AI in Procurement: Transparency, Evaluation Risk and Officer Assurance

Why this matters now

Artificial intelligence is no longer a distant procurement issue. It is already present in bid writing, specification drafting, market research, evaluation preparation, data analysis, supplier products and service delivery models.

For councils, this creates opportunity and risk.

AI may help suppliers produce clearer tender responses, analyse requirements, improve service design and automate aspects of delivery. It may also create unsupported claims, generic answers, inaccurate content, confidentiality concerns, hidden subcontracting, data protection issues, bias risk and uncertainty about whether the bidder genuinely has the capability described in its submission.

The Cabinet Office PPN 017: Improving transparency of AI use in procurement provides optional questions and practical guidance to help identify AI use in procurements and in the delivery of government services. It applies to central government departments, executive agencies and non-departmental public bodies, but other public sector contracting authorities may wish to apply the approach.

This makes the PPN relevant for councils that need a proportionate way to handle AI without overreacting.

The aim should not be to ban AI. The aim should be to understand when AI use matters, where it creates risk and what evidence is needed to protect a fair, transparent and defensible procurement process.

AI use in bids is not automatically wrong

PPN 017 makes an important point: suppliers' use of AI to develop tender responses is not prohibited during the commercial process. This is sensible.

Suppliers have long used bid writers, templates, consultants, grammar tools, design support and internal knowledge libraries. AI tools are another form of assistance. A council should not reject a tender merely because AI helped draft text.

The real issue is whether the response is accurate, specific, confidentially prepared and supported by genuine capability.

An AI-assisted tender may be strong if the supplier has used the tool to structure information, improve clarity and check consistency while retaining control over the content.

An AI-assisted tender may be weak if it contains generic commitments, invented evidence, unrealistic delivery claims, copied content, unsupported methodologies or statements that do not reflect the supplier's actual resources.

The evaluator should still assess the tender against the published criteria. AI use may justify clarification or due diligence where it creates uncertainty, but it should not become an undisclosed scoring factor.

The confidentiality risk

One of the most important issues is the potential misuse of confidential information.

Suppliers may input tender documents, draft responses, pricing details, client information or authority-specific material into external AI tools. Officers may also use AI tools to summarise bids, draft specifications or develop evaluation guidance without fully understanding data retention, training, hosting or confidentiality implications.

This can create serious risk.

Confidential tender documents may be uploaded into systems outside the authority's control.

Supplier confidential information may be processed through tools not approved for that purpose.

Commercially sensitive pricing or solution details may be exposed.

Personal data may be used without a lawful basis or appropriate controls.

The authority may be unable to explain how information was processed or protected.

Councils should therefore set clear rules for their own officers and, where proportionate, for suppliers. Tender documents can include wording asking suppliers to disclose AI use in bid preparation and confirming that confidential information must not be used as training data for AI systems.

Internally, procurement teams should follow the council's information governance, cyber and digital policies before using AI tools on procurement material.

The evaluation risk

AI affects evaluation in two main ways.

First, suppliers may submit more polished but less specific responses. Evaluators may find it harder to distinguish genuine capability from well-written generic content.

Second, authorities may be tempted to use AI to support the evaluation process itself.

The first risk can be managed through stronger question design and evidence requirements. A quality question that asks for generic methodology will invite generic AI-assisted answers. A better question asks for contract-specific understanding, practical delivery detail, relevant evidence, named controls, measurable outputs and examples that can be tested.

The second risk requires greater caution. Evaluation decisions must be made in accordance with the published criteria by the authority's appointed evaluators. AI may support administrative tasks in tightly controlled circumstances, but it should not replace evaluator judgement, apply undisclosed criteria or produce unexplained scoring.

If AI is used internally to summarise material, identify issues or support drafting, the authority should retain a clear record of what was used, by whom, for what purpose and how the human evaluator reached the final judgement.

A council should be able to explain the evaluation process without relying on a black-box tool.

AI in service delivery is different from AI in bid writing

A supplier's use of AI to write or structure a tender is one issue. A supplier's use of AI to deliver the contract is another.

Where AI forms part of the service, the authority may need to consider technical, ethical, legal and operational issues.

What function will the AI perform?

Will it make recommendations or decisions affecting residents?

Will it process personal or special category data?

How will bias, accuracy and error be monitored?

Who is accountable for outputs?

Can a human review or override the system?

What data will be used to train, test or operate the model?

Where will data be hosted and processed?

How will the authority exit the contract or move to another provider?

These questions may be central to social care, revenues and benefits, customer contact, fraud, planning, housing, environmental services and corporate analytics procurements.

A council should not rely on general assurances that AI is secure, ethical or innovative. Requirements should be converted into specification, evaluation, contract terms and contract management controls.

Procurement documentation needs to catch up

Many tender packs were drafted before AI became a routine operational issue.

Councils should consider whether their standard procurement documents address AI disclosure, confidentiality, data use, evaluation evidence, cyber assurance, intellectual property, bias monitoring, audit rights, explainability, human oversight and exit arrangements where relevant.

This does not mean every procurement needs an AI schedule. A stationery purchase or basic maintenance contract will not require the same controls as an AI-enabled triage platform.

The principle is proportionality.

Low-risk procurements may require no AI-specific wording.

Moderate-risk procurements may require disclosure of AI use in bid preparation or service delivery.

High-risk procurements may require detailed technical, data protection, ethical, cyber and contract management controls.

The procurement strategy should identify the level of risk and explain the chosen approach.

Officers need clear internal rules

Supplier AI use is only half the issue. Councils also need internal controls for officer use of AI.

Procurement officers may use AI to draft first versions of specifications, summarise guidance, produce timetables, rewrite evaluation questions or prepare market research. These uses may be helpful, but they need boundaries.

Officers should not upload confidential supplier information into unapproved systems.

Officers should not rely on AI outputs without verification.

Officers should not use AI to create undisclosed evaluation criteria.

Officers should not allow AI to make scoring decisions.

Officers should not assume legal or procurement advice generated by AI is accurate.

Officers should record where AI has materially supported procurement documents or decision-making, particularly where it affects content that suppliers will rely upon.

The solution is not fear. The solution is governance that lets officers use tools safely and professionally.

AI and conflicts of interest

AI can also create conflict and equal-treatment issues that are less obvious than confidentiality or data protection risks.

A supplier may use AI trained on previous public tender responses, including material prepared for other authorities. This may not be improper, but it can create responses that look tailored while containing assumptions from another contract.

A consultant supporting a supplier may use AI tools across multiple bid clients. The supplier still needs to ensure that confidential information from another opportunity has not been reused.

A council officer may use AI to help draft documents and inadvertently create wording that favours a particular solution, incumbent supplier or market approach if the source material was too narrow.

An evaluator may use AI to summarise a long tender and miss nuance that should have been assessed manually.

These risks reinforce the need for human accountability. Procurement teams should not view AI as neutral simply because it produces fluent text. The source material, instructions, limitations and checking process matter.

Assessment summaries and audit trails

The Procurement Act 2023 places greater emphasis on clear communication of assessment outcomes. That makes the audit trail around AI use more important.

Where AI has been used internally to support drafting or administration, councils should be able to show that the final procurement decision was made by appropriate officers using the published criteria. The authority should retain enough information to explain how tender material was assessed and how scoring rationale was developed.

If a supplier challenges an evaluation outcome, the council will need to rely on human assessment records, not an unexplained AI-generated summary.

This does not mean every administrative use of technology must be disclosed publicly. It does mean that material use should be governed, authorised and capable of explanation.

For higher-risk procurements, a short internal AI use record may be sensible. It can identify whether AI was used, the tool category, the purpose, whether confidential information was processed, who reviewed the output and how final human judgement was applied.

What councils should do now

Create an AI procurement position statement.

Councils should set out how AI may be used by procurement officers, what tools are approved, what information must not be entered into AI systems and what records should be retained.

Update tender templates where proportionate.

Standard documents should include optional AI disclosure wording and confidentiality controls for procurements where AI use is reasonably relevant.

Improve quality question design.

Questions should test contract-specific capability, evidence and deliverability rather than inviting generic written responses that can be produced easily by AI.

Train evaluators to spot weak AI-assisted content.

Evaluators should look for unsupported claims, generic methodology, inconsistency, lack of evidence, unrealistic commitments and responses that do not answer the specific requirement.

Assess AI used in delivery separately.

Where AI forms part of the service, the authority should consider data, cyber, bias, explainability, accountability, audit and exit requirements as part of procurement planning.

Keep human accountability clear.

AI should not replace the authority's judgement. Decision records should show how officers reached conclusions against the published criteria.

Use practical support where internal policy is still developing.

Prestige Commercial Consulting's governance and procurement support can help authorities review procurement templates, evaluation models and audit trails. Officers who need structured learning on procurement process and evaluation discipline can use the PCC Learning Platform.

The supplier perspective

Suppliers should assume that AI-assisted tendering will become more visible.

A supplier that uses AI well should still own the response. It should verify facts, remove generic content, align the submission to the requirement and ensure that every claim can be evidenced.

Suppliers should not upload confidential council documents into public AI tools where that would breach tender conditions, confidentiality obligations or basic commercial prudence.

Where AI is part of the service being offered, suppliers should be prepared to explain how it works, what data it uses, how errors are managed, how humans remain accountable and how the authority can audit and exit the arrangement.

Good suppliers should welcome proportionate questions. They create an opportunity to distinguish credible capability from superficial AI-generated responses.

Closing takeaway

AI is already changing public procurement. Councils should not respond with panic or prohibition. They should respond with practical transparency, proportionate controls and stronger evidence standards.

The key distinction is simple. AI can support procurement activity, but it must not obscure accountability, weaken confidentiality, distort evaluation or replace officer judgement.

Authorities that update templates, train evaluators, clarify internal rules and manage AI-enabled services properly will be better placed to benefit from innovation while protecting fairness and public trust.

For support with AI-related procurement controls, tender documentation, evaluation assurance or governance review, contact Prestige Commercial Consulting Limited.