Building the Business Case

Why Business Cases Matter for AI

AI projects are easy to justify abstractly (“AI could help us!”) but harder to justify concretely. Without a solid business case, your initiative will compete for resources with proven alternatives. You need to demonstrate not just that AI is cool, but that it creates business value worth the investment.

A good business case answers the question: Why should we spend $X on this AI project instead of on something else?

ROI Frameworks for AI

Direct Cost Savings

What it is: AI replaces human effort on tasks, reducing labor costs.

How to calculate:

Identify the task being automated
Determine current cost: (# of people × salary) + tools + infrastructure
Estimate reduction: How much less labor do we need?
Calculate savings: Current cost × reduction percentage
Subtract AI costs: Savings minus API/infrastructure/team costs

Example: Email Triage Automation

Current state:

3 support staff @ $50K each = $150K
They spend 40% of time triaging/routing emails
Current cost: $150K × 40% = $60K/year

With AI:

AI classifies and routes emails, reducing triage work by 80%
Savings: $60K × 80% = $48K
AI costs: GPT-4 API for 500 daily emails = $1,200/year
Net savings: $48K - $1.2K = $46.8K/year

ROI: $46.8K savings on $15K implementation = 3.1x return year 1

Watch out for:

Assuming people disappear when tasks are automated (they usually shift to other work)
Overestimating how much time is actually saved
Forgetting that AI quality might be 80%, requiring human verification of 20%
Not accounting for training/ramp time during transition

Productivity Gains

What it is: AI helps people do more in less time, enabling them to take on more work or move to higher-value tasks.

How to calculate:

Identify the impact on throughput or quality
Translate to business value: More customers served? Faster delivery? Higher quality?
Calculate revenue or cost impact
Subtract AI costs

Example: Sales Productivity Improvement

Current state:

Sales team: 20 people averaging $500K revenue each = $10M
They spend 20% of time on research, lead qualification, data entry (low-value work)

With AI:

AI handles lead research and initial qualification
Sales reps can focus 100% on selling vs. 80%
Productivity increase: 20% × 20 reps = 4 reps’ worth of additional capacity
Revenue impact: 4 reps × $500K = $2M additional revenue

Gross benefit: $2M × 30% profit margin = $600K AI cost: Lead scoring and CRM integration = $50K + $200K/year Year 1 ROI: ($600K - $200K) / $50K initial = 8x

Watch out for:

Assuming productivity gains are linear (they often plateau)
Market saturation (you can’t sell 20% more if there’s no market)
Team attrition risk (people might leave if their role changes dramatically)
Ramp time (productivity doesn’t improve day 1; it takes weeks)

Quality Improvements

What it is: AI improves accuracy or consistency, reducing errors and their costs.

How to calculate:

Identify the error or quality gap
Calculate cost of current errors: (error rate × volume × cost per error)
Estimate AI impact: By what % would AI reduce errors?
Calculate benefit: Current cost × improvement %
Subtract AI costs

Example: Loan Application Quality

Current state:

10,000 loan applications/year
5% error rate (500 errors) = ~50 loans incorrectly approved, 450 declined that shouldn’t have been
Cost of false approvals: 50 × $20K loss = $1M
Cost of false rejections: 450 × $2K lost opportunity = $900K
Total error cost: $1.9M

With AI:

AI reviews applications, reduces false approval rate by 80%
Reduces false rejection rate by 40%
Savings: ($1M × 80%) + ($900K × 40%) = $800K + $360K = $1.16M
AI cost: Model training + API calls = $100K + $50K/year
Year 1 benefit: $1.16M - $50K = $1.11M

ROI: $1.11M / $100K = 11x

Watch out for:

Overestimating how much AI improves quality (it usually doesn’t reach 100%)
Costs changing behavior (customers discover errors anyway, offsetting savings)
Implementation friction (if system is too conservative, you lose value)

Time-to-Value and Speed Improvements

What it is: AI enables faster execution, which has business value (first-mover advantage, faster customer response, etc.).

How to calculate:

Identify what speeds up (response time, decision time, delivery time)
Quantify value of speed: What’s worth for faster response?
Translate to business outcome: customer retention, market share, etc.
Calculate benefit
Subtract costs

Example: Customer Support Response Time

Current state:

Average support response time: 8 hours
Customer satisfaction scores correlate with response time
Each 4 hours of delay ≈ 2% increase in churn
Current churn: 5% = $500K annual loss

With AI:

AI provides immediate response to 50% of questions
Remaining escalations handled within 2 hours instead of 8
Average response time drops from 8 hours to 4 hours
Churn reduction: 2% = $200K saved
AI cost: $100K setup + $50K/year
Year 1 benefit: $200K - $50K = $150K

ROI: $150K / $100K = 1.5x (lower ROI but improves customer satisfaction)

Watch out for:

Churn reduction being nonlinear (some customers are sticky regardless)
Satisfaction not always translating to retention
Customer lifetime value changing over time
Competitive pressure (competitors improve simultaneously)

Cost Modeling: The Hidden Costs

Most AI cost models miss the “small” costs that add up.

Direct Costs (Usually Easy to Identify)

Model/API costs: GPT-4 at $0.03/1K input tokens, $0.06/1K output tokens
Infrastructure: Servers, databases, monitoring ($100-1K/month)
Tooling: Prompt management, testing, version control ($500-5K/month)
Initial development: Engineering time to build and integrate (3-6 months of salary)

Hidden Costs (Often Missed)

Data preparation: Cleaning, labeling, formatting data (can be 50% of project cost)
Monitoring and maintenance: Continuous monitoring, alert handling, fixes ($5-20K/month at scale)
Model retraining: As data patterns shift, models degrade and need retraining ($10-100K/year)
Team learning curve: Time before team is truly productive (4-12 weeks)
Governance and compliance: Policy development, review processes, audit trails ($20-50K)
Human oversight: Someone verifying results, handling edge cases (ongoing)
Vendor lock-in costs: Migrating from one provider is expensive (plan for this)

Total Cost of Ownership Example: Document Analysis System

Year 1:

API costs: 10,000 documents × $0.10/doc = $1,000
Server/DB: $500/month × 12 = $6,000
Tooling: $2,000
Development: $150,000 (4 months of 2 engineers)
Data preparation: $40,000 (cleaning training data)
Monitoring setup: $10,000
Total Year 1: $209,000

Year 2+:

API costs: $1,000
Infrastructure: $6,000
Maintenance: $20,000
Model improvements: $30,000
Monitoring/incident response: $10,000
Total Year 2+: $67,000/year

Break-even at roughly 6 months if creating $35K+/month in value.

Funding Strategies

Once you have a business case, how do you get it funded?

Option 1: Efficiency Play (ROI-Driven Funding)

When: You have clear cost savings or productivity gains.

Pitch: “This costs $X but saves $Y, so it pays for itself in Z months.”

Who funds: CFO, business unit leads

Timeline: Often funded immediately if ROI is clear and risk is low

Example: “Support automation saves $50K/year for $15K implementation. Six-month payback. Let’s go.”

Option 2: Capability Building (Strategic Funding)

When: You’re building organizational capability with longer-term value.

Pitch: “This doesn’t pay for itself in year 1, but it builds capability we’ll use across multiple initiatives over time.”

Who funds: CTO, strategic innovation budget

Timeline: Typically 12-24 month commitment with checkpoints

Example: “AI platform costs $400K year 1 but enables 5 future initiatives worth $2M+ together. Let’s invest.”

Option 3: Venture Model (Experiment Funding)

When: You’re exploring whether something is possible.

Pitch: “We don’t know if this works. This pilot ($50K) will tell us. If it works, we’ll scale it.”

Who funds: Innovation budget, business unit

Timeline: 8-12 week pilots with go/no-go decision

Example: “Personalization might increase revenue 3-5%, but we need to validate the model works. Here’s our $75K pilot plan.”

Presenting the Business Case

Structure

Problem statement (1 paragraph)
- What problem are we solving?
- Why does it matter?
- What’s the business impact of not solving it?
Proposed solution (2-3 paragraphs)
- What’s our approach?
- Why AI?
- Why this approach vs. alternatives?
Expected outcomes (specific numbers)
- Cost savings/revenue impact
- Timeline
- Success metrics
Investment required (specific costs)
- Development cost
- Ongoing costs
- Infrastructure/team costs
- Timeline to positive ROI
Risk assessment (what could go wrong)
- Technical risks (model accuracy, integration)
- Organizational risks (adoption, team capability)
- Business risks (market changes, competition)
- Mitigation plan
Go/no-go decision criteria (when to stop)
- If accuracy is below X%, we stop
- If ROI projects below Y%, we pivot
- If adoption is below Z%, we reconsider

Sample Business Case (2 Pages)

SUBJECT: Proposal to Implement AI-Powered Customer Support Triage

Problem: Support team spends ~$60K/year on email triage and routing (40% of 3 FTEs). Routing errors cause 15% of escalations. Response time averages 8 hours, correlating with 2% increase in churn per 4-hour delay.

Solution: Implement Claude API to automatically classify incoming support emails and route to appropriate team. System handles triage for all emails; humans do response.

Expected outcomes:

Reduce triage workload by 70% (savings: $42K)
Improve routing accuracy from 85% to 95% (reduce escalations 15% → 5%, saving ~$30K in rework)
Reduce response time from 8h to 3h average (reduce churn by 1.5%, saving ~$150K)
Total year 1 benefit: $222K

Investment:

Development: 6 weeks × 2 engineers = $60K
Initial infrastructure setup: $15K
Annual API costs: $1.5K (based on usage model)
Year 1 total: $76.5K

ROI: $222K / $76.5K = 2.9x year 1

Timeline: 8 weeks to launch, 12 weeks to collect sufficient usage data for validation

Key risks:

Routing accuracy might not reach 95% (mitigation: initial human review of all AI decisions)
Integration complexity higher than estimated (mitigation: 2-week prototype first)
Team adoption slower than expected (mitigation: gradual rollout with champions)

Success criteria:

Triage accuracy ≥90% at month 2
User adoption ≥80% of support staff by month 4
ROI projection validated with actual data by month 3

Recommendation: Approve Phase 1 (8-week pilot) with mandatory go/no-go decision before Phase 2.

Making Conservative Assumptions

The best business cases are conservative. If you overestimate benefits, you’ll look bad when results don’t match projections.

Conservative practices:

Use past performance data, not optimistic projections
Assume slower adoption than you think
Model costs higher than you estimate (they usually are)
Factor in implementation delays (they happen)
Plan for 6-month ramp before you see full benefits

Example conservative estimate:

Optimistic estimate: 80% cost savings, 3-month payback
Conservative estimate: 50% cost savings, 6-month payback
Plan on conservative; celebrate if optimistic wins out

Strategic Questions for Your Business Case

What’s the primary value driver? Cost savings? Productivity? Quality? This changes funding approach.
How confident are we in the numbers? If we’re wrong by 50%, does the business case still work?
What would change our mind? Set decision criteria up front.
How does this fit strategic priorities? Does it align with CEO priorities? That matters for funding.
Who would champion this internally? Do you have executive sponsorship?

Key Takeaway: Build business cases grounded in quantified value and realistic costs. Use frameworks for cost savings, productivity, quality, and speed. Account for hidden costs. Be conservative in projections. Present clearly with specific success criteria and risk mitigation. The best business case is one where conservative assumptions still show strong ROI.

Discussion Prompt

For your priority AI initiative: What’s the primary value driver (cost, revenue, quality, speed)? What would a conservative business case look like with real numbers from your business?