AI business case
An AI business case is the evidence used to decide whether an AI initiative should be funded, tested, expanded, redesigned, or stopped. It connects a defined business problem with available options, expected benefits, full costs, delivery requirements, risks, and a plan for measuring results.
A useful business case does more than estimate AI ROI. It explains why the organization should act, why the proposed workflow is a better target than other options, and what evidence will determine whether the investment continues.
What should an AI business case include?
Start with the decision and the workflow. Name the current problem, the people and systems involved, the accepted unit of work, and the outcome the initiative is expected to change.
Then compare credible options. These may include improving the existing process without AI, buying a tool, building an internal system, changing the workflow, or doing nothing. The business case should explain the costs, benefits, constraints, and risks of each option instead of treating AI as the predetermined answer.
HM Treasury's Green Book defines appraisal as an assessment of the costs, benefits, and risks of different options. It also requires a business-as-usual option as the benchmark for comparison (HM Treasury Green Book). That principle translates well to enterprise AI decisions.
How should costs and benefits be estimated?
Use the AI total cost of ownership rather than license price alone. Include implementation, model usage, infrastructure, integration, security review, training, human review, maintenance, and internal labor where they are material.
Separate benefit types. Faster work, higher accepted throughput, better quality, lower unit cost, avoided spending, and incremental revenue require different evidence. Do not convert reported time savings into cash unless the operating plan shows how the organization will use or remove that capacity.
Document every important assumption, its source, and its confidence range. NIST's AI Risk Management Framework calls for expected benefits and costs to be examined against appropriate benchmarks, with intended scope and human oversight documented (NIST AI RMF Core).
Why does baseline measurement belong in the business case?
Baseline measurement records how the workflow performs before the intervention. Without it, a later improvement cannot be compared fairly with prior performance.
The business case should define the baseline period, eligible work, metric formulas, source systems, quality guardrails, and known factors that may affect the comparison. It should also specify the measurement window and decision threshold before results are available.
This turns evaluation into part of the investment design. It also prevents teams from selecting favorable metrics after a pilot has finished.
How does the business case connect to realized value?
A forecast is not an outcome. The business case should assign owners for each expected benefit and describe the operational changes needed to capture it. That plan becomes the basis for benefits realization.
After implementation, compare actual costs and outcomes with the approved case. Realized AI value is the value the organization has captured in operations, not the value predicted in a spreadsheet or observed temporarily during a pilot.
Microsoft's guidance on measuring agent impact recommends connecting adoption and operational indicators to business outcomes through explicit value drivers (Microsoft guidance). Adoption can help explain a result, but it is not the benefit itself.
Common AI business-case mistakes
Common failures include starting with a vendor rather than a workflow, comparing AI only with the current process, omitting internal effort and review costs, counting all saved time as money, and presenting one ROI estimate without uncertainty.
Another mistake is approving the investment without an evaluation and exit plan. A decision-ready AI business case states what evidence will support expansion, redesign, renewal, or cancellation. That makes the case a living decision record rather than a one-time funding document.