Skip to main content

    AI strategy and governance

    AI operating model: design around the work AI will change

    An AI operating model sets how AI capability becomes a reliable business outcome. It connects people, processes, technology, information and controls, and assigns responsibility for the decisions between them. It also sets how an idea becomes funded work, a supported service and a result someone owns.

    In short

    Start with one consequential workflow, from trigger to result

    • Start with one consequential workflow, traced from its trigger to the accepted result, and identify what AI changes in it.
    • Six operating choices settle the design. They cover accountability, what the system may decide or do, shared capabilities, acceptance of work, sources of information, and value and failure.
    • A decision-rights map names who proposes, approves, performs, checks and can stop. It tests whether the permissions actually implemented match the intended design.

    Detail

    Six operating choices settle the design

    Follow a process from its trigger to the accepted result. Record decisions, hand-offs, exceptions and dependencies. Then identify what AI changes: whether it produces an output, recommends an action or executes a permitted step. Each design creates different demands on quality, authority, review and recovery. Conventional integration or assisted work may meet the objective without an autonomous agent.

    Six operating choices then settle the design.

    Accountability: name the business owner of the outcome. Separate responsibility for the workflow from responsibility for the model, integration and security controls, and make both explicit.

    What the system may decide or do: reading a purchase order, drafting a change and approving payment are different permissions. Decide which actions may occur within tested limits, which require meaningful approval and which remain prohibited. Enforce the limits in access controls as well as in the prompt.

    Shared capabilities: common platforms, evaluation, procurement and security standards can reduce duplication. Process expertise and benefit ownership often need to stay close to the business.

    Acceptance of work: define a good outcome and who can judge it, including difficult cases and the consequences of error. If AI removes preparation but increases senior review, review capacity becomes part of the design.

    Sources of information: identify authoritative sources, permissions, version control and the owners of updates. A knowledge base needs someone to look after it, because adding documents can make the context worse if their contradictions stay unresolved.

    Value and failure: measure accepted outcomes, full cost and material exposure. Set the triggers to investigate, roll back, re-approve or retire the system. Once deployed, the system needs an owner, a budget and a retirement trigger.

    Commercial impact

    An operating model gives AI work an owner and limits

    Without an operating model, AI work can reach teams with no owner for the result and no limit on what the system may do. The design has to cover the people as well as the system. Plan supervision, exception handling, customer relationships, professional judgement and learning into work shared by people and agents.

    Base any change to roles on the organisation's obligations and on how the work changed. A comparison between a licence price and a salary shows neither. Resilience can also require keeping people who understand the process, because they must be able to run the fallback.

    Begin with a scope the organisation can supervise and measure. Resolve its dependencies, then expand when the evidence supports it. Record the wider effects a local pilot will have, and show the next decision, its evidence and the owner responsible for producing it.

    Managers

    How AI changes the work of managers and executives

    AI changes what managers spend time on more than whether the role exists. Drafting, summarising and first-pass analysis move to the tool. Setting the brief, checking the output, deciding where released time goes and owning the result stay with the manager. Workforce decisions should follow evidence of how the work changed.

    Area of workMoves to the toolStays with the managerNew for the manager
    DraftingFirst drafts of papers, updates and repliesSetting the brief and approving the final versionChecking a draft for accuracy before it leaves the team
    SummarisingFirst-pass summaries of meetings, documents and dataDeciding what mattersChecking what a summary left out
    AnalysisFirst-pass analysis and optionsJudging the evidence and making the decisionAsking what information the tool could not reach
    Team timeRepeatable steps in recurring tasksBeing accountable for the resultDeciding where released time goes and which measure that work changes

    Ramon Rodriguez describes the executive side of this change in 'Loop Engineering: the AI skill executives need next'. The podcast episode 'Does AI always mean fewer people?' covers how chief executives move teams into higher-value work. Executive AI Coaching works on each executive's own papers, decisions and stakeholders.

    In practice

    A worked example of decision rights in a supplier workflow

    For each important step, identify who proposes, approves, performs, checks and can stop. Compare that map with the permissions actually implemented.

    Take a hypothetical supplier-record workflow. An assistant identifies possible duplicates and shows the evidence. A finance owner decides whether to merge them. A controlled system performs and records the approved change.

    Giving an agent broad access and asking it to clean up the supplier list differs greatly. The objective can be similar while authority and risk differ.

    What to do

    Steps to build ownership, limits and review into the model

    • Start from one important workflow, traced from trigger to accepted result. An organisation chart or a platform decision is the wrong starting point.
    • Separate ownership of the outcome from ownership of the model, integration and security controls, and make both explicit.
    • Where appropriate, enforce limits on what the AI may do in technical controls, as well as in the prompt that instructs it.
    • Plan review capacity, maintenance of information sources and recovery as part of the operating model.
    • Sequence change around constraints: start with a scope the organisation can supervise and measure, then expand when the evidence supports it.

    Last reviewed

    Manager tasks change first, and headcount is a separate decision

    1. How does AI change the role of a manager?

      The tasks change before the role does. A manager who used to draft a paper now briefs the tool, checks the draft and signs it off. Review takes a larger share of the week.

    2. Will AI mean fewer managers?

      Not necessarily. Headcount is a separate decision, made once the organisation can show how the work changed. Time released by AI counts as capacity until someone decides where it goes. It becomes a saving only when a cost falls.

    Put this topic against your own numbers, owners and timelines.