Executive AI mandate and first 90 days

How to use this checklist

For executives, business-unit leaders and transformation sponsors in industrial B2B and complex enterprises.

Work through the six decisions with the business sponsor, workflow owners, finance, technology and risk teams. Mark each item as agreed, missing or contested; name an owner and a decision date. Use the 90-day sequence as a planning guide, adapted to your organisation's starting point.

• Downloads

1. Agree the mandate

Start with the operating problem and the authority to change it.

A one-page mandate with sponsor, outcomes, scope, resources and review dates.

• Name the executive sponsor, business scope, target outcomes and constraints. State which responsibilities are in the mandate and which remain elsewhere.
• Identify the workflows that may change and the customer or employee affected. Compare AI with process repair, integration and simpler alternatives.
• Agree resources, access to business owners and the review cadence. Record conflicts that require sponsor escalation.

2. Assign decision rights

Accountability needs authority over process, funding and deployment.

A decision-rights table that names the decision, accountable owner, contributors and escalation path.

• Name one business owner for each workflow outcome, a delivery owner and the people accountable for data, controls and adoption.
• Record who funds, approves a release, accepts residual risk, changes the process and can stop or roll back the system.
• Set escalation deadlines and approval evidence. Separate shared platform standards from local business decisions.

3. Choose a disciplined portfolio

Fund a small set of opportunities with evidence and a route to use.

A ranked opportunity register with owners, assumptions, costs and investment gates.

• For each candidate, record the performance gap, baseline, affected volume, data access and dependencies. Keep uncertain estimates visible.
• Compare expected benefit with implementation, integration, inference, review, change and operating costs. Test downside assumptions.
• Define the next experiment, a decision date and explicit continue, change or stop criteria. Identify what work will be retired if the new process succeeds.

4. Establish value baselines

Separate time saved, capacity released and value actually captured.

A value scorecard with baseline, target, observation period, source, owner and attribution limits.

• Have the business owner and finance agree the current performance, measurement period, data source and counterfactual before the intervention.
• Track adoption, cycle time, quality, exceptions and review burden alongside the financial or operating outcome. Use comparable cohorts where feasible.
• Record attribution limits and all operating costs. Decide who validates benefits and what evidence changes the investment decision.

5. Design controls into the workflow

Governance has to work during an actual decision or exception.

A release and incident record connecting purpose, evaluations, permissions, approvals and monitoring.

• Inventory intended uses, data sources and system permissions. Ask qualified specialists to assess applicable legal, privacy and security duties.
• Define evaluations, release criteria, human review, customer disclosure where applicable and restricted actions. Preserve versions and evidence.
• Test failure and incident scenarios. Name the escalation owner, monitoring thresholds, rollback method and a safe manual fallback.

6. Own adoption and capability

The new capability must survive after the project team leaves.

An adoption and operating plan with user cohorts, feedback, service ownership and transition criteria.

• Co-design the change with users and local managers. Identify incentives, training, workload and legacy habits that affect actual use.
• Pilot with a defined cohort and a feedback route. Measure repeat use and task completion as well as registrations or licences.
• Assign ongoing service ownership, support and budget. Transfer skills, document exceptions and agree when the old process can be retired.

Days 1–30: Diagnose and establish authority

Sponsor decision: approve the mandate, accountable owners and investigation priorities.

• Interview sponsors and workflow owners; observe the current process and customer journey.
• Complete the mandate and decision-rights table; inventory existing pilots, vendors and dependencies.
• Agree baseline measures with finance and identify material data, control and capability gaps.
• Select a small portfolio for investigation; explain which initiatives will pause and why.

Days 31–60: Test the intervention and its conditions

Investment decision: continue, change or stop each experiment against pre-agreed evidence.

• Compare alternatives and validate the highest-risk assumptions with users and delivery teams.
• Define experiment cohorts, evaluation criteria, permissions, human escalation and measurement.
• Estimate total costs and agree deployment, adoption and service requirements before expanding scope.
• Review evidence with business, finance, technology and risk owners; document unresolved trade-offs.

Days 61–90: Decide what becomes an operating capability

Operating decision: stage the rollout, investigate further or stop. Ninety days is a review horizon, not a promise of production readiness.

• Assess observed adoption, quality, exceptions and economics against the baseline and counterfactual.
• Run the release and incident review; confirm support, monitoring, rollback and ownership.
• Fund a staged rollout only where the evidence supports it; retain a clear stop condition.
• Publish the next-quarter plan with owners, resources and gates; transfer lessons and retire duplicate work.

Scope and citation

An original practitioner decision aid by Juan Beltrán. It is a synthesis of executive operating practice, not audited research, a maturity certification or a guarantee of results. The public industrial case illustrates selected principles; it does not validate this checklist as a universal model. Tailor it to your context and use qualified advice for consequential legal or risk decisions.

Beltrán, Juan. Executive AI mandate and first 90 days. Revised 2 October 2026. https://juanbeltran.ch/resources/executive-ai-mandate

Sources

• NIST AI Risk Management Framework
• European Commission: AI Act implementation
• McKinsey: publicly reported Holcim ordering pilot