Overview

A supply chain operations director proposed a demand forecasting AI system with expected savings of "$4 million annually." The CFO asked for details: what specifically would save money? How was the $4M calculated? What were the costs? The director realized he'd made assumptions about AI's capabilities without actually calculating the business impact. He couldn't explain his number to the CFO, which meant he couldn't sell the investment.

A credible business case doesn't require perfect accuracy. It requires transparent methodology that leadership can understand and challenge. The best cases explain not just the expected return, but how you calculated it and what assumptions could be wrong.

Building Your AI Investment Cost Model

Every AI investment has multiple cost categories. Create a detailed cost model that accounts for them all.

Software Licensing Costs (Year 1 and ongoing):

Software costs include the AI platform itself, integration tools, and supporting software.

• AI platform licensing: Most platforms charge per user, per transaction, or as fixed annual fee. Typical range: $50K-$500K annually depending on scale
  - Integration software: APIs, middleware, data pipelines might require ETL tools or iPaaS platforms. Typical: $20K-$100K annually
  - Data infrastructure: Cloud storage, compute, databases. Typical: $30K-$200K annually depending on data volume
  - Supporting tools: Version control, monitoring, security. Typical: $10K-$50K annually

Build a three-year model showing licensing costs year by year, accounting for growth in users or transaction volume.

Implementation Services (typically Year 1):

Implementation costs include professional services, consulting, and custom development.

• Requirements and design: $30K-$75K (1-2 months consulting)
  - Model development and training: $50K-$150K (building the AI models)
  - System integration: $40K-$120K (connecting to your systems)
  - Testing and validation: $20K-$50K (ensuring it works correctly)
  - Pilot execution: $10K-$30K (small-scale test before full rollout)

These are largely one-time costs, with some ongoing maintenance work in years 2-3.

Internal Labor Costs (Years 1-3):

Your team's time implementing and supporting AI:

• Project management: 0.5 FTE for 12 months = $75K
  - Business analyst: 1 FTE for 6 months planning + 0.5 FTE> - Data engineer: 1 FTE for data work = $120K
  - Training and change management: 0.5 FTE for 6 months = $50K
  - Ongoing operations/support: 0.25 FTE Year 2+ = $30K annually

Total Year 1 internal labor typically $250K-$400K depending on team size and seniority.

Infrastructure Costs:

Physical infrastructure for the AI system:

• Servers/cloud: $30K-$100K annually (depending on compute needs)
  - Network upgrades: $10K-$50K one-time
  - Security controls: $20K-$60K (licensing + configuration)
  - Disaster recovery: $5K-$15K annually
  - Monitoring and alerting: $10K-$20K annually

Training and Change Management:

Helping people work with the new system:

• Curriculum development: $10K-$30K
  - Training delivery (instructor-led or online): $20K-$50K
  - Train-the-trainer program: $10K-$20K
  - Change management: $30K-$60K (communications, resistance management)

Ongoing Operations (Years 2-3+):

Maintaining and improving the system:

• License renewals: Full cost each year
  - Infrastructure: Continued hosting, security, backups
  - Support and maintenance: Bug fixes, updates, 0.2-0.3 FTE ongoing
  - Model retraining: Periodic updates as business context changes
  - Governance and compliance: Ensuring system is working as intended

Critical insight: Most organizations underestimate internal labor and change management costs by 40-60%. Software and consulting are visible costs. Your team's time and the work to change behavior are less visible but often exceed software costs. Build this explicitly into your model.

Quantifying AI Investment Benefits

Benefits are harder to quantify than costs because they depend on adoption and execution. But without quantification, you can't defend your investment.

Time Savings Benefits:

The most common AI benefit is reducing time on manual work.

Identify the manual process: "Analysts spend 15 hours per week reviewing reports and extracting key insights for executive summary."

Calculate baseline cost: 15 hours/week × 52 weeks × $40/hour labor cost = $31,200 annually

Estimate AI impact: "AI automation will reduce this to 2 hours per week (for review and refinement)."

Calculate benefit: 13 hours/week saved × 52 weeks × $40/hour = $26,960 annually

Be conservative on adoption. If you predict 100% adoption, plan for 60-80%. If you predict 80% benefit realization (some people still do things the old way), your actual benefit is $26,960 × 0.70 = $18,900 annually.

Error Reduction Benefits:

AI often improves accuracy compared to human decisions.

Identify the error type: "Customer service reps misclassify ticket urgency 12% of the time, causing escalation delays."

Calculate current error cost: 12% × 5,000 tickets/year × $150 cost per escalation = $90,000 annually

Estimate AI improvement: "AI will reduce misclassification to 3%."

Calculate benefit: 9% × 5,000 tickets × $150 = $67,500 annually

Again, be conservative on adoption. Reps might distrust AI accuracy initially. Plan for 60-70% adoption of AI recommendations.

Capacity Gains:

AI can help you handle more volume without more headcount.

Current state: Team processes 1,000 contracts/month with 5 people = 200 contracts per person per month

AI assistance: "AI-assisted contract review reduces time by 40%."

Benefit calculation:
- Same headcount can now process 333 contracts per month
- Increased volume handling = 1,667 contracts/month vs 1,000
- At $200 per contract processed = $133,800 additional revenue annually

Or: Maintain current volume with fewer people. 1,000 contracts ÷ 333 per person = 3 people instead of 5 people = $120K in labor cost savings.

Quality Improvements:

Better decisions create long-term competitive advantage but are hard to quantify short-term.

"Demand forecasting AI reduces forecast error by 15%, which reduces excess inventory."

Current excess inventory carrying cost: $2M in excess inventory × 15% carrying cost = $300K annually

Improvement: 15% × $300K = $45K annual savings

These are real but smaller than time savings and error reductions.

Building Your Business Case Model

Create a spreadsheet showing:

Year 1:
- All costs (software, implementation, labor, infrastructure, training)
- All benefits (time savings, error reduction, capacity gains)
- Net Year 1 impact (typically negative due to upfront costs)

Year 2:
- Reduced costs (no implementation services, no training, but ongoing support)
- Higher benefits (adoption improves as team gets comfortable)
- Net Year 2 impact (typically breakeven or small positive)

Year 3:
- Ongoing costs only
- Stable or growing benefits
- Net impact shows strong positive returns

ROI calculation:

ROI = (Total 3-Year Benefit - Total 3-Year Cost) / Total 3-Year Cost × 100%

Example:
- Year 1: -$350K (costs exceed benefits)
- Year 2: +$80K (benefits begin exceeding ongoing costs)
- Year 3: +$120K (strong ongoing returns)
- Total 3-year: -$350K + $80K + $120K = -$150K

Wait, that's negative ROI! This is why "quick wins" approach works better than betting on year 3 payoff. But recalculate if adoption happens faster:

If adoption reaches 80% by month 6 instead of month 12:
- Year 1: -$200K
- Year 2: +$150K
- Year 3: +$150K
- Total 3-year: +$100K

That's breakeven at 2.5 years, with positive trajectory for years 4-5.

Payback Period:

How many months until cumulative benefit exceeds cumulative cost?

Month 1-8: Negative (implementation costs exceed early benefits)
Month 9-14: Approaching breakeven
Month 15-18: Positive
Payback period: 18 months

Sensitivity analysis shortcut: Show best case (90% adoption), base case (70% adoption), and worst case (50% adoption). This demonstrates you understand risks and creates credibility with leadership. The base case becomes your commitment.

Handling Uncertain Benefits

Many AI benefits are uncertain. You might save 30-50% of the time, not a fixed amount. How do you account for uncertainty?

Use probability weighting:

"We estimate 70% probability of 40% time savings, 20% probability of 30% time savings, and 10% probability of no benefit realization."

Expected benefit = (0.70 × 40%) + (0.20 × 30%) + (0.10 × 0%) = 28% + 6% + 0% = 34% expected benefit

This is more honest than assuming 40% benefit will definitely happen.

For the business case, use the expected value calculation. This acknowledges uncertainty while giving a clear number for financial planning.

The Business Case Document

Your business case becomes your commitment document to leadership.

Include:
1. Executive summary: Investment amount, expected ROI, payback period
2. Problem statement: What's the current pain? What does it cost us?
3. Proposed solution: High-level description of the AI system
4. Cost analysis: Detailed cost model for years 1-3 with assumptions stated
5. Benefit analysis: How we calculated each benefit with adoption assumptions
6. Sensitivity analysis: Best case, base case, worst case scenarios
7. Risk analysis: What could go wrong and how we'll mitigate it
8. Success metrics: How we'll measure if the investment is working
9. Timeline: When costs occur and when benefits begin
10. Recommendation: Clear recommendation to proceed or not

The document is dense but defensible. When someone asks "why should we spend $400K on this?", you have a data-driven answer.

Real Business Case Example**

Here's a complete business case for an invoice automation project:

Investment: Invoice Processing Automation

Executive Summary:

• Investment: $320K (Years 1-3)
  - Expected Benefits: $420K (Year 1 conservative), increasing to $800K annually by Year 3
  - Payback Period: 14 months
  - 3-Year ROI: 62%
  - Recommendation: Proceed immediately

Year 1 Cost Breakdown:

• AI Platform licensing: $60K (per-transaction pricing)
  - Implementation services: $80K (requirements, setup, testing)
  - Internal labor: $100K (project management 0.5 FTE, data prep 1 FTE, training delivery)
  - Infrastructure: $30K (cloud setup, security, backup)
  - Change management: $30K (communications, training materials, support)
  - Total Year 1: $300K

Year 2-3 Costs:

• Licensing: $60K annually
  - Support: 0.2 FTE = $25K annually
  - Infrastructure and maintenance: $15K annually
  - Year 2-3 total: $100K annually

Year 1 Benefits Calculation:

Current state: 5 AP analysts process invoices. Each analyst spends 30% of their time on data extraction and validation.

• Current cost: 5 analysts × $80K salary × 30% time = $120K annually (40% adoption = $48K realized)
  - AI impact: Reduces extraction time from 6 hours to 1 hour per invoice (invoice complexity varies)
  - Volume: 10,000 invoices annually
  - Time saved: 50,000 hours saved × $40/hour labor = $2M potential benefit
  - With conservative 20% realization (only 2,000 invoices automated in Year 1 as adoption ramps): $400K benefit

Year 2-3 Benefits:

• Year 2: 80% adoption = $1.2M benefit (plus $50K error reduction from fewer manual errors)
  - Year 3: 85% adoption = $1.3M benefit (plus $100K from upstream process improvements enabled)

Cash Flow Summary:

• Year 1: -$300K costs + $400K benefits = $100K net positive
  - Year 2: -$100K costs + $1.25M benefits = $1.15M net positive
  - Year 3: -$100K costs + $1.4M benefits = $1.3M net positive
  - 3-Year Total: $2.45M benefits - $500K costs = $1.95M net benefit
  - ROI: ($1.95M / $500K) × 100% = 390% over 3 years

Payback Analysis:

• Month 1-6: Implement, ramp adoption to 10%. Benefits: $40K. Costs: $150K. Cumulative: -$110K
  - Month 7-12: Adoption reaches 40%, benefits accelerate. Benefit: $300K. Costs: $150K. Cumulative: +$40K
  - Payback: Month 14 (when cumulative benefit exceeds cumulative costs of $300K)

This example shows realistic assumptions (conservative adoption ramp), transparency (here's how we calculated each number), and credibility (financial planning leadership can trust).

Common Business Case Mistakes**

Mistake 1: Over-Optimistic Adoption**

You assume 100% of invoices will be processed by AI immediately. Reality: adoption ramps slowly. Ramp adoption conservatively (10% month 1, 20% month 3, 40% month 6, 70% month 12). This matches real human behavior change patterns.

Mistake 2: Ignoring Downtime During Implementation**

You estimate benefits starting month 1. Reality: Project takes 3 months to implement, test, and train. Benefits start month 4. Factor in ramp time for realistic projections.

Mistake 3: Only Counting Direct Benefits**

Time savings of $400K is good. But also count: error reduction from fewer manual mistakes, capacity freed (can AP team handle more volume?), process improvement enabled (cleaner data improves upstream accounting?). Comprehensive benefits are often 20-30% higher than just time savings.

Mistake 4: No Sensitivity Analysis**

You show one ROI number: 50%. Leadership asks "what if adoption is lower?" and you have no answer. Always show three scenarios: best case (90% adoption), base case (70%), worst case (50%). This builds credibility.

Mistake 5: Treating Business Case as "Set It and Forget It"**

You finalize the business case, get approval, and never update it. Six months in, actual costs are 20% higher and adoption is slower than planned. You should have updated the case to reality and repriced. Keep the business case alive. Update it monthly, especially in year 1.

Defending Your Business Case**

Leadership will challenge your assumptions. Here's how to defend them:

Assumption: Adoption reaches 70% by month 12**

Challenge: "How do you know adoption will be that high?"

Defense: "We have three data points: (1) Similar automation projects we've done had 65-75% adoption in first 12 months. (2) We're building in change management (training, support, early wins), which accelerates adoption. (3) We have a rollback plan if adoption stalls." This shows evidence-based assumptions, not wishful thinking.

Assumption: Time savings will be $400K in Year 1**

Challenge: "How do you know time will be freed and not just lost?"

Defense: "We're being intentional about how freed time is used. (1) We have a backlog of high-priority work (month-end close improvements) that AP team will transition to. (2) AP manager is committed to measuring time freed and reassigning it to strategic work. (3) We're tracking time savings as a success metric." This shows you're not just assuming time magically becomes productive. You're managing the transition.

Assumption: Software costs $60K annually**

Challenge: "Is that market rate? Have you compared vendors?"

Defense: "Yes, we evaluated three platforms. Platform A is $45K but has limited customization. Platform B is $75K but more robust. We selected Platform C at $60K as the value-for-money choice. Here's the comparison matrix." This shows due diligence, not random estimates.

Deliverable: AI Investment Business Case Document**

Your business case document is presented to leadership for approval to proceed.

The document includes:

• Executive summary: Investment amount, expected ROI, payback period, clear recommendation
  - Problem statement: What's the current pain? What does inefficiency cost us? Why now?
  - Proposed solution: High-level description of the AI system, how it solves the problem
  - Cost analysis: Detailed cost model for years 1-3 with assumptions stated explicitly (can leadership challenge each assumption)
  - Benefit analysis: How we calculated each benefit with adoption assumptions and evidence for those assumptions
  - Sensitivity analysis: Best case (90% adoption), base case (70%), worst case (50%) with clear ROI for each
  - Risk analysis: What could go wrong (adoption slower than expected, benefits lower, costs higher) and how we'll mitigate
  - Success metrics: How we'll measure if the investment is working (adoption rate, time saved realized, adoption curve)
  - Timeline: Gantt chart showing when costs occur, when benefits begin, when milestones are achieved
  - Recommendation: Clear, confident recommendation to proceed with conditions (if conditions exist)

The document is dense (10-15 pages) but defensible. Every number has a source. Every assumption can be challenged. Leadership walks away confident they understand what they're approving.

What to Do Monday Morning**

• List all cost categories for your planned AI investment (software, implementation, labor, infrastructure, training, change management).
  - Get actual vendor estimates for software and implementation services. Don't guess.
  - Calculate internal labor costs using real team members' salaries and estimated time commitment.
  - Identify the manual process that will be automated and measure its current cost (time spent × hourly rate).
  - Research industry benchmarks for similar AI implementations to validate your benefit assumptions.
  - Estimate AI impact percentage conservatively (time reduction, error reduction). If you think it will save 50%, budget for 35% to be conservative.
  - Model adoption conservatively (ramp slowly, don't assume 100% immediately). Real adoption curves are 10%-20%-40%-70% over 12 months.
  - Build 3-year cash flow model showing costs and benefits by month for Year 1, by quarter for Years 2-3.
  - Calculate ROI and payback period. If payback exceeds 24 months, you need a compelling strategic reason to proceed.
  - Create sensitivity analysis showing best/base/worst case scenarios. Use the base case as your commitment to leadership.

Key Takeaways**

• Quantify both costs and benefits. Costs are easy; benefits require disciplined estimation and research.
  - Be conservative on adoption rates. Plans assuming 100% adoption immediately always disappoint. Model slow adoption ramps (10%-70% over 12 months).
  - Model uncertainty explicitly. Use probability weighting for uncertain benefits or sensitivity analysis showing best/base/worst cases.
  - Show your work. Document how you arrived at every cost and benefit number. Leadership can challenge your logic, but they'll respect transparency.
  - Include all cost categories. Internal labor and change management often represent 50% of total cost but are easy to underestimate.
  - Keep business case alive. Update it monthly in year 1, comparing actual costs and benefits to projections. Business case is a living document, not a static artifact.
  - Target reasonable ROI thresholds. Strategic AI investments should aim for 30%+ ROI by year 3. Quick-win projects should aim for 50%+ in year 1. Below 20% ROI, the investment is usually not justified.

FAQs**

Q: Should we include "soft benefits" like improved morale or brand perception?**

A: Mention them in the narrative but don't rely on them for ROI justification. Soft benefits are real but nearly impossible to quantify. Use them to explain why the financial case is even better than the numbers show. Example: "This project saves $400K annually (hard benefit) PLUS improves team satisfaction and reduces turnover (soft benefit worth additional $50K annually in hiring costs avoided, though we're not including in ROI calculation)."

Q: What if we can't quantify benefits because we haven't done AI before?**

A: Use industry benchmarks. Demand forecasting projects typically improve accuracy 10-20%. Invoice automation projects typically reduce processing time 60-80%. Process automation projects typically reduce cycle time 40-60%. Search for published case studies from similar organizations and use their results as your baseline. Then adjust down 20% for conservative estimation.

Q: How do we handle intangible benefits like "faster decision-making"?**

A: Quantify where possible. "Faster decision-making" is vague. "Faster decision-making enables us to respond to competitive pricing changes 1 week earlier, preventing 2-3% revenue loss that would cost $500K annually" is quantifiable. If benefits are truly intangible, acknowledge them but base your business case on quantified benefits to maintain credibility.

Q: Should we fund the full 3-year project, or commit year by year?**

A: Year 1 funding is a full commitment (software, implementation, labor). Year 2 and Year 3 funding should be contingent on achieving Year 1 success metrics. This creates accountability for execution and benefits realization while providing flexibility to adjust if circumstances change. It also protects leadership: if Year 1 doesn't deliver, they're not on the hook for Years 2-3 spending.

Q: What's a reasonable ROI threshold for AI projects?**

A: Strategic AI investments should target 30%+ ROI by year 3 (long payback acceptable for transformation). Quick-win projects should target 50%+ ROI in year 1 (fast payback required for quick wins). Below 20% ROI, the project usually isn't justified unless it's a strategic imperative (required by regulation, competitive response, etc.). Make this clear in your business case.