Public-Sector AI Procurement and Assurance Adviser

The mandate

The council repeatedly asks how to buy AI capabilities whose model, data and operating behaviour will change after award. Draft procurements specify broad accuracy and innovation goals but leave public-service outcomes, cohort performance, decision challenge, supplier evidence, model updates, data reuse and exit mechanics too vague to compare bids or hold a provider accountable.

The adviser will reserve three days monthly for requirements and evidence review, a chair or council session, and preparation or individual challenge with service and procurement leaders. Six council meetings are included; urgent questions during a live procurement receive acknowledgement within one business day and a written view within three, subject to equal-treatment and confidentiality controls.

The appointment runs for twelve months through award and initial assurance design for two priority services. The council chair may propose one renewal of up to six months if a named procurement challenge or acceptance question remains unresolved; renewal requires the council’s recorded approval and is not caused automatically by supplier delay.

The adviser holds no line authority, evaluation vote, procurement delegation or executive responsibility for requirements, bidder communication, scoring, award, service decisions, data processing or expenditure. Accountable officials retain every decision, and advice must be available through the approved record rather than private influence over an evaluator or bidder.

Other public-interest and non-bidding advisory work may continue after disclosure. A role, investment, referral arrangement or research sponsorship involving a bidder, major subcontractor, assurance provider or competing public buyer using the same confidential proposal creates a conflict requiring information restriction, recusal or termination.

Why the board wants this voice

Procurement leaders understand fair process and service owners understand public outcomes, but the council lacks a member who has contracted evolving AI systems with reproducible acceptance and exit. Current drafts either freeze technical detail prematurely or defer every consequential control to post-award design. The chair wants a practitioner who can preserve competition while making performance, change and accountability contractible.

What you will own

• Press service owners to define the decision or workflow outcome, affected users, safe alternative, accountable official and consequence of model failure.
• Test requirements for representative cohorts, uncertainty, accessibility, human review, challenge, incident response and measurable public-service impact.
• Shape bid evidence around runnable evaluations, data lineage, model and prompt identity, operating capacity and independently inspectable limitations.
• Challenge contract terms for training and improvement rights, subcontractors, model change, audit, incident notice, correction, records and public explanation.
• Probe commercial models for hidden inference, annotation, integration, assurance, retraining and exit costs under changing demand.
• Design acceptance and renewal gates that distinguish demonstration, shadow operation, limited deployment and wider service authority.
• Frame supplier-exit provisions for data return, artefact portability, replacement cooperation, model access, secure deletion and continuity of public service.

Candidate qualifications

• Led procurement, commercial assurance or board oversight for AI used in a consequential public or regulated service.
• Converted policy and service outcomes into testable model, workflow, cohort and human-review acceptance criteria.
• Negotiated AI terms covering training rights, model change, evidence access, subcontractors, incidents, portability and secure exit.
• Identified a low headline bid whose inference, integration, assurance or switching economics changed the award recommendation.
• Preserved fair competition and documented challenge while technical and political stakeholders preferred a named approach.
• Advised accountable officials without taking procurement authority or representing board advice as bidder endorsement.

Non-negotiables

• Can attend all six Delhi NCR council meetings and the three approved bidder, service-design or assurance sessions.
• Will disclose vendor, subcontractor, investor, public-buyer and assurance relationships before any procurement paper is shared.
• Accepts no scoring vote, bidder contact or private instruction outside the documented governance and equal-treatment process.
• Brings direct AI contracting evidence; general public policy, technology strategy or conventional software procurement is insufficient.

1. 49 words maximum. Describe one AI procurement term that changed after a model or operating risk became contractible.
2. 49 words maximum. Which current vendor, investor, public-buyer or assurance relationship could constrain your independence?
3. 49 words maximum. Confirm the Delhi NCR cadence and name the acceptance evidence you would require before wider service deployment.