“This will transform our operations.” I’ve heard that pitch fail more times than I can count. CFOs don’t fund transformation—they fund returns.

If you want to get an AI project approved in manufacturing, you need a business case that speaks the language of finance: costs, benefits, risks, and returns. Here’s how to build one.

Start with a specific problem, not a technology

The worst business cases start with “We should implement AI” and then go looking for applications. That’s backwards.

Start with a problem:

• “Unplanned downtime on Line 3 cost us $420,000 last year”
• “Quality rejections are running at 3.2%, versus our target of 2%”
• “We’re spending $180,000 annually on manual inspection that could be automated”

A specific problem gives you measurable outcomes. That’s what finance cares about.

Quantify the current state

Before you can show improvement, you need to establish the baseline. This means actual numbers, not estimates.

For downtime reduction:

• Total hours of unplanned downtime (from maintenance records)
• Cost per hour of downtime (production value, labour, ripple effects)
• Pattern of failures (which equipment, how often)

For quality improvement:

• Current defect rates (by type, by line, by shift)
• Cost of quality (scrap, rework, customer complaints, warranty)
• Where in the process defects are caught

For labour efficiency:

• Current staffing on target activities
• Fully loaded labour cost (wages, super, overheads)
• What else that labour could be doing

Get the real numbers from your systems. If you don’t have good data, that’s a problem to solve before the AI project.

Be realistic about benefits

Here’s where most business cases go wrong: they assume the AI will work perfectly from day one.

In reality:

• Implementation takes longer than planned
• Initial performance is lower than eventual steady-state
• There’s a learning curve for staff
• Not all promised benefits materialise

A credible business case acknowledges this. I typically recommend:

Conservative scenario: Assume 60% of target benefits, 120% of expected costs Base case: Full expected benefits and costs Optimistic scenario: 120% of benefits, 80% of costs

Present all three. Your CFO will appreciate the honesty, and you won’t be held to an unrealistic number.

Calculate the full cost

The software licence is just the beginning. A complete cost picture includes:

One-time costs

• Software and hardware purchase
• System integration (often the biggest hidden cost)
• Data preparation and migration
• Infrastructure upgrades
• Training and change management
• Project management and contingency

Ongoing costs

• Annual licensing/subscription fees
• Maintenance and support
• Additional staffing (if needed)
• Data storage and compute
• Model retraining and updates
• Technology refresh

A rule of thumb: the total cost of ownership over five years is typically 2-3x the initial purchase price. If your business case only includes the purchase price, you’re underestimating.

Build the ROI calculation

Now you can do the maths. The basic framework:

Net Present Value (NPV): Total benefits minus total costs, adjusted for the time value of money. This should be positive.

Payback Period: How long until cumulative benefits exceed cumulative costs? Most CFOs want this under 24 months for discretionary investments.

Internal Rate of Return (IRR): The effective return on the investment. Compare this to your company’s hurdle rate for capital projects.

Let me show a simplified example:

Project: AI-assisted predictive maintenance for critical production line

Costs:

• Year 0: $150,000 (implementation)
• Years 1-5: $25,000/year (ongoing)

Benefits:

• Years 1-5: $80,000/year (downtime reduction, conservative estimate)

Payback: ~2.5 years 5-year NPV (at 10% discount): ~$145,000 IRR: ~28%

Those are numbers a CFO can work with.

Address the risks

A business case that ignores risks isn’t credible. Common risks in manufacturing AI projects:

Technical risks:

• Integration challenges with existing systems
• Data quality insufficient for AI
• Performance doesn’t meet expectations

Operational risks:

• Staff resistance or lack of adoption
• Process changes harder than expected
• Vendor support inadequate

Strategic risks:

• Technology becomes obsolete
• Vendor goes under or pivots
• Regulatory changes affect approach

For each risk, identify:

• Likelihood (low/medium/high)
• Impact (low/medium/high)
• Mitigation strategy

This shows you’ve thought it through and aren’t just presenting happy-path numbers.

Consider the alternatives

A strong business case compares options:

Option 1: Do nothing What happens if you don’t invest? Costs continue. Competitors may gain advantage. But also, no implementation risk.

Option 2: Non-AI alternatives Could you solve the problem with traditional automation? Different processes? More people? Sometimes a lower-tech solution makes more sense.

Option 3: Phased approach Instead of full implementation, could you pilot on one line first? Lower upfront cost, reduced risk, but also delayed full benefits.

Option 4: Full implementation The proposed project.

Showing you’ve evaluated alternatives strengthens your credibility.

Make the intangible tangible

Some benefits are hard to quantify but still matter:

• Improved safety from automated monitoring
• Better data for future decisions
• Staff satisfaction from modern tools
• Customer perception of innovation

Don’t ignore these, but don’t rely on them either. Mention them as additional upside beyond the quantified benefits.

Structure the document

A typical structure that works:

1. Executive summary: Problem, solution, key numbers, recommendation (one page)
2. Current situation: The problem with data
3. Proposed solution: What you want to do
4. Costs: Full breakdown
5. Benefits: Conservative, base, optimistic
6. Financial analysis: NPV, payback, IRR
7. Risks and mitigations
8. Alternatives considered
9. Recommendation and next steps
10. Appendices: Detailed calculations, vendor comparisons, technical specs

Keep the main document short—10 pages maximum. Put detail in appendices for those who want it.

Get support before the meeting

Don’t surprise your CFO. Before the formal presentation:

• Get operations leadership aligned
• Have IT review technical feasibility
• Brief finance on the approach
• Address objections in advance

The business case meeting should be a confirmation, not a negotiation.

If you get stuck

Building a credible business case takes work. If you don’t have the internal expertise, it’s worth getting help. Firms that specialise in manufacturing AI—including AI consultants in Melbourne—can help structure the analysis and provide realistic benchmarks.

Just make sure whoever helps you is independent of the vendor you’re evaluating. You want honest numbers, not a sales pitch dressed up as analysis.

A well-built business case won’t just get your project approved. It’ll set you up for success by establishing clear expectations and accountability from the start.