Selective AI Policy Pilot Portfolio: A method and eighteen instruments for deciding which AI pilots should proceed

A decision method for public institutions that receive more AI proposals than they can responsibly support. It sets out how to decide which proposals proceed to bounded testing, which need redesign, which should remain research or policy work, and which should stop — and it supplies the eighteen instruments that carry the decision from first idea to closure. Selection is treated as a public-value judgement, not a competition for the most technically ambitious idea. The controlling question at every stage is whether a small, controlled test can answer an important policy question without exposing people or institutions to unjustified risk. The portfolio is organised in six parts: Part I — Purpose and operating model. What counts as a pilot and what does not; ten portfolio principles; an eight-stage pathway in which every stage has a controlling question and leaves a required record; five decision families from proceed to design to do not proceed Part II — Selection and authorisation. Eligibility gates, non-compensable conditions, comparative scoring, panel decision records and bounded authorisation Part III — Pilot design and safeguards. Readiness, theory of change and assumption ledger, participation and inclusion, risk and remedy, data governance, vendor and sovereignty conditions Part IV — Monitoring, learning and decision gates. Measurement and learning plans, incident and complaint logging, and the gate review itself Part V — Scale, transfer and closure. Why scale requires a fresh decision, and how an intervention ends, continues, transfers or closes Part VI — Reusable instruments. The eighteen forms, screens, scorecards, registers, ledgers and records used throughout, numbered and printable Two appendices follow: a worked fictional example running one proposal through the full pathway, and a rapid reference and glossary. Ten principles govern the method — problem before technology, selectivity, proportionality, inclusion, authority, reversibility, remedy, evidence humility, local agency, and no automatic scale. Each is stated with its operational meaning, so it can be applied rather than merely endorsed. The portfolio was developed from the eight capstone projects submitted by Cohort 1 of the AI Literacy Fellowship for African Policymakers, written by forty-one fellows and published in full in the companion volume. It completes a collection of five reusable artefacts derived from that volume: a governance blueprint, a practitioner toolkit, four policy briefs, a simulation casebook, and this pilot portfolio. Published by the OpenSchool Initiative, Abuja, Nigeria. First open publication edition, 2026. Licensed CC BY 4.0. This is an open policy and capacity-building resource. Anyone applying the method in a real jurisdiction must substitute verified local law, authorities, data-protection requirements and institutional arrangements for the illustrative material, and engage affected groups and relevant specialists before authorising a pilot.