When every bid reads perfectly, you need a different way to find the right supplier

When every bid reads perfectly, you need a different way to find the right supplier

Something has shifted in the past 18 months in the quality of tender responses coming across my desk, and colleagues across the public sector are telling me the same thing. Bids are more polished than they used to be. They’re better structured, more closely aligned to evaluation criteria, and written with a fluency that wasn’t there a few years ago. On the surface, that sounds like a good thing. In practice, it’s creating a problem that most procurement teams haven’t fully grappled with yet.

The reason for the shift is obvious: suppliers are using AI to write bids. For many firms, particularly smaller ones without dedicated bid teams, AI has been genuinely valuable. It’s helped them produce submissions that compete on a more equal footing with larger organisations, and that’s good for competition and good for the public sector buyers who benefit from a broader supply base. I don’t think anyone should be criticised for using AI in bid writing. PPN 017, the government’s procurement policy note on AI transparency, is clear that AI use in tenders isn’t prohibited, and the Scottish Government’s own AI guidance on Public Contracts Scotland takes the same position.

But here’s the consequence nobody planned for. When AI raises the floor on bid quality across the board, the written submission alone stops being a reliable way to distinguish between suppliers. And for public sector procurement teams in local government, education, healthcare and blue light services, where getting the right supplier is often the difference between a service that works and one that fails, that’s a problem worth taking seriously.

The evaluation gap

I’ve been working in public procurement long enough to remember when the quality of a tender response genuinely told you something about the firm behind it. A well-written, thoughtful bid usually came from an organisation that had invested time in understanding your requirements and had the people and systems to back up what they were saying. A patchy, generic response usually told you something too.

That signal has weakened considerably. AI can take a mediocre understanding of a client’s requirements and produce a submission that reads as though the firm has thought deeply about every aspect of the specification. It can generate case studies that sound specific but are actually assembled from generic patterns. It can align responses to evaluation criteria with a precision that human writers rarely achieved, because it’s processing the entire tender document at once rather than working through it section by section.

The result is that evaluation panels are sitting in front of a stack of responses that all read well, all hit the criteria, and all sound like they come from capable, committed organisations. The differentiation that should emerge through quality scoring gets compressed, and competitions start drifting towards outcomes determined almost entirely by price. That’s a concern across every part of the public sector, because price-only outcomes rarely deliver the service quality that the people relying on those services actually need.

Why reading harder isn’t the answer

The instinct when faced with polished AI-generated bids is to scrutinise the written submissions more closely. Look for tells. Spot the generic language. Try to separate the bids that were written by people who genuinely understand the work from the ones that were assembled by a machine.

I don’t think that approach works, for a couple of reasons. AI-generated text is getting harder to distinguish from human-written text with every model update. Even experienced evaluators can’t reliably tell the difference, and asking them to try adds time and subjectivity to a process that’s supposed to be objective. More importantly, it misses the point. The question isn’t whether a supplier used AI to write their bid. The question is whether they can actually deliver what the bid promises. Those are different things, and the second question is the one your evaluation needs to answer.

What needs to change in practice

If the written submission alone can no longer carry the weight it used to, procurement teams need to put more emphasis on the parts of the evaluation process that AI can’t fake.

Clarification interviews. Asking shortlisted suppliers to talk through their responses in person, or by video, is one of the most effective ways to test whether the people behind the bid actually understand what they’ve committed to. A well-written method statement might describe a brilliant mobilisation plan, but if the project manager who’ll be running it can’t explain how it works without reading from notes, that tells you something the written submission never would. The Procurement Act’s competitive flexible procedure makes this easier than it used to be, and Scotland’s procurement framework has always allowed for dialogue at appropriate points.

Reference and track record checks. AI can write a convincing case study. It can’t invent a track record. Going deeper on references, not just collecting them but actually speaking to previous clients, asking specific questions about delivery quality, responsiveness and problem-solving, gives you evidence that no amount of bid-writing polish can manufacture. For contracts where the stakes are high, a school’s catering service, a hospital’s facilities management, a fire service’s fleet maintenance, this due diligence is worth the time it takes.

Site visits and capability assessments. Where practical, visiting a supplier’s premises or an active project site gives evaluation panels information they can’t get any other way. You can see the systems they actually use, meet the people who’ll actually be doing the work, and get a sense of whether the organisation’s culture and capability match what the bid describes. This isn’t possible for every procurement, but for larger or higher-risk contracts it should be standard practice.

Probing questions that go beyond the tender document. Rather than asking suppliers to repeat what they’ve already written, evaluation panels should ask questions that require suppliers to think on their feet. How would you handle a specific scenario? What would you do if demand exceeded what you’ve planned for? Walk us through a time when something went wrong on a similar contract and how you dealt with it. These questions test understanding, experience and honesty in ways that a pre-prepared written response, whether human or AI-generated, simply can’t.

The supplier conversation matters too

This isn’t only about what happens on the buyer’s side. The sector needs a more open dialogue with suppliers about AI use in bidding. Tender documents should include clear wording reminding firms that they’re contractually required to deliver any commitments made in their responses, whether those commitments were drafted by a person or a machine. That’s not about discouraging AI use. It’s about making sure everyone understands that a well-written bid creates a real obligation.

PPN 017’s disclosure questions are a reasonable starting point. They ask suppliers to confirm whether AI was used in preparing their response, and that disclosure is treated as information rather than scored. For contracts procured under Scotland’s procurement legislation, similar transparency principles apply. But disclosure alone doesn’t solve the evaluation challenge. Knowing that a supplier used AI to write their bid doesn’t tell you whether they can deliver the service. The evaluation methods I’ve described above are what actually answers that question.

Getting ahead of this rather than reacting to it

The public sector organisations that handle this transition well will be the ones that recognise what’s changed and adapt their evaluation approach before it causes problems, rather than after a poorly performing supplier has already been appointed on the strength of a bid they couldn’t live up to.

That means building clarification interviews into your procurement timelines as standard. It means budgeting time for meaningful reference checks and site visits on higher-value contracts. It means training evaluation panels to ask probing questions rather than relying solely on scoring written responses. And it means having an honest conversation within your organisation about how AI is changing the dynamics of your procurement process, so your approach evolves with the market rather than falling behind it.

AI has made it easier for suppliers to write good bids. That’s not a bad thing. But it means procurement teams need to get better at looking beyond the words on the page to find out what’s really there.

Kathryn Irons is Head of Key Accounts at Inprova Group

This article draws on themes from an opinion piece published in Housing Digital.

How Inprova can help

Quantum, our spend and contract management platform, lets you track supplier performance against what was committed at tender stage, giving you the data to hold suppliers to account on what they promised, not just what they wrote.

Procurement consultancy including support with evaluation framework design, clarification interview processes and supplier assessment, making sure the decisions that matter are grounded in evidence rather than bid-writing quality.

Frameworks, DPS and Dynamic Markets where suppliers have already been through our independent vetting and due diligence, reducing the risk that a polished bid doesn’t reflect genuine delivery capability. We work with customers across local government, education, healthcare and blue light services throughout the UK, including Scotland.

---

---

Sources:

Cabinet Office, PPN 017: Improving transparency of AI use in procurement, February 2025

Public Contracts Scotland, Guidance on the use of Artificial Intelligence (AI), September 2024

Trowers & Hamlins, The AI advantage: artificial intelligence and public procurement, February 2025