Where overpricing happens most often
An inflated price enters a procurement at one of three stages. At planning, the buyer sets a start price above the market. At the tender, the winner offers a price with almost no cut from the estimate. After signing, the price rises through contract amendments.
The Open Contracting Partnership describes 73 red flags across all stages of procurement, and several of them are directly about price: an item price far above the average for its category, a winning price too close to the estimate or above it, and growth of the contract amount after the winner is chosen.
The risk is higher where competition is low. World Bank indicators place single-bidder purchases and one supplier’s share of a buyer’s purchases next to contract price growth: if nobody can bring the price down at the tender, the inflated start price survives into the contract.
The data you need: purchase items with a classifier code, quantity and unit price; the estimated price and its justification; the prices of all bids; the amount when the winner was chosen and the final contract amount; and all contract amendments with their reasons.
How AI builds a market price
For AI, a market price is not one number but a distribution. The system collects unit prices for similar items: the same classifier code, close characteristics, a comparable volume, the same period and region. The result is a corridor where prices usually lie.
Then each new purchase is placed on this distribution. OCP suggests flagging an item if its price goes above the upper limit of the usual spread for the product category: such a price may mean that a bidder inflates it on purpose. The same method works for the ratio of the winning price to the estimated price.
The main difficulty here is not the calculation but comparability. Item names are written in different ways, units of measure get mixed up, and goods and delivery are put in one line. A language model reads the descriptions and the technical specification, brings the items to a common form and separates services from goods, and only then does the system compare prices.
A made-up example: Buyer B purchases laptops, and the unit price is a third higher than in similar contracts of the same quarter. The system flags the item and shows the reviewer the ten closest contracts for comparison. If the technical specification has characteristics the similar items lack, the difference may be justified, and the flag is removed.
AI also looks at the winning price separately. Competition usually brings the price below the estimate, so, as OCP describes it, a winning bid too close to the estimated price or above it may point to a leak of budget information.
The reviewer sees not only the flag but also its basis: the item, its price, the corridor of similar prices and the list of contracts the corridor is built on. A flag without an explanation cannot be checked, while a flag with a list of similar items can be checked against documents and the market.
Price growth after signing
Some overpricing appears not at the tender but later. A contract is won at a low price, and then the amount grows through additional agreements. OCP says directly that amendments can be lawful, but they can be used to raise the price without new competition.
The signs in the data are these. The final contract amount differs a lot from the amount when the winner was chosen; the user sets the deviation threshold, for example 30%. The contract is amended with a price increase. After a direct purchase, amendments push the amount above the threshold that requires a competitive method.
A similar indicator, the difference between the amount when the winner was chosen and the final contract amount, is calculated on the red flag dashboard of ANAC, Italy’s anti-corruption authority. In World Bank indicators, contract price growth sits next to performance delays.
For AI this is a task of linking documents. It must match the contract, all its agreements, the acceptance acts and the payments, and then show the reviewer the chain: what it was, what it became and what justifies each change. If the work is paid for but there is no result, it is already a matter of performance control.
Then a person takes over: checks the justification of each change, compares the acts with the actual volume and, if needed, asks the buyer for calculations. If the changes are lawful, the flag is closed with a note, and the note stays in the purchase history so the same contract does not come up again.
The experience of Kazakhstan
Kazakhstan shows that price monitoring gives a measurable result. According to its Anti-Corruption Agency, monitoring of inflated prices in the Single Procurement Window prevented wasteful spending of more than KZT 67 billion over two years: KZT 29 billion in 2022 and KZT 38.4 billion in 10 months of 2023.
It matters that this is prevented spending. The signal about an inflated price comes before payment, and the buyer has time to revise the purchase instead of recovering money afterwards.
The second part of the experience concerns suppliers. Kazakhstan’s Ministry of Finance selects them with an AI rating system that uses 28 databases. A supplier’s price and reliability are checked together: a low price from a supplier with a poor record is also a risk.
How control systems work in Russia and Kazakhstan is covered on the page about Russia and the CIS.
World Bank indicators
On the ProACT platform the World Bank collected open procurement data from 110 countries for 2006–2020 and calculated risk indicators from it. They include a single bidder, bid submission periods, deviation of prices from Benford’s law, a supplier’s share of a buyer’s purchases, offshore registration, delays and contract price growth.
Benford’s law describes how often first digits appear in natural sets of numbers. If bid prices in one market are distributed differently, it is a reason to check whether they were adjusted by hand. Such a shift alone proves nothing.
For a buyer, the value of ProACT is that the indicators are calculated on open data and in the same way for all countries. You can apply the same logic to your own purchases: take the fields already in the register and calculate the risk for each procedure from them.
The indicators add to each other. A high price with a single bidder and later growth of the contract amount weighs more than each sign alone, so the system adds them into an overall risk score for the procedure.
How to check your own purchase
You can start a price check with the data the buyer already has.
1. Export the items of your contracts for two or three years with the classifier code, quantity and unit price.
2. Find similar items in open procurement data for the same period and region.
3. For each item, calculate where its price stands in the distribution, and flag those outside the usual spread.
4. Compare the amount when the winner was chosen with the final contract amount, and list all agreements that raised the price.
5. For the flagged items, pull the justification of the start price and the commercial offers it refers to.
A flag means risk, not proof. A high price can have honest reasons: urgency, special characteristics, delivery to a remote place. The reviewer decides on a violation from the documents, and the system only shows where to look first. How to set up such flags and thresholds is described on the page about red flags.