AI is changing the way tenders are written but not how they are assessed in Ireland. That gap is becoming a problem, says BidReview founder Tony Corrigan.
Until recently, the biggest challenge in public procurement was getting SMEs to compete at all. The submission process was abusive, the success rate was low, and many business owners looked at one process and decided their time was better spent elsewhere. I spent decades trying to simplify the workflow, but nothing changed.
Then came AI, and everything changed.
In 2024 and 2025, companies that had never tendered before began to provide ChatGPT with sales collateral and forwarded the results to the post. Today, an applicant can choose from a number of AI-powered proposal platforms that will generate a comprehensive response to the barest archetype of input.
I test tenders for a living, and from the buy side of the table, volume has exploded. Tournaments that attracted three or four bids in 2023 are now attracting twelve or fifteen. In one recent experiment I participated in, about 30pc of the content posted was completely AI-generated, another 40pc was mostly AI-generated, and only the last 30pc was meaningfully written by a human.
Consumer concerns
Consumers have started to take notice. They are worried about two things.
The first is what happens to their RFP documents once the supplier has fed them a multi-tenant AI model they don’t control. Clauses limiting the use of AI in proposal preparation are appearing in more RFPs. In theory, providers are not penalized for reporting it. Actually, the AI-generated copy has a fingerprint, and I wouldn’t bet that some testers would be more skeptical about the claims in the submissions that obviously ChatGPT’d the end so that they don’t even have to go into the proposal testing.
The second and more serious concern is that AI postulates are never true. The polite term is hallucination. In the context of shopping, that term is too soft. What’s really happening is that a supplier with real gaps in experience, capacity or resource has the gaps covered by a model that’s been trained to produce a realistically sound response. The bid reads well. The business behind it may not be able to deliver. In the best case scenario, the buyer spends time evaluating a proposal that has never been raced. In the worst case scenario, they give a supplier who can’t fulfill the contract.
The supply side of the market has changed. The testing side has not gone away. All the tenders I have reviewed this year have been read, scored and debated by people, sitting in a room or on the phone, working line by line. There are no widely accepted AI testing tools, and I don’t expect them anytime soon. If the buyer gives his decision to the model and the missing supplier finds out, the failure can be very serious. AI’s well-known tendency to produce subjective responses is a feature when writing sales copy and a fatal bug when deciding who will get a €250,000 contract.
So, now we have a market where it takes no time at all to generate a bid that looks credible, and as long as you have done a proper evaluation of one. The volume is quadrupled. Indeed, there is a tendency among inspectors to issue bids on the basis of eligibility criteria and prior contracts. And the buyer’s best defense against the flood of common, sound, changeable submissions is one thing that’s been a feature of public procurement for as long as I’ve worked: falling back on suppliers they already know.
Previous delivery success
Fewer than 3,000 businesses have recorded public sector contract wins on the island in the past two years. That’s a fraction of the companies that are completely capable of delivering. Evaluation panels score points on evidence of past delivery, and past delivery almost always means someone else’s past delivery. The result, over time, is the concentration of the supply chain: buyers end up relying on a smaller pool of older suppliers, changing pricing power, less flexibility, and when disruptions strike, there are no well-developed relationships to fall back on.
The advent of AI did not break this pattern. It strengthened you. If the examiner is faced with fifteen proposals, several of which are produced by a visible template and are not visible to each other, the logical answer is to weight the most well-known providers even more.
None of this is an argument against AI in tendering. The first draft of a modern proposal should probably be written with the help of AI; anyone competing without it is really in trouble. But the quality gap between “the AI wrote a comprehensive answer” and “this is the bid we’re really going to win” is a whole game, and nothing in the current toolkit is closing it.
In a €22bn market on the island, where one in four competitions still attract one or no bids, the problem is not that we need more proposals. We need a way to tell which ones we have are good. Until the testing side reaches the writing side, the asymmetry will continue to widen, and companies that are already profitable will continue to widen it.
Tony Corrigan is the founder of BidReview.ai, an AI-powered platform that automates the bidding and review process. He previously founded TenderScout, having started his career at IBM.
Don’t miss out on the information you need to succeed. Sign up for Daily BriefSilicon Republic’s digest of must-know sci-tech news.
