# Build an AI business case that survives scrutiny

AI-assisted editorial draft • Bill Allred review pending • Draft and sources checked 19 October 2026

*All amounts and operating assumptions below are hypothetical, not prices, forecasts, or client results.*

## Executive answer

A defensible AI business case explains what changes operationally, which benefits can actually be realized, what delivering them costs, and what evidence could invalidate the proposal. Keep cash savings, released capacity, additional contribution, service improvement, and working-capital effects separate. Count each benefit once.

A strong case can justify an investment through growth or service even when payroll does not fall. It should also make a disappointing result visible before the organization commits to scale.

## Start with the mechanism of value

Use a chain that another person can challenge:

**Changed activity → changed operating result → realization action → financial or service effect.**

For example: shorter preparation creates reviewer capacity; that capacity supports additional orders; verified demand and fulfillment capacity permit those orders; the orders earn incremental contribution after variable costs. Each arrow contains an assumption.

Avoid multiplying all hours saved by a salary rate and calling the result cash. If the same employees remain on the same payroll, the immediate benefit is capacity. Cash savings require an identifiable reduction in spending, such as avoided external work, and a credible decision to realize it. If the hours instead support growth, count the resulting contribution, without adding the same hours again as a labor saving.

Financial statements distinguish profit from cash flows. The SEC’s educational guide explains that income statements report earnings and expenses while cash-flow statements track cash moving into and out of the business. This guide uses simplified decision economics; finance should determine the appropriate accounting and cash-flow treatment. [Source: SEC](https://www.sec.gov/about/reports-publications/investorpubsbegfinstmtguide)

## Worked example: capacity converted into growth

**Hypothetical illustration, USD, annual steady state unless stated otherwise.** A service operation handles 12,000 eligible requests a year. AI-assisted preparation reduces total human handling time, including review and correction, from 18 to 12 minutes.

- Capacity released: 12,000 × 6 minutes ÷ 60 = **1,200 hours**.
- At an illustrative $40 hourly loaded cost, this is **$48,000 of capacity value**, reported separately and excluded from cash benefits.
- Verified incremental demand is assumed to be 1,500 requests, each requiring 12 minutes. Serving it consumes **300 hours** of the released capacity.
- At $50 revenue and $20 incremental variable cost per additional request, contribution is **$30 per request**, or **$45,000**. Existing payroll is unchanged; the variable-cost assumption must cover all additional fulfillment costs.
- A separate external administration expense falls by **$12,000**, supported by a planned scope reduction. Assume it covers different work and does not overlap the capacity calculation.
- Annual recurring technology, support, evaluation, and operating costs are **$24,000**.
- One-time implementation, integration, training, and validation costs are **$30,000**.

Recurring net benefit is $45,000 + $12,000 − $24,000 = **$33,000**. A full year at steady state, less one-time costs, yields **$3,000** before tax, financing, discounting, and any other excluded effects. That is a scenario, not a first-year forecast: rollout delays or gradual adoption would reduce first-year benefits.

The remaining 900 released hours are unmonetized capacity. Do not add their salary equivalent to the $33,000. Management must decide whether they improve service, absorb variability, or support another activity.

## Test the assumption most likely to fail

Hold other assumptions fixed and vary incremental demand:

| Additional requests | Incremental contribution | Recurring net benefit | Net after $30,000 setup* |
|---:|---:|---:|---:|
| 500 | $15,000 | $3,000 | −$27,000 |
| 1,500 | $45,000 | $33,000 | $3,000 |
| 2,500 | $75,000 | $63,000 | $33,000 |

*Full-year steady-state scenario, not a rollout forecast.*

Recurring operating break-even requires 400 additional requests: ($24,000 − $12,000) ÷ $30. Recovering setup within the same full steady-state year requires 1,400: ($24,000 + $30,000 − $12,000) ÷ $30. These thresholds assume the separate $12,000 saving occurs and the operation can fulfill demand at the assumed margin.

Also stress review effort, adoption, errors, vendor costs, and downstream capacity. A favorable demand scenario cannot rescue an unworkable operating process.

## Reusable assumptions worksheet

Copy one row for every benefit and cost. Mark each assumption as measured, quoted, estimated, or unknown; attach its source and owner.

| Line | Required evidence | Avoid this mistake |
|---|---|---|
| Cash savings | Spend baseline and realization action | Counting unchanged payroll |
| Released capacity | Net time after review and rework | Counting gross drafting speed |
| Additional contribution | Demand and incremental margin | Treating revenue as profit |
| Service or quality | Baseline, target, affected customers | Inventing a dollar conversion |
| One-time costs | Integration, training, validation, transition | Omitting internal effort |
| Recurring costs | Usage, support, monitoring, maintenance | Using model fees alone |
| Working capital | Timing and cash-balance assumptions | Calling a one-time release annual profit |
| Uncertainty | Downside scenario and stopping condition | Reporting one precise ROI |

## Measurement and limits

Before testing, specify eligible cases, baseline period, comparison method, exclusions, and who verifies outcomes. Compare end-to-end handling and completed work, including correction and escalation. Separate adoption problems from model performance; record both.

This is an illustrative management method, not accounting, investment, or tax advice. It omits financing, discounting, tax, and deployment ramp effects. Use your finance team’s investment conventions. Approve the next stage only when the evidence supports its assumptions, and connect the case to [a defined workflow](https://altiport.ai/resources/redesign-workflow-before-choosing-tool/).