Module 1: Building the AI Business Case · AI for Enterprise Leaders

Six months after approving a $2 million AI pilot, your CFO walks into the quarterly review and asks one question: "What did we actually get?" If you cannot answer that question before you start the project, you will not be able to answer it after. This module gives you the four-part framework that makes AI business cases credible, not just compelling.

Why Most AI Business Cases Fail

Most AI business cases fail not because the technology does not work, but because the case was built backward. A leader sees a compelling demo, estimates a rough number, and then works backward to make that number fit a slide. The CFO sees through this immediately. The project gets approved on enthusiasm, the ROI is never tracked, and six months later the same leader is explaining why the original estimate was "aspirational."

A credible AI business case is built forward: start with a specific problem, form a measurable value hypothesis, model the real costs, and register the real risks. This sequence forces you to confront uncomfortable questions before you spend money, not after.

The one-sentence test If you cannot complete this sentence, you do not have a business case yet: "We will use AI to do X, which will reduce/increase Y by Z, which we will measure with metric W, and it will cost $A with a payback period of B months."

The Four Business Case Components

Component 1

Problem Definition

A specific, bounded problem with a measurable current state. "Our contract review team takes an average of 4.2 days per contract" is a problem definition. "We want to use AI for legal" is not.

Component 2

Value Hypothesis

A falsifiable prediction of what will change. "AI-assisted review will reduce average contract cycle time to 1.5 days" is a value hypothesis. It can be tested and proven or disproven.

Component 3

Cost Model

Every cost item across five categories: compute, data preparation, talent, integration with existing systems, and change management. Most business cases omit the last two and then overspend on them.

Component 4

Risk Register

A named list of the risks that could prevent value realization, with a likelihood and impact score for each. Without a risk register, surprises are guaranteed. With one, they become manageable.

Quantifying the Value Hypothesis

AI value falls into four categories. Each requires a different measurement approach, and each has a different credibility challenge with finance teams.

Productivity uplift: Time saved per task multiplied by volume multiplied by fully-loaded cost per hour. If your contract team handles 400 contracts per month and AI reduces review time from 4.2 days to 1.5 days, that is 1,080 lawyer-hours per month recovered. At a fully-loaded rate of $150 per hour, that is $162,000 per month in recoverable capacity. Finance will ask whether that capacity actually translates to revenue or whether it disappears into other work. You need an answer.

Error reduction: Current error rate multiplied by cost per error multiplied by volume. If 3% of invoices contain errors and each error costs $800 to remediate across 10,000 monthly invoices, you are spending $240,000 per month on invoice error remediation. AI that reduces the error rate to 0.5% eliminates $200,000 in monthly remediation cost. This is the most credible value category for finance teams because the cost of errors is already on the books.

Speed: Reduced time-to-market or time-to-decision multiplied by the value of time. A pharmaceutical company that reduces clinical trial protocol review from 6 weeks to 2 weeks may accelerate product launch by a meaningful margin. The value of that time is calculable, but requires assumptions that need to be made explicit.

Revenue enablement: New revenue that is only possible because of AI capability. This is the hardest category to defend because it requires attribution assumptions. State the assumptions explicitly and use conservative estimates. Finance teams will cut your number in half anyway; start with the honest version.

Building the Cost Model

Most AI cost models include compute and maybe data. They miss the three categories that cause projects to overspend: talent, integration, and change management.

Compute: API costs or infrastructure costs for inference. These scale with usage and are often the easiest to model. Start with a cost-per-query estimate and multiply by projected volume. Add a 40% buffer for usage spikes during the first six months.

Data: Data preparation, cleaning, labeling, and ongoing data quality maintenance. This is often 30-40% of the total project cost and is consistently underestimated. Get an estimate from the team that will actually do the work, not from the vendor.

Talent: ML engineers, data scientists, prompt engineers, and AI product managers. Include both internal hire costs and external consulting costs. Also include the cost of retraining existing staff who will work alongside the AI system.

Integration: Connecting the AI system to existing data sources, workflows, and enterprise systems. Integration is almost always more expensive and slower than the AI component itself. Budget twice what the vendor estimates.

Change management: Communication, training, process redesign, and the productivity dip that occurs when people learn a new way of working. This is the category most likely to determine whether value is actually realized. Under-invest in change management and you will have an AI system that nobody uses.

Common mistake The change management budget is not the training budget. Training is a one-time event. Change management is an ongoing investment in helping people adopt new workflows, address concerns, celebrate early wins, and sustain new behaviors. They are different categories with different cost structures.

The 3-Horizon Framework

Not every AI initiative has the same time horizon for value delivery, and mixing horizons in a single business case creates confusion about what success looks like in year one versus year three.

Horizon 1 (0-6 months): Quick wins. Narrow, well-defined problems with measurable current states. High confidence in value delivery. Examples: document classification, FAQ automation, data extraction. These should show positive ROI within the first six months.

Horizon 2 (6-18 months): Scale. Expanding proven solutions across more users or more use cases, or building capabilities that enable Horizon 3. Medium complexity and medium confidence. Examples: expanding a document review pilot from one team to the enterprise, or building a data platform that future AI initiatives will depend on.

Horizon 3 (18-36 months): Transform. Fundamental changes to how the business operates. High complexity, high potential value, lower confidence because the technology and your organization's capability are both still developing. Examples: autonomous decision-making in underwriting, AI-native customer journeys, fully automated supply chain optimization.

A strong business case has initiatives in all three horizons, with Horizon 1 wins funding the capability building required for Horizon 3.

The ROI Calculator

Use the interactive below to model the annual value of an AI productivity initiative. Adjust the team size, weekly hours saved per person, and fully-loaded cost. The calculator shows projected annual value and an estimated payback period based on a simplified cost assumption.

Python · ROI Calculation

# AI ROI Business Case Calculation
# Every formula made explicit so finance teams can verify

def calculate_ai_roi(
    team_size: int,
    hours_saved_per_week: float,
    hourly_cost: float,
    implementation_cost: float,
    annual_run_cost: float,
    weeks_per_year: int = 48  # exclude holidays and ramp-up
) -> dict:
    """
    Calculate simple AI ROI for a productivity initiative.
    All numbers should be fully-loaded (benefits + overhead).
    """

    # Annual gross value: time recovered at fully-loaded cost
    weekly_value = team_size * hours_saved_per_week * hourly_cost
    annual_gross_value = weekly_value * weeks_per_year

    # Total first-year cost: implementation + annual run
    total_year1_cost = implementation_cost + annual_run_cost

    # Net first-year value
    net_year1_value = annual_gross_value - total_year1_cost

    # Payback period in months
    monthly_value = annual_gross_value / 12
    if monthly_value > 0:
        payback_months = implementation_cost / monthly_value
    else:
        payback_months = float('inf')

    # 3-year NPV (simplified, 10% discount rate)
    discount_rate = 0.10
    npv = -implementation_cost
    for year in range(1, 4):
        net_annual = annual_gross_value - annual_run_cost
        npv += net_annual / ((1 + discount_rate) ** year)

    return {
        'annual_gross_value': annual_gross_value,
        'total_year1_cost': total_year1_cost,
        'net_year1_value': net_year1_value,
        'payback_months': round(payback_months, 1),
        '3yr_npv': round(npv),
        'roi_year1_pct': round((net_year1_value / total_year1_cost) * 100, 1)
    }

# Example: 50-person team, 4 hrs/week saved, $100/hr
result = calculate_ai_roi(
    team_size=50,
    hours_saved_per_week=4,
    hourly_cost=100,
    implementation_cost=200_000,
    annual_run_cost=60_000
)
# annual_gross_value: $960,000
# payback_months: 2.5 months

Try this

Pick one AI initiative your organization is considering or has discussed in the last six months. Write out the four business case components: (1) the specific problem in one sentence with a measurable current state, (2) the value hypothesis with a specific number you predict will change, (3) a rough cost model across the five categories, and (4) the top three risks. Doing this exercise will reveal whether the initiative is a Horizon 1, 2, or 3 play, and whether the current thinking is ready for a CFO conversation.