What is AI opportunity sizing?

Start with the workflow and decision. Identify the business problem, owner, eligible population, completed unit of work, current volume, baseline outcome, and material constraints.

Then estimate a range for the change. Relevant inputs may include:

• –eligible work volume and expected adoption;
• –baseline cycle time, quality, throughput, or unit cost;
• –plausible improvement supported by tests or comparable evidence;
• –implementation and recurring cost;
• –correction, review, exception, and rework effort;
• –benefit timing and duration; and
• –uncertainty, risk, and dependencies.

Keep observed, externally benchmarked, assumed, and unknown inputs labeled. One precise total can hide weak evidence. Define the metrics for each input in an AI metric dictionary.

The calculation depends on the benefit type. A general operating structure is:

Addressable improvement = eligible volume x plausible improvement per unit x expected adoption

Then compare the supported benefit range with the fully loaded cost and relevant disbenefits. AI total cost of ownership describes the lifecycle cost boundary, and AI ROI compares net monetized benefit with total investment. Do not monetize quality, risk, or time automatically. Preserve those outcomes in their original units unless an approved method converts them.

Use scenarios rather than one forecast when uncertainty is material. A conservative case can use lower adoption and improvement with higher cost. An upside case should remain grounded in credible evidence rather than the maximum technically possible result.

HM Treasury's Green Book recommends comparing options, including business as usual, across costs, benefits, risks, and uncertainty rather than relying on one summary metric (HM Treasury Green Book).

Opportunity sizing is an early screen. It asks whether the workflow deserves deeper analysis or a bounded test. An AI business case compares options in greater detail and covers affordability, delivery, governance, procurement, risk, and evaluation.

Use the AI initiative inventory to check for related pilots, duplicate tools, shared dependencies, and historical evidence. A proposed opportunity may be smaller when another initiative already addresses the same workflow.

Translate the sizing assumptions into AI pilot success criteria. Test the most uncertain and decision-relevant inputs, such as eligible volume, accepted-output improvement, review burden, quality, adoption, or cost.

After the pilot, replace assumptions with observed AI value where the data supports it. Keep attribution and realization limits visible. A pilot result from one team or case type may not generalize to enterprise volume.

NIST's AI Risk Management Framework calls for intended scope, expected benefits and costs, benchmarks, human oversight, and deployment conditions to be documented (NIST AI RMF Core).

Compare workflows using consistent categories, not a single ROI score. Consider strategic relevance, addressable volume, evidence confidence, time to useful evidence, data readiness, implementation effort, quality and risk, total cost, and reversibility.

An opportunity with moderate value and strong measurement readiness may deserve priority over a larger estimate built on unavailable data and uncertain adoption. An AI measurement readiness assessment tests that readiness.

Common failures include using total organizational volume instead of eligible volume, assuming full adoption, applying vendor productivity claims without context, excluding review and implementation cost, and treating all saved time as cash.

Another mistake is ignoring the do-nothing or process-only option. The best response to a workflow problem may be simplification, automation without AI, or no investment.

1. 01HM Treasury Green BookGOV.UKIndependent
2. 02NIST AI RMF CoreNISTIndependent