Capability records, perspective 1: Strategy and value
Part of the SEI AI Adoption Framework (AIAF).
1.1 AI vision and business strategy
Purpose
Prevent AI activity from drifting away from strategy, so that investments, policies, and priorities are set tool by tool instead of from stated outcomes.
Intended outcome
The organization can explain where AI supports its strategy, which outcomes matter, and how adoption priorities will change as capabilities evolve.
Decisions enabled
- which outcomes AI should support and which it should not pursue;
- how much strategic weight AI initiatives receive against other investments;
- when priorities change because capabilities, regulation, or competitors change; and
- who may commit the organization to a direction.
Applicability
Workloads: All
Delivery modes: All
Typical state: Required
Key activities
- describe the outcomes that matter to customers, employees, owners, and the public;
- state where AI is expected to help, where it is not, and why;
- name accountable owners and decision rights for AI direction;
- connect the strategy to the portfolio, funding, and risk-appetite decisions it should guide;
- set a review cadence and the events that force an earlier review; and
- communicate the direction so that teams can tell whether a proposal fits.
Required evidence
- approved strategy or direction statement;
- stated objectives and the outcomes they serve;
- decision owners and review cadence;
- record of changes to direction with reasons; and
- evidence that portfolio decisions refer to the strategy.
Common failure patterns
- a strategy of general ambition that cannot rule any proposal in or out;
- AI direction set by the most enthusiastic team or vendor;
- no review, so the strategy stays fixed while conditions change;
- strategy that never reaches funding or portfolio decisions; and
- treating a technology forecast as a strategic commitment.
Capability levels
| Level | Observable condition |
|---|---|
| Emerging | Leaders describe AI ambitions in speeches or slides, but no owner, scope, or review exists and teams cannot tell what fits. |
| Defined | An accountable owner maintains an approved direction with objectives, boundaries, and a review date that proposals can be checked against. |
| Managed | Portfolio, funding, and risk decisions cite the direction. Proposals that do not fit are declined or reshaped, and the direction is reviewed on schedule. |
| Adaptive | The direction changes in response to evidence from deployments, providers, regulation, and the workforce, and leaders revise priorities across workloads and delivery modes without waiting for an annual cycle. |
Review and exit triggers
- a material change in business strategy, market, or regulation;
- evidence that realized results differ from the outcomes assumed;
- a significant change in provider capability, price, or terms;
- repeated proposals that the direction cannot decide; or
- a scheduled review date.
1.2 Opportunity discovery and portfolio management
Purpose
Prevent AI opportunities from being pursued in isolation, duplicated, hidden, or left running without an owner or a decision to continue.
Intended outcome
AI opportunities enter one visible portfolio with an owner, objective, stage, dependencies, and disposition.
Decisions enabled
- which opportunities enter the portfolio and which are declined;
- how they are ordered, combined, or sequenced;
- which items continue, pause, or stop at each review; and
- which dependencies or conflicts require coordinated action.
Applicability
Workloads: All
Delivery modes: All
Typical state: Required
Key activities
- provide one place where any person can propose an AI opportunity;
- record owner, objective, workload, delivery mode, stage, dependencies, and disposition;
- include unplanned and adjacent uses, such as features that arrive inside existing products;
- screen and order opportunities using the strategy, value, risk, and readiness;
- review the portfolio on a set cadence and record stop and retire decisions; and
- share portfolio results with the people who proposed opportunities.
Required evidence
- opportunity register with owner, stage, and disposition;
- prioritization rationale;
- portfolio review decisions;
- records of items stopped, merged, or retired; and
- dependency and conflict notes.
Common failure patterns
- parallel registers that disagree;
- opportunities that never receive a disposition;
- only new projects are tracked while embedded features and informal use are missed;
- prioritization by sponsor seniority rather than evidence; and
- stopped items disappear without recording what was learned.
Capability levels
| Level | Observable condition |
|---|---|
| Emerging | Opportunities are tracked by individual teams in lists or messages, and leaders cannot see the full set or its status. |
| Defined | One register lists each opportunity with an owner, objective, stage, dependencies, and disposition, and it is reviewed on a schedule. |
| Managed | Prioritization uses consistent criteria. The portfolio shows overlaps, dependencies, and stopped items, and review decisions are recorded and acted on. |
| Adaptive | Portfolio criteria and cadence adapt to what deployments teach. Leaders rebalance across workloads and delivery modes as evidence, capacity, and provider conditions change. |
Review and exit triggers
- an opportunity has no owner or disposition at the next review;
- two items serve the same outcome or depend on the same scarce resource;
- the strategy or risk appetite changes;
- a stage gate shows the item no longer meets its objective; or
- a provider feature makes a planned item unnecessary.
1.3 Value definition and realization
Full record: worked example 1 of 5 in Part 3, Capability model. The catalog entry in Part 4 gives the intended outcome, workloads, delivery modes, and condition.
1.4 Workload and delivery-mode classification
Purpose
Prevent decisions from treating all AI the same, so that the work AI performs and the way the organization receives and controls it are confused and the wrong controls are applied.
Intended outcome
Decision makers distinguish the work AI performs from how the organization receives and controls it.
Decisions enabled
- which workload and authority level applies to a use;
- which delivery mode applies and who controls what;
- which capabilities apply and at what depth; and
- whether a change in use requires reclassification.
Applicability
Workloads: All
Delivery modes: All
Typical state: Required
Key activities
- describe the work AI performs and who receives the result;
- set the workload (W1 to W4) and, for W4, the authority held (Observe, Communicate, or Commit);
- set the delivery mode (A to D) and where control sits;
- record the reasoning, including what would change the classification;
- classify embedded features and each use of a shared tool separately when the work differs; and
- reclassify when scope, authority, or mode changes.
Required evidence
- workload, authority level, and delivery mode for each use;
- classification rationale;
- reclassification records; and
- link from classification to the applicable capabilities.
Common failure patterns
- classifying by product name rather than by the work performed;
- giving an assistant read access and later granting actions without reclassifying;
- assuming a purchased service moves the responsibility to the provider;
- one classification for a tool used in very different ways; and
- classifications that are never revisited.
Capability levels
| Level | Observable condition |
|---|---|
| Emerging | Uses are described by product or enthusiasm, and workload, authority, and delivery mode are inferred differently by different people. |
| Defined | Each in-scope use has a recorded workload, authority level, and delivery mode with a short rationale, and the record drives which capabilities apply. |
| Managed | Classification is checked by someone other than the proposer. Reclassification is triggered by changes in scope, authority, or provider, and reviews confirm it. |
| Adaptive | Classification practice is tuned from incidents and review findings. Leaders refine the criteria and use the pattern of classifications to spot where the organization's exposure is shifting. |
Review and exit triggers
- the AI is given new tools, data, or authority;
- the provider changes where processing or control sits;
- users apply the tool to different work;
- an incident reveals the classification was wrong; or
- the use moves between delivery modes.
1.5 Solution and sourcing pattern selection
Purpose
Prevent sourcing from being decided by familiarity, demonstration quality, or price alone, leaving the organization with a provider, pattern, or level of customization that does not fit its objective and constraints.
Intended outcome
The organization selects a provider, acquisition approach, architecture pattern, and customization level that fit the objective and constraints.
Decisions enabled
- whether to buy, configure, build, assemble, or subscribe;
- which provider and architecture pattern to use;
- how much to customize; and
- what would cause the choice to be revisited.
Applicability
Workloads: All
Delivery modes: All
Typical state: Shared
Key activities
- state the objective, constraints, and consequence before comparing options;
- compare alternatives, including a non-AI approach and not acting;
- estimate total cost, including integration, evaluation, support, change, and exit;
- use findings from provider risk, portability, and legal review in the decision;
- record the choice, the reasons, and the conditions for revisiting it; and
- connect the decision to funding and gate decisions.
Required evidence
- decision record with alternatives considered;
- total-cost comparison;
- constraints and assumptions;
- revisit triggers and owner; and
- links to risk, portability, and legal findings.
Provider evidence
Ask the provider for:
- provider documentation of capabilities, limits, terms, and roadmap commitments;
- independent evidence of performance in comparable use; and
- contract terms that support the assumptions in the decision.
Common failure patterns
- selecting the provider before defining the objective;
- comparing only AI options;
- counting license price while omitting integration and exit cost;
- customizing beyond what the organization can maintain; and
- no revisit trigger, so the choice outlasts its reasons.
Capability levels
| Level | Observable condition |
|---|---|
| Emerging | Sourcing choices follow the preference of a sponsor or the first acceptable vendor, and reasons are not recorded. |
| Defined | Each in-scope choice has a decision record with objective, alternatives, cost, and constraints. |
| Managed | Decisions use consistent criteria and findings from portability, provider risk, and legal review. Revisit triggers are monitored. |
| Adaptive | Sourcing criteria change as markets and internal capability change. Leaders move work between delivery modes and providers on evidence rather than inertia. |
Review and exit triggers
- provider terms, price, features, or location change materially;
- realized cost or performance departs from the decision record;
- the objective or risk appetite changes;
- a better alternative becomes available; or
- the provider's viability or ownership changes.
Boundary
Decides which provider, acquisition approach, and pattern to use. 5.4 designs for the ability to switch, 6.8 watches change after selection, and 7.8 assesses provider risk.
1.6 Investment governance and decision gates
Purpose
Prevent investments from continuing by default, or being stopped by chance, because decisions are not tied to evidence suited to cost and consequence.
Intended outcome
Investments continue, change, pause, or stop based on evidence appropriate to their cost and consequence.
Decisions enabled
- whether an item advances to the next stage;
- what funding it receives and under what conditions;
- whether to change scope, pause, or stop; and
- who may approve at each level of cost and consequence.
Applicability
Workloads: All
Delivery modes: All
Typical state: Required
Key activities
- define stage gates and the evidence each requires, scaled to cost and consequence;
- name approvers and their authority;
- link gate decisions to funding releases;
- record conditions attached to approvals and check they are met;
- make stop and pause decisions routine and visible; and
- reconcile gates with the checkpoints (Investigate, Select, Pilot, Expand, Continue) without making them mandatory boards.
Required evidence
- gate criteria by consequence;
- approval and rejection records;
- funding and stop decisions;
- conditions and their closure; and
- gate evidence packages.
Common failure patterns
- gates that approve every proposal;
- the same heavy process for low-consequence uses and light process for high ones;
- conditions that are never checked;
- stopping treated as failure, so items drift on; and
- funding released without evidence.
Capability levels
| Level | Observable condition |
|---|---|
| Emerging | Approvals happen through informal discussion or sponsor judgment, and evidence requirements are unclear. |
| Defined | Gates, evidence expectations, and approvers are defined for in-scope investments, and decisions are recorded. |
| Managed | Gate depth scales with cost and consequence. Conditions are tracked to closure, and stop and pause decisions occur when evidence calls for them. |
| Adaptive | Gate design changes in response to outcomes. Leaders shorten, remove, or deepen gates where evidence shows they add or lack value. |
Review and exit triggers
- an item advances without the required evidence;
- cost or consequence exceeds the level at which it was approved;
- conditions of approval are not met;
- gate decisions are overturned repeatedly; or
- a checkpoint passes with no recorded decision.
1.7 Risk appetite and prohibited uses
Purpose
Prevent the organization from discovering its limits only after harm, by stating in advance which outcomes, uses, data, actions, and authority levels are unacceptable or require escalation.
Intended outcome
The organization states which outcomes, uses, data, actions, and authority levels are unacceptable or require escalation.
Decisions enabled
- which uses are prohibited;
- which require escalation or executive approval;
- how much residual risk is acceptable; and
- when an exception may be granted and by whom.
Applicability
Workloads: All
Delivery modes: All
Typical state: Required
Key activities
- state unacceptable outcomes in terms of people, rights, safety, finances, and reputation;
- define prohibited and escalation-required uses, data, actions, and authority;
- set limits by workload and authority level, including what AI may commit externally;
- set the process and authority for exceptions and record them;
- review the statement against incidents, regulation, and stakeholder expectations; and
- translate the statement into rules others can apply.
Required evidence
- risk-appetite statement approved by accountable leaders;
- prohibited-use rules;
- escalation criteria and paths;
- exception records; and
- review dates and changes.
Common failure patterns
- a statement too general to decide any case;
- limits that exist in policy but not in review practice;
- exceptions granted informally;
- appetite stated once and never revised; and
- appetite that differs by team without anyone deciding it should.
Capability levels
| Level | Observable condition |
|---|---|
| Emerging | Leaders hold views about acceptable risk, but they are unwritten and applied inconsistently. |
| Defined | An approved statement names prohibited uses, escalation criteria, and exception authority. |
| Managed | Reviews cite the statement when deciding. Exceptions are recorded, counted, and reviewed, and the statement is revised on schedule. |
| Adaptive | Appetite is adjusted from evidence about incidents, outcomes, and stakeholder trust, and leaders test whether limits are working before a failure forces the question. |
Review and exit triggers
- an incident or near miss shows a limit was missing or wrong;
- new authority levels or tools are introduced;
- law, regulation, or stakeholder expectations change;
- exceptions become frequent; or
- a scheduled review date.
Boundary
States what is unacceptable or needs escalation. 3.1 tells people the rules, and 3.3 sets how review depth scales.
1.8 Market and capability horizon
Purpose
Prevent decisions from resting on stale assumptions or unexamined forecasts about technology, providers, regulation, customers, workforce, and competitors.
Intended outcome
Leaders monitor technology, provider, regulatory, customer, workforce, and competitive change without treating forecasts as facts.
Decisions enabled
- which changes warrant action now and which warrant watching;
- when to revisit strategy, sourcing, or risk appetite;
- what to start, accelerate, or stop; and
- how much confidence to place in a forecast.
Applicability
Workloads: All
Delivery modes: All
Typical state: Lightweight
Key activities
- choose the changes worth watching and who watches them;
- record each as a claim with evidence, confidence, and a date;
- define signals that would change a decision;
- separate observed facts from forecasts and opinion;
- feed findings to strategy, portfolio, sourcing, and workforce decisions; and
- retire items that no longer matter.
Required evidence
- horizon register with sources and confidence;
- signals and triggers;
- review dates and outcomes; and
- decisions that cited the register.
Common failure patterns
- collecting news without connecting it to a decision;
- treating a vendor roadmap as a commitment;
- forecasts never compared with what happened;
- the register owned by nobody; and
- attention concentrated on technology while workforce and regulatory change are ignored.
Capability levels
| Level | Observable condition |
|---|---|
| Emerging | Leaders follow industry news informally, and nobody is responsible for connecting it to decisions. |
| Defined | A named owner keeps a register of important changes, with evidence, confidence, and review dates. |
| Managed | Signals and triggers are defined and checked. Findings reach the decisions they affect, and past forecasts are compared with outcomes. |
| Adaptive | The watch list and methods adapt as forecasts are tested. Leaders act early on weak signals when the cost of waiting is high and avoid reacting to noise. |
Review and exit triggers
- a signal reaches its trigger;
- a forecast is shown to be wrong;
- the strategy or portfolio changes what matters; or
- an item has no review date.
