Build the case around a measurable business problem, not the appeal of a model. Before funding deployment, define the current baseline, compare AI with business as usual and credible alternatives, estimate full costs, test uncertain assumptions, and set risk controls and decision gates. The result should be a conditional decision—pilot, fund, defer, or reject—not a guaranteed return forecast.
1. Define the problem and the comparator
Describe the workflow that needs improvement, who experiences the problem, how it works today, and what the problem costs in money, time, quality, or capacity. Be specific enough to identify a baseline: for example, current error or rework rates, turnaround time, throughput, or customer satisfaction.
As an Amazon Associate I earn from qualifying purchases.
Compare the proposed AI system with a credible business-as-usual case and with alternatives such as process redesign, conventional automation, or buying an existing capability. The UK government’s impact-evaluation guidance says to document precisely what business as usual means when it is the comparator. If there is no defensible baseline or meaningful outcome to improve, defer the investment case until one can be established.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
2. Define how success will be measured
Translate the desired outcome into measures, a target, a time window, and a method of observation before deployment. Pair financial measures—such as revenue, cost, or rework—with operational and human outcomes where they matter, including quality, decision-making, staff capacity, confidence, and satisfaction. Australia’s National AI Centre notes that measures can include fewer errors, faster turnaround, increased revenue, better decisions, and more time for higher-value staff work. See its ROI guidance.
#1 Best Overall
Do not count time saved as cash savings unless you explain the mechanism. It might reduce paid hours, increase throughput with existing staff, or free capacity for higher-value work. Label estimates as assumptions and use a range when uncertainty warrants it. There is no universal AI ROI percentage or payback period established here; calculate against your own baseline and assumptions.
3. Test feasibility before scaling
When data readiness, technical feasibility, or likely benefit is uncertain, test the key assumption before asking for a full deployment budget. GOV.UK recommends initial data analysis and a small proof of concept to explore feasibility and support the business case. Its guidance also warns that AI discovery can take longer than comparable non-AI work. See Assessing if artificial intelligence is the right solution.
A proof of concept tests a hypothesis; it does not by itself establish production-scale value. Set the data requirements, affected users, duration, evaluation method, success threshold, and stop conditions in advance. Do not assume a limited pilot result will hold at production volumes, with different edge cases, or at different operating costs.
4. Estimate the full cost and test sensitivities
Estimate costs over the same period used to estimate benefits. Separate one-time build or setup costs from recurring costs, and include the categories relevant to your project:
- Data preparation, model or service charges, cloud and infrastructure.
- Integration, security, legal work, procurement, and licensing.
- Internal staff time, training, and change management.
- Human review, monitoring, support, maintenance, and overhead.
- Exit or replacement costs if the organization changes supplier or approach.
These are cost categories to examine, not a universal checklist or published benchmark. OECD guidance on public-sector investment and adoption highlights uncertainty around costs such as licensing, cloud, staffing, procurement, overhead, and maintenance; organizations need to estimate their own context. See OECD’s 2025 discussion of trustworthy AI investment and OECD.AI’s overview of AI in government.
Show base, optimistic, and downside assumptions where useful. Test what happens if adoption is slower, quality is lower, human review remains intensive, volumes change, or costs rise. State whether the proposal depends on a particular vendor, data source, or infrastructure commitment, and how the organization could change course. OECD’s guidance is about public administrations; use its cost uncertainties as prompts, not as cost estimates for a private-sector project.
5. Compare options on the same basis
If there are at least two credible approaches, compare them against the same business outcome and time horizon. Include business as usual or doing nothing if it is a realistic comparator. Use consistent criteria rather than giving the AI option a different or more favorable yardstick.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problems| Comparison area | Question for the business case |
|---|---|
| Expected value | What financial and non-financial outcomes could each option deliver? |
| Evidence and likelihood | What supports the estimate, and which assumptions remain untested? |
| Full cost | What are the initial, recurring, internal, and exit costs? |
| Implementation and fit | How long might implementation take, and how well does the option fit the workflow? |
| Risk and mitigation | What could go wrong, and what controls or review effort would be needed? |
| Reversibility | How easily can the organization stop, switch, or redesign the approach? |
| Evaluation quality | Can the proposed evaluation credibly establish whether the option achieved its intended outcomes? |
OECD advises public administrations to plan and monitor AI investments for value for money, risk management, timely implementation, and realization of intended benefits. Its 2025 figures—88% of OECD countries had a standardized approach to developing digital-government value propositions, while 41% had developed a risk-assessment mechanism for digital-government investments—describe government investment practices, not individual AI project success or ROI. The source is OECD’s 2025 publication.
6. Assess risks, controls, and accountability
Map who may be affected, how the system will be used, and what could go wrong in the data, model, interface, workflow, or supplier relationship. Consider privacy, security, reliability, bias or disparate impact, explainability, human oversight, misuse, and service continuity as applicable. Identify an accountable owner, reviewers, an escalation route, a monitoring plan, and conditions for pausing or stopping.
Rank #4
NIST’s voluntary AI Risk Management Framework organizes risk work under four functions: Govern, Map, Measure, and Manage. It encourages considering trustworthiness throughout design, development, deployment, use, and testing or evaluation. NIST states that the framework is for voluntary use; it does not replace applicable law. Use the NIST AI RMF Playbook and NIST FAQs as structuring resources, and check for current revisions when applying them.
For enterprise due diligence, the OECD’s 2026 Responsible AI Due Diligence Guidance offers a risk-based process to identify and address potential adverse impacts across relevant activities and business relationships. These frameworks help organize analysis; determine legal obligations separately for the deployment’s geography, sector, and use case.
7. Set decision gates and plan evaluation before launch
Decide in advance what evidence permits expansion, what should trigger redesign, and what means stopping. Plan how outcomes and unintended effects will be tracked, who will evaluate them, and when the investment case will be revisited as evidence accumulates.
HM Treasury’s Guidance on the Impact Evaluation of AI Interventions, updated 15 May 2026, frames impact evaluation as a systematic assessment of whether, to what extent, how, and why an intervention produced its intended impacts. It is written for central government and public services; its comparator and evaluation principles can inform other settings when appropriately adapted.
Quick Recap
A reusable business-case outline
- Decision requested: pilot, buy, build, scale, defer, or stop.
- Problem and affected workflow: who experiences it, the current process, and the baseline.
- Why AI may help: the proposed mechanism and why simpler alternatives may be less suitable.
- Options and comparator: business as usual plus credible non-AI and AI options.
- Target outcomes: measures, baseline, target, time window, and data owner.
- Feasibility evidence: data readiness, pilot design, assumptions, and limitations.
- Benefits: financial and non-financial outcomes, attribution, and confidence level.
- Costs: one-time, recurring, internal effort, human review, operations, and exit costs.
- Risks and controls: affected people, accountable owner, mitigations, review, and stop conditions.
- Evaluation plan: comparator, method, timing, metrics, and responsible evaluator.
- Decision gates: evidence needed to proceed, change course, or stop.
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.
