Capability records, perspective 2: People and operating model

Part of the SEI AI Adoption Framework (AIAF).

2.1 AI literacy

Purpose

Prevent harm and wasted effort from people using AI without understanding what it can do, where it fails, and what they remain responsible for.

Intended outcome

People understand the opportunities, limits, responsibilities, and verification practices relevant to their roles.

Decisions enabled

  • which roles need which learning;
  • who may use AI for which work;
  • when learning must be refreshed; and
  • whether people can verify AI output well enough to rely on it.

Applicability

Workloads: All
Delivery modes: All
Typical state: Required

Key activities

  • define role-based expectations, including leaders, builders, reviewers, and ordinary users;
  • teach capabilities, limits, data handling, verification, and escalation using real tasks;
  • assess whether people can apply what they learned;
  • refresh learning when tools, policies, or risks change;
  • give people a place to ask questions and report problems; and
  • record who has completed what.

Required evidence

  • role-based learning content and assignments;
  • assessment of applied skill;
  • completion records;
  • refresh cadence; and
  • feedback and question logs.

Common failure patterns

  • one generic course for every role;
  • completion counted while ability is not tested;
  • learning that stops after launch;
  • leaders and reviewers excluded; and
  • content that ignores the tools people actually use.

Capability levels

Level Observable condition
Emerging Learning is optional or ad hoc, and people rely on what colleagues or vendors tell them.
Defined Role-based learning exists for in-scope users and records show who has taken it.
Managed Learning is tied to access and tested through applied tasks. Gaps found in reviews and incidents change the content.
Adaptive Learning adapts to new tools and observed errors, and the organization uses assessed skill to decide where to extend access and autonomy.

Review and exit triggers

  • a new tool, feature, or authority level is introduced;
  • incidents or reviews show misuse or over-reliance;
  • policy or law changes; or
  • assessment results fall below the expected level.

2.2 Workforce impact and role redesign

Purpose

Prevent workforce effects from being discovered after the fact, when tasks, roles, workload, opportunity, and staffing have already changed without anyone planning for them.

Intended outcome

The organization anticipates how AI changes tasks, roles, workload, opportunity, and staffing rather than treating workforce effects as an afterthought.

Decisions enabled

  • which tasks and roles AI changes;
  • whether to redesign, redeploy, reskill, or reduce;
  • how benefits and burdens are shared; and
  • when to slow or change a deployment because of its effect on people.

Applicability

Workloads: All
Delivery modes: All
Typical state: Required

Key activities

  • assess how each in-scope use changes tasks, skill needs, workload, autonomy, and opportunity;
  • involve affected people and their representatives early;
  • identify who gains, who loses, and who absorbs hidden work such as checking;
  • plan transition actions including learning, redeployment, and support;
  • follow legal and consultation obligations; and
  • review actual effects after adoption.

Required evidence

  • workforce-impact assessment;
  • role and task changes;
  • consultation and engagement records;
  • transition actions and outcomes; and
  • post-adoption review.

Common failure patterns

  • assuming time saved becomes value without examining the effects on people;
  • discovering role changes when output metrics fall;
  • hidden review work assigned to the same people;
  • consultation after the decision; and
  • attributing workforce decisions to AI without evidence.

Capability levels

Level Observable condition
Emerging Workforce effects are discussed informally and are addressed only when complaints or departures occur.
Defined In-scope uses have a workforce-impact assessment with named roles and a transition plan.
Managed Assessments involve affected people, actual effects are reviewed, and findings change deployment, staffing, or support decisions.
Adaptive Practice adapts to what is learned across deployments. Leaders redesign roles and work early and share benefits deliberately.

Review and exit triggers

  • a deployment changes tasks, workload, or staffing materially;
  • affected people report unplanned effects;
  • legal or consultation requirements change; or
  • actual effects differ from the assessment.

2.3 Roles, accountability, and decision rights

Purpose

Prevent decisions about AI from being unowned, so that nobody can approve, stop, answer for, or fix the AI when something goes wrong.

Intended outcome

Every AI capability has accountable business, technical, risk, data, security, and operating decisions assigned to named roles.

Decisions enabled

  • who is accountable for outcomes, risk, data, security, and operation;
  • who may approve, change, pause, or stop;
  • how decisions are escalated; and
  • how decision rights are held when the provider or integrator is involved.

Applicability

Workloads: All
Delivery modes: All
Typical state: Required

Key activities

  • assign accountable business, technical, risk, data, security, and operating roles by capability;
  • separate those who build or propose from those who approve;
  • record decision rights and limits;
  • include provider and integrator responsibilities and the organization's counterpart;
  • update the assignments when people or scope change; and
  • test the assignments in reviews and incidents.

Required evidence

  • responsibility and decision-rights map;
  • role descriptions;
  • delegation records;
  • escalation paths; and
  • evidence of use in reviews or incidents.

Common failure patterns

  • accountability shared by a committee and held by no person;
  • owners who lack authority or time;
  • builder and approver are the same person on consequential work;
  • assignments that lapse when people move; and
  • provider responsibilities assumed but not agreed.

Capability levels

Level Observable condition
Emerging Ownership is assumed or inferred, and people disagree about who decides or answers.
Defined Each in-scope capability has named roles with documented decision rights.
Managed The map covers business, technical, risk, data, security, and operating decisions, is kept current, and is tested in reviews and incidents.
Adaptive Decision rights are adapted as autonomy, scale, and provider arrangements change, and leaders delegate by evidence of competence and consequence.

Review and exit triggers

  • an owner leaves or changes role;
  • authority or scope of the AI changes;
  • an incident reveals a missing or disputed owner;
  • the provider or integrator changes; or
  • the operating model changes.

2.4 Capability and talent strategy

Purpose

Prevent the organization from committing to AI capability it cannot build, buy, run, or oversee because the needed skills are missing.

Intended outcome

The organization can acquire, develop, retain, or source the skills needed for its target AI capability.

Decisions enabled

  • which skills to build, hire, borrow, or buy;
  • how much dependence on external expertise is acceptable;
  • where to invest in learning; and
  • whether a planned capability is feasible with current people.

Applicability

Workloads: All
Delivery modes: All
Typical state: Required

Key activities

  • describe the skills needed for the target capability, including engineering, evaluation, risk, operations, and domain expertise;
  • assess current skills against the need;
  • decide how each gap is closed and who is accountable;
  • keep critical knowledge from sitting with one person or provider;
  • plan learning, hiring, and sourcing in step with the portfolio; and
  • review results and revise the plan.

Required evidence

  • skills requirements by capability;
  • gap assessment;
  • learning, hiring, and sourcing decisions;
  • succession or coverage for critical skills; and
  • review of whether gaps closed.

Common failure patterns

  • plans that list tools but not people;
  • reliance on a single expert or vendor team;
  • hiring that follows hype;
  • skills for oversight and evaluation overlooked; and
  • skills gaps found at deployment.

Capability levels

Level Observable condition
Emerging Skills are discovered when projects stall, and responses are individual and reactive.
Defined Required skills and gaps are listed for in-scope capabilities with an owner and plan.
Managed The plan aligns with the portfolio, covers critical dependence, and is reviewed against results.
Adaptive The plan adapts to technology and workforce change, and the organization builds skills before the need becomes urgent.

Review and exit triggers

  • a planned capability requires a skill the organization lacks;
  • a key person leaves;
  • the portfolio or delivery mode changes;
  • provider support is reduced; or
  • a gap closure plan misses its date.

2.5 Change and adoption management

Purpose

Prevent adoption from failing or harming people because they were not told why work is changing, how it will change, where to get help, or how their feedback matters.

Intended outcome

Affected people understand the reason for change, how work will change, where to get help, and how feedback influences adoption.

Decisions enabled

  • who must be engaged and when;
  • how to sequence communication and support;
  • whether adoption is on track or needs intervention; and
  • how feedback changes the plan.

Applicability

Workloads: All
Delivery modes: A, B, C
Typical state: Required

Key activities

  • identify affected groups and their concerns;
  • explain the reason for change, the effect on work, and what is not changing;
  • provide training, support, and ways to raise issues;
  • use champions and managers in a way that does not hide problems;
  • measure adoption, comprehension, and sentiment; and
  • act on feedback and tell people what changed.

Required evidence

  • stakeholder and communication plan;
  • support arrangements;
  • adoption and feedback measures;
  • records of changes made in response; and
  • results of engagement activities.

Common failure patterns

  • announcing a tool without explaining the work change;
  • counting logins as adoption;
  • support that ends at launch;
  • feedback collected and not used; and
  • forcing use without addressing real objections.

Capability levels

Level Observable condition
Emerging Communication is a launch announcement, and support is handled by individuals when problems arise.
Defined A plan identifies affected groups, messages, support, and measures for in-scope changes.
Managed Adoption evidence includes understanding and sentiment, and feedback changes the plan and is reported back.
Adaptive Change practice is refined from results across deployments, and leaders adjust pace and approach by group instead of one schedule for everyone.

Review and exit triggers

  • adoption or sentiment falls below expectation;
  • support demand exceeds capacity;
  • a significant feature or policy change occurs; or
  • feedback shows a misunderstanding that communication has not resolved.

2.6 Human-AI work design

Purpose

Prevent poorly allocated work between people and AI, where checking is nominal, handoffs are unclear, or exceptions fall to no one.

Intended outcome

Work assigns appropriate responsibilities to people and AI, including verification, handoffs, escalation, and exception handling.

Decisions enabled

  • which responsibilities belong to people and which to AI;
  • where verification is required and by whom;
  • how exceptions and failures are handled; and
  • whether the work design is workable for the people who carry it out.

Applicability

Workloads: All
Delivery modes: A, B, C
Typical state: Required

Key activities

  • describe the work as it will be done, including review points and handoffs;
  • assign verification so reviewers have the time, information, and authority to do it;
  • define escalation and exception handling;
  • design for the failure modes of the AI, including confident error;
  • test the design with the people who will use it; and
  • observe use and revise.

Required evidence

  • work design and responsibilities;
  • review and verification points;
  • escalation and exception paths;
  • user testing results; and
  • observations of the work in use.

Common failure patterns

  • review as a formality;
  • reviewers who cannot see what the AI did;
  • exceptions that return to the least-informed person;
  • design that assumes the AI is right; and
  • no test with real users.

Capability levels

Level Observable condition
Emerging AI is added to existing work and people decide informally how to use it.
Defined The work design names responsibilities, review points, and escalation for in-scope uses.
Managed Design is tested with users, verification is resourced, and observed use leads to changes.
Adaptive Designs adapt to observed error and user skill, and the organization adjusts the balance between people and AI by evidence.

Review and exit triggers

  • errors reach outcomes despite review;
  • reviewers cannot keep up or stop checking;
  • AI behavior or authority changes;
  • volume or work mix changes; or
  • users report that the design does not fit the work.

2.7 AI operating model

Purpose

Prevent AI from being run through scattered teams, duplicated services, and unfunded commitments because structure, funding, and governance do not match the scale of the portfolio.

Intended outcome

Structure, funding, ownership, shared services, federated responsibilities, and governance match the organization's portfolio and scale.

Decisions enabled

  • what is centralized, shared, or federated;
  • how AI work is funded and who owns the cost;
  • which services teams may rely on; and
  • when the model should change as the portfolio grows.

Applicability

Workloads: All
Delivery modes: All
Typical state: Required

Key activities

  • describe the structure, services, and decision forums for AI;
  • assign ownership and funding for shared and local responsibilities;
  • define service boundaries and how teams request support;
  • align governance with the portfolio, scale, and risk;
  • check that responsibilities do not fall between teams; and
  • review the model as the portfolio and organization change.

Required evidence

  • operating model description;
  • service boundaries and responsibilities;
  • funding and ownership model;
  • governance forums and decisions; and
  • review of fit.

Common failure patterns

  • a central team that becomes a bottleneck;
  • no central function, so each team solves the same problems;
  • shared services without funding;
  • governance that does not match the scale of use; and
  • responsibilities that exist on paper only.

Capability levels

Level Observable condition
Emerging AI work is organized by whoever starts it, and shared needs are met informally.
Defined A defined operating model names structure, ownership, funding, and shared services.
Managed The model fits the portfolio, services are used, and responsibilities are checked for gaps.
Adaptive The model changes as scale, regulation, and capability change, and leaders move responsibilities between central and federated teams on evidence.

Review and exit triggers

  • portfolio size or risk changes materially;
  • duplicated or orphaned services appear;
  • funding for shared services is withdrawn;
  • a reorganization changes ownership; or
  • governance becomes a bottleneck.

2.8 Productivity-tool strategy and equitable access

Purpose

Prevent personal productivity tools from being bought, spread, and renewed through enthusiasm, leaving unequal access, unmeasured benefit, and licenses that no one needs.

Intended outcome

Productivity tools are selected, piloted, allocated, supported, measured, renewed, and retired according to work needs and outcomes.

Decisions enabled

  • which tools to offer and to whom;
  • how licenses are allocated, moved, or withdrawn;
  • whether a pilot should expand; and
  • whether to renew, replace, or retire a tool.

Applicability

Workloads: W1
Delivery modes: B, C
Typical state: Conditional, when the organization provides productivity tools to people

Key activities

  • define selection criteria tied to work needs;
  • pilot with cohorts that represent the roles that would use the tool;
  • set an allocation policy that avoids unexplained advantage;
  • provide support and learning for the tool;
  • measure outcomes in addition to usage; and
  • review renewals and retire tools that no longer serve a need.

Required evidence

  • selection criteria;
  • pilot cohorts and results;
  • license policy and allocations;
  • outcome and renewal reviews; and
  • retirement decisions.

Common failure patterns

  • licenses allocated by seniority or request speed;
  • usage counted as value;
  • tools renewed automatically;
  • pilots limited to enthusiasts; and
  • no way for people with a work need to request access.

Capability levels

Level Observable condition
Emerging Tools are acquired by teams or individuals, and access depends on who asked.
Defined Selection criteria, an allocation policy, and a renewal date exist for each tool in scope.
Managed Pilots represent real roles, outcome measures inform renewals, and access decisions can be explained.
Adaptive Allocation and selection adapt to measured benefit and changing needs, and licenses move to where they produce value.

Review and exit triggers

  • usage or outcomes fall below the renewal threshold;
  • a cohort lacks access without good reason;
  • price or terms change;
  • a better tool becomes available; or
  • a tool begins to be used for work beyond its approved scope.