iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Manage AI as a portfolio of initiatives tied to specific organizational objectives—not as a collection of models or software purchases. For each proposed use, define the result it should deliver, establish a baseline, assess feasibility and risk, fund the capabilities it depends on, and review evidence before expanding it. This approach helps leaders decide which initiatives to fund, what to monitor, and when to revise or stop work.
Start with an organizational problem, not an AI tool
Describe the business or operational problem before selecting a model, vendor, or technology. A useful investment proposal names the people affected, the process that needs to change, the intended result, and why AI is a plausible way to achieve it. If the proposal cannot explain what should improve, it is not ready for an investment decision.
Make the intended outcome specific enough to test. “Use AI to improve customer service” is a direction, not a measurable target. A stronger proposal identifies the service measure to change, the population or workflow in scope, and the period over which the result will be assessed. Possible measures include resolution time, error rates, service availability, or staff time spent on a defined task; choose those that fit the use case rather than treating any one metric as universal.
Set a baseline and define what counts as value
Record the current state before deployment. Depending on the initiative, that may include performance, cost, quality, throughput, user experience, or existing risk. Specify how each measure will be collected and who owns it. Without a baseline, a favorable result after launch may be impossible to distinguish from normal variation or changes unrelated to the AI system.
#1 Best Overall
State the expected benefit and its assumptions. Include the counterfactual—what is likely to happen without the initiative—and consider costs across the system’s lifecycle, not only the initial purchase or build. Relevant costs can include data preparation, integration, infrastructure, workforce training, evaluation, oversight, maintenance, and eventual replacement or retirement. Do not present a projected benefit as a realized return until the organization has measured outcomes against its baseline.
The OECD’s 2025 review of AI in core government functions emphasizes strategic planning, monitoring, value for money, and impact assessment. It also notes that measurement matters for demonstrating return on investment and prioritizing future investment. These principles are useful for business decision-making, but the published figures in that review concern government use cases; they are not benchmarks for company ROI.
Compare proposals on the same decision dimensions
Use a consistent set of questions to compare initiatives, while allowing decision-makers to weigh factors according to organizational priorities and risk tolerance. OECD and NIST guidance supports these dimensions at a framework level, but does not establish a universal scoring formula or a reliable private-sector ROI benchmark.
| Dimension | Questions to answer |
|---|---|
| Strategic fit and outcome | Which organizational objective does the use case support? What specific outcome should change, and for whom? |
| Value and evidence | What is the baseline? What is the counterfactual? Which measures will show whether benefits and costs materialized? |
| Feasibility | Are the data, infrastructure, skills, integrations, procurement path, and operational ownership in place or realistically attainable? |
| Lifecycle sustainability | Can the organization operate, evaluate, maintain, and update the system over time? What continuing resources and dependencies will it require? |
| Risk and controls | What operational, financial, legal, security, and societal risks could arise? Who is accountable, and are proposed controls proportionate to the context? |
A proposal that scores well on potential benefit but depends on unavailable data, unassigned ownership, or unmanageable exposure is not automatically a good investment. Make those dependencies visible so leaders can decide whether to fund them, narrow the scope, or defer the initiative.
Fund the capabilities that make delivery possible
AI investment is an organizational capability decision as well as a technology decision. The OECD’s government-focused guidance identifies governance, data, digital infrastructure, skills, purposeful investment, procurement, partnerships, guardrails, oversight, and stakeholder engagement as enablers of trustworthy AI. For a business, these are useful planning prompts rather than direct government requirements.
- Governance and ownership: Assign responsibility for the business outcome, system operation, risk decisions, and escalation. Make clear who can approve a change or pause use.
- Data and infrastructure: Confirm that data can be accessed and used appropriately, and that infrastructure can support the intended workflow and oversight.
- People and process: Identify the skills required to build or procure the system, validate its outputs, support users, and handle exceptions.
- Procurement and partnerships: Assess vendor and partner dependencies alongside internal capabilities, including how the organization will maintain sufficient oversight.
- Stakeholder engagement: Involve affected teams and users early enough to surface operational needs, potential harms, and adoption barriers.
Manage risk throughout the AI lifecycle
Risk review should not be a one-time approval before launch. Consider how risks can emerge during design, development, procurement, deployment, use, evaluation, and change. The appropriate controls depend on the system’s context and potential effects; overly broad or poorly tailored guardrails can create unnecessary barriers, while inadequate controls can leave material risks unmanaged.
The NIST AI Risk Management Framework is a voluntary framework intended to improve how trustworthiness considerations are incorporated into AI systems’ design, development, use, and evaluation. Its companion Playbook suggests actions organized around four functions:
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →- Govern: Establish accountability, policies, roles, and oversight.
- Map: Understand the system’s context, intended use, affected parties, and potential impacts.
- Measure: Assess performance and risks using appropriate evidence.
- Manage: Prioritize and address risks, monitor the system, and respond when conditions change.
These functions can help structure a company’s risk process; they are not a mandated universal standard. The NIST Playbook is based on AI RMF 1.0. For enterprise due diligence, the OECD’s 2026 guidance connects responsible business conduct with the OECD AI Principles and is aimed at enterprises involved in developing and using AI across the value chain.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Run a bounded implementation and make a scale decision
Before committing to broad deployment, define a bounded implementation that can produce decision-quality evidence. Specify its scope, duration, users, outcome measures, risk checks, and conditions for pausing or ending it. Monitor both intended benefits and adverse or unexpected effects; a system can meet a narrow performance target while creating problems elsewhere in the workflow.
Rank #4
- Define the problem and target outcome. Name the process, objective, affected users, and result the initiative is intended to change.
- Establish baseline and value assumptions. Record current performance, measurement methods, expected benefits, lifecycle costs, and the counterfactual.
- Assess feasibility and risk. Check data, infrastructure, integration, skills, procurement, ownership, and context-specific risks before approving the scope.
- Fund enabling work. Include governance, data readiness, training, infrastructure, evaluation, and operational support in the investment decision.
- Implement with monitoring. Collect the agreed measures and review performance, costs, incidents, user experience, and changing conditions.
- Decide whether to scale, revise, or stop. Expand only when evidence supports the intended value and the organization can sustain the system and its controls. If results fall short, revise the use case or stop rather than treating continued spending as proof of commitment.
What the available government figures do—and do not—show
The OECD’s 2025 publication analyzed 200 AI use cases in government. Among those cases, 57% supported automated, streamlined, or tailored processes and services; 45% enhanced decision-making, sense-making, or forecasting; and 30% aimed to improve accountability or anomaly detection. The same publication reports that 15% of governments had an AI investments framework in 2023.
These figures describe public-sector use cases and government frameworks, not business outcomes or commercial returns. They can help illustrate the variety of purposes AI initiatives serve, but they cannot predict whether a particular company project will deliver value.
Recommended Free Tools
Use a portfolio view, not a single-project bet
Manage initiatives together so leaders can compare their strategic contribution, evidence, dependencies, and risks. A portfolio view makes it easier to see whether several projects depend on the same data or infrastructure, whether enabling capabilities are underfunded, and whether the organization is concentrating investment in one type of outcome. Revisit priorities as implementation evidence arrives; the portfolio should reflect what the organization learns, not only the assumptions made at approval.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
