AI in the Bidding Process

HM Government is putting a lot of effort into trying to anticipate the impact of AI on the economy generally, on specific sectors within the economy and on public services.

They have already made available guidance and policy documents via organisations such as the Cabinet Office, the Office for AI and the Government Digital Service. These documents offer insights into how various organisations might sensibly use AI in projects without building in unnecessary bias and other risks.

Now the Government has put its mind to the expectation that companies who supply goods and services to the public sector might well start to use (or significantly increase their use of) AI in the creation of their bid responses.

With that in mind they have published a new Procurement Policy Note – Improving Transparency of AI use in Procurement (March 2024) (PPN02/24).

This is aimed only at Central Government Departments and agencies, not the NHS or local authorities, although they can be expected to take a similar approach shortly. The PPN specifically states that “other public sector contracting authorities may wish to apply the approach set out in this PPN”.

Companies that regularly bid for government contracts will want to familiarise themselves with the PPN and consider how they will answer the additional disclosure questions that the PPN suggests might sensibly be added to the ITT. Example questions appear in the PPN and are designed to flush out how far bidders have relied on AI to develop their responses to questions in the procurement documentation.

The key risks that the PPN is intended to address seems to be the risk that bidders who have relied on generative AI to produce their bid responses will not have checked that the responses accurately reflect what the bidder can actually deliver.  AI’s tendency to produce so-called ‘hallucinations’ is clearly (and rightly) of concern to those involved in running government procurements. Their response is to seek explicit confirmation from bidders that the response has been properly checked by human beings and does not contain misleading claims. Two sample questions are included that ask the bidder to verify that their AI-generated responses have been checked and verified for accuracy.

Arguably, most tender processes will already include a section requiring each bidder to confirm that its bid is accurate. It’s not entirely clear whether the new sample questions will add a great deal except as a timely reminder to the bidders to check the AI’s homework.

Interestingly, although the PPN is almost entirely focussed on the use of AI in generating bid responses (which might include purely procedural aspects), one sample question addresses the use of AI in the offered goods and services themselves. It invites bidders to “describe how AI technologies are integrated into your service offerings…”.

Another risk addressed by the PPN (almost in passing and with no suggested wording to help departments to manage it) is the risk that bidders might use ‘confidential contracting authority information…as training data for AI systems…”. This is a real risk, and should therefore be addressed in all public procurements as part of a much wider set of measures, but this PPN could arguably go further to help authorities with their mitigation measures by, for example, suggesting additional provisions on confidentiality and intellectual property rights. We will wait to see what else emerges and how well the PPN fits into the expanding suite of guidance and policy documents on AI.

Tim Heywood, is a partner at gunnercooke llp specialising in public procurement, public law, technology and regulatory matters. He was formerly head of the procurement and commercial law team at the Treasury Solicitors Department and supports bidders and purchasers on complex high value procurement.

This article is not offered as legal advice and must not be relied on as such. Specific legal advice should be sought in relation to your particular circumstances.