Bid Rigging in Procurement: What AI Can See in Losing Bids

A tender closes with four bids. One supplier wins, the price is within budget and the file looks competitive, so attention moves to award and implementation. From the perspective of that single procurement, there may be nothing obviously wrong.

That is precisely why bid rigging is difficult to detect. Suppliers can coordinate who wins, submit deliberately uncompetitive cover bids, stay out of selected tenders or rotate opportunities while preserving the appearance of competition. The buyer sees one event; the suspicious pattern may only exist across many tenders, suppliers and contracting authorities.

On 8 September 2026, the UK's Competition and Markets Authority argued for a much more data-driven response. Its paper puts striking numbers around the problem: UK public procurement is roughly £400 billion a year, bid rigging can inflate prices by 20 percent or more, and the CMA estimates a possible annual impact of roughly £1 billion to more than £3 billion under its assumptions. The more interesting lesson for procurement, however, is what the CMA says about the data needed to find it.

The winning bid is only part of the evidence

Procurement naturally spends most of its time on the winning offer. Was the price competitive, did the bidder meet the specification, and can the award be defended? For cartel screening, the losing bids may be just as important because the signal is often relational.

Three suppliers may repeatedly appear together. One wins while the others submit higher prices, then the winner changes on the next tender. That pattern can be completely legitimate, perhaps because capacity, geography or cost exposure varies. It can also be worth examining if the rotations, pricing gaps or participation patterns recur in ways that are difficult to explain competitively.

No single tender necessarily proves anything. The information emerges when bid-level data is compared over time and across buyers, which is why the CMA stresses the importance of retaining losing-bid information in a consistent, machine-readable form.

AI changes the scale of the question

Historically, cartel investigations have often started with whistleblowers, complaints or evidence found once suspicion already existed. Data screening creates another route by asking whether suspicious patterns recur across a volume of procurement data that no individual buyer could realistically inspect.

The CMA's Bid Rigging Intelligence Tool, BRIT, combines cartel-enforcement expertise with data science and is being piloted with public bodies including the Departments for Education, Work and Pensions and Justice. According to the CMA, those pilots have begun to generate enforcement leads.

That is a meaningful use of AI in procurement because the technology is not trying to replace a buyer's judgment in one tender. It extends visibility across thousands of observations and highlights where specialist human attention may be justified.

A suspicious pattern is not evidence of guilt

The evidentiary boundary has to remain sharp. Similar bids, repeated pairings or rotating winners can have innocent explanations, and a screening model cannot establish collusion simply by flagging an anomaly.

BRIT is described as a tool for identifying suspicious patterns and generating leads. Investigation, evidence, legal assessment and due process come afterwards. Procurement teams need the same discipline: a strange pattern is a reason to preserve the record and use the correct legal or competition route, not a reason to confront the supplier across the table.

The CMA explicitly warns public buyers not to discuss suspected bid rigging with suppliers because tipping them off can make investigation harder. That boundary should be designed into any internal alerting process, not left to the intuition of an individual category manager.

The real bottleneck may be data rather than the model

The strongest part of the CMA argument is almost mundane: sophisticated analytics do not help if procurement does not retain the right data. Losing bids are not routinely collected centrally in the UK in a consistent machine-readable form, which limits screening across the public sector.

Enterprise procurement has the same structural problem. Tender histories are spread across sourcing tools, email, spreadsheets and local drives; supplier names are inconsistent; some systems retain final rankings while discarding the detail of unsuccessful bids. An organization can buy advanced analytics and still be unable to ask the interesting question because the historical evidence is fragmented.

That makes data architecture part of competition capability. The decision about what procurement retains today determines what patterns it can investigate tomorrow.

Cross-buyer visibility changes what becomes observable

A buyer may see that supplier A lost to supplier B and think nothing of it. Another authority sees A lose to C, and a third sees B lose to C. None has enough information alone to detect a broader pattern. A combined dataset may reveal repeated participation, winner rotation or correlated price behavior that warrants specialist review.

This is one reason centralized or interoperable procurement data has value beyond spend analysis. It creates questions that cannot be answered from one tender file at a time.

It also illustrates the wider judgment problem in procurement AI. As our article on AI and procurement judgment argues, a system can surface a pattern that deserves attention without deciding what the pattern means. The professional still has to understand the evidence boundary before acting.

Procurement capability still matters after detection improves

If screening becomes more automated, buyers still need to understand the signs of bid rigging, preserve records, keep the process competitive and know when to involve legal or competition specialists. They also need to avoid the opposite error: assuming that a clean dashboard proves the market is genuinely competitive.

Voice2Evolve is not a bid-rigging detection product, and this problem is primarily one of competition enforcement and data. The conversation link appears later, when a professional has to challenge unusual supplier behavior, escalate a concern internally or resist drawing conclusions beyond the evidence. Those moments still require disciplined language under pressure.

AI can make hidden patterns more visible. What it cannot do is turn a statistical signal into proof, or relieve procurement of the responsibility to handle the signal correctly.

The losing bids matter because collusion is designed to hide inside apparently normal competition. Sometimes the information procurement usually treats as unsuccessful is exactly the information that makes the market intelligible.

Sources

• UK Competition and Markets Authority (8 September 2026). Rigged bids, real costs: a case for urgent action on bid rigging in UK public procurement.
• Sarah Cardell, CMA (8 September 2026). Rigged bids, real costs: the case for tackling bid rigging in public procurement.