Procurement, Corruption, and the Case for AI Watchdogs in Africa

Award data, even in its current messy state, carries strong integrity signals. Threshold clustering, an unusual density of contracts priced just below the value that would trigger open tender, is trivially computable and consistently informative. Supplier network analysis surfaces companies sharing directors, addresses, or phone numbers while bidding against each other. Timing analysis flags entities incorporated shortly before an award they subsequently win.

None of these prove wrongdoing. Each of them raises the prior probability enough to justify a human audit, which is the correct and only defensible use. The value is triage: turning ten thousand awards into a prioritised list of fifty that a small integrity unit can actually examine.

Most African procurement systems publish some data, but rarely in a form that supports analysis. Supplier names vary across records. Amounts appear in inconsistent currencies and formats. Awards are published as scanned PDFs. Entity resolution, deciding that four differently spelled names are the same company, is where the majority of the engineering effort goes and where most civic tech projects underestimate the work by an order of magnitude.

This is unglamorous and decisive. An integrity model built on unresolved entities produces false positives that discredit the entire programme on first contact with a genuine dispute.

Any system that flags a company or an officer will be challenged, and it should be. That means every flag must be explainable in plain language, traceable to specific records, and reversible. Black-box scoring in this domain is indefensible and will be defeated the first time it reaches a tribunal.

It also means governance about who can see what. A public transparency dashboard and an internal investigative tool are different products with different risks. Publishing raw anomaly scores against named suppliers invites defamation exposure and, worse, gives sophisticated actors a free map of what the detector notices.

The instinct is to place these tools with anti-corruption agencies. In practice the more durable home is often the procurement authority itself, where the incentive is process quality rather than prosecution, and where flags can trigger a pre-award check instead of a post-award scandal. Prevention is cheaper politically and far cheaper fiscally.

Civil society has a distinct and complementary role: independent analysis of published data, which keeps the official system honest. Both need the same underlying cleanup, resolved entities, structured awards, consistent identifiers, which is the piece worth funding regardless of who ends up operating the detector.