagged tenders, and we report both the share of flags they confirm and the cases the model missed.

What decision-makers get

The most valuable flag is the one raised before award.

• Procurement teams see risk on open tenders while there is time to act.
• Audit offices get a ranked list of past contracts to examine, each with its evidence.
• Finance ministries get price benchmarks and an estimate of what more competition would save.
• Sustainability teams can track green purchasing. The OECD found that 35 of 38 member countries have green public procurement frameworks, but only 3 have methods to measure environmental impact.

A first project picks one buyer or one category, such as medical supplies. We would load the published data, compute flags and benchmarks, back-test them against past audit findings and hand over a scorecard. The buyer then decides whether to continue.

How it runs safely

A flag starts a review. It never cancels a tender or accuses a firm.

• An officer or auditor reviews each flag, and a named official decides any action.
• Commercially sensitive bid data stays in the buyer's systems.
• Fairness checks watch whether flags fall more often on small or new suppliers.
• We log every flag with its inputs and indicator version, so an auditor can trace it.
• Procurement law differs by country, and this page is not legal advice.

Open-source building blocks

Buyers, auditors and suppliers can rerun the same open checks. The Open Contracting Partnership maintains the first four tools.

• Kingfisher Collect, which downloads OCDS data from publishers.
• OCDS Kit, command-line tools for working with OCDS data.
• Cardinal, which calculates red flags such as single bid received and short submission period.
• lib-cove-ocds, which checks OCDS data against the standard.
• Docling, started at IBM Research Zurich, which converts PDFs and other documents into structured text.
• Splink, for linking supplier records across registers.