Listen to this lesson
Course: Measuring AI ROI Pathway: Business (Paid) Level: Intermediate Estimated Reading Time: 8 minutes
Every investment decision comes down to the same question: is this worth it?
With AI, that question is harder to answer than it should be. Vendors talk in generalities. Case studies are suspiciously perfect. And the costs — the real, ongoing costs — are often buried in technical jargon that makes it difficult to build a proper business case.
This lesson strips the noise away. We'll look at what AI actually costs, what returns are realistic, and how to think about the investment case in terms your board and your CFO will recognise.
AI projects have a different cost structure to traditional software. Understanding this upfront saves painful conversations later.
Infrastructure and compute. AI models need processing power. If you're running models in the cloud — which most NZ businesses are — you're paying per use. A mid-sized company using AI for customer service might spend $2,000–$8,000 NZD per month on API calls and cloud compute. That's not a one-off capital expense; it's an ongoing operational cost.
Data preparation. This is where most budgets blow out. Your data is almost certainly not ready for AI. It's in spreadsheets, legacy systems, PDFs, and people's heads. Cleaning, structuring, and connecting that data typically costs 2–3 times what people budget for it. A realistic data preparation project for a mid-market NZ business runs $30,000–$150,000 NZD, depending on complexity.
Integration. AI doesn't work in isolation. It needs to connect to your existing systems — your CRM, ERP, accounting software, whatever your team actually uses day to day. Integration work is often underestimated. Budget $20,000–$80,000 NZD for a meaningful integration project.
People. You need someone who understands the technology, someone who understands your business, and ideally someone who understands both. Whether that's internal hires, contractors, or a mix, the people cost is real. A capable AI engineer in New Zealand commands $130,000–$180,000 NZD per year. A data scientist, $110,000–$160,000 NZD.
Ongoing maintenance. AI systems aren't "set and forget." Models drift. Data changes. The world moves on. Plan for 15–25% of your initial build cost annually in maintenance.
Here's where honest conversations matter most.
AI returns generally fall into four categories:
1. Cost reduction. This is the easiest to measure. If an AI system handles 40% of your customer enquiries that previously required a human, you can calculate the saving directly. A NZ contact centre handling 10,000 enquiries per month at an average cost of $12 NZD per interaction could save $48,000 NZD monthly by deflecting 40% to AI. That's measurable, defensible, and real.
Valuation benchmarks from the market. The AI industry now has real-world examples of scale and sustainability. OpenAI, generating over $25 billion in annualized revenue as of early 2026, demonstrates that AI-native business models can achieve substantial economic value. When building your investment case, these public benchmarks help contextualise projections:
Use these benchmarks when presenting to boards or CFOs who need reassurance that AI isn't just hype — it's already generating multi-billion dollar revenues for established companies.
2. Revenue growth. Harder to isolate, but real. AI-driven personalisation in e-commerce typically lifts conversion rates by 10–30%. For a NZ online retailer doing $5 million in annual revenue, even a conservative 10% lift is $500,000 NZD. The challenge is proving the AI caused the lift versus other factors.
3. Speed improvements. Time-to-decision, time-to-market, time-to-resolution. A law firm using AI for document review might cut due diligence time from 40 hours to 8 hours per transaction. At $350 NZD per hour, that's $11,200 NZD saved per deal — and the ability to take on more deals.
4. Risk reduction. Fraud detection, compliance monitoring, quality control. The value here is often in what didn't happen — the fraud that was caught, the compliance breach that was avoided. Important, but harder to put a dollar figure on.
AI investments rarely pay off in month one. A realistic timeline for a meaningful AI project in a NZ business:
If someone promises you returns in 30 days, be sceptical. If someone says 18 months, they're probably being honest.
A credible AI investment case has three components:
Conservative projections. Use the low end of any range. If the case still works at the conservative end, it's a good investment. If it only works at the optimistic end, it's a gamble.
Clear cost boundaries. Include everything: infrastructure, people, data prep, integration, maintenance, and a contingency of at least 20%. AI projects overrun more often than they don't.
Defined success criteria. Before you spend a dollar, agree on what "success" looks like. Is it a 20% reduction in processing time? A 15% improvement in customer satisfaction scores? A $200,000 NZD annual saving? Write it down. Everyone signs off.
For most NZ mid-market businesses, a well-scoped AI project should target a payback period of 12–24 months. That's comparable to other significant technology investments.
If your payback period calculation stretches beyond 36 months, either the project scope is too large, the returns are too uncertain, or both. Consider breaking it into smaller phases with independent value.
The AI investment case isn't fundamentally different from any other technology investment. It just has some characteristics that catch people out: ongoing compute costs, data preparation requirements, and a longer ramp to full returns.
The businesses that get this right are the ones that go in with realistic expectations, conservative numbers, and clear success criteria. The ones that struggle are the ones chasing vendor promises without doing their own arithmetic.
Build a rough AI investment case for your organisation.
Write this up in a single page. You're not committing to anything — you're practising the discipline of putting real numbers on the table.
Question 1: What percentage of initial build cost should you budget annually for AI system maintenance?
A) 5–10% B) 15–25% C) 30–50% D) Less than 5%
Answer: B) 15–25%. AI systems require ongoing maintenance as models drift and data changes over time.
Question 2: Which category of AI return is typically easiest to measure?
A) Revenue growth B) Risk reduction C) Cost reduction D) Speed improvements
Answer: C) Cost reduction. It can be directly calculated from the volume of work deflected from human handling to AI.
Question 3: A realistic payback period for a well-scoped AI project in a NZ mid-market business is:
A) 1–3 months B) 6–9 months C) 12–24 months D) 36–48 months
Answer: C) 12–24 months. This is comparable to other significant technology investments and reflects the time needed for data preparation, integration, and scaling.