Hoe Een Echte AI-Businesscase Eruitziet (Met Getallen) What a Real AI Business Case Looks Like (With Numbers)

De meeste AI-businesscases zijn te vaag om op te handelen. Ze spreken over 'efficiëntiewinsten', 'strategisch concurrentievoordeel' en 'toekomstbestendigheid' — maar geven geen enkel concreet getal. Dit artikel laat zien hoe een werkbare businesscase eruitziet.

A convincing AI business case is not a slide deck with impressive statistics about global AI adoption. It is a specific, credible financial argument that your AI investment will generate returns that justify the cost, risk, and organisational disruption involved. Most AI business cases we review fail this standard — not because the opportunity isn't real, but because the analysis is too generic, too optimistic, and missing too many real costs. Here is how to build one that holds up to scrutiny.

Why Most AI Business Cases Are Unconvincing

Waarom de meeste AI-businesscases niet overtuigen

The typical AI business case draws on industry reports projecting transformative productivity gains, selects the most optimistic vendor case studies, and presents a simple payback calculation that ignores integration costs, change management, maintenance, and the almost universal reality that AI deployment takes longer than planned.

These cases fail for three reasons: they conflate potential with probability, they ignore total cost of ownership, and they don't specify the mechanism by which value will be captured — they assume it. A credible AI business case does the opposite: it presents conservative estimates, accounts for all costs, and describes precisely how and when value will flow.

Step 1: Define the Value Levers

Stap 1: Definieer de waardehefbomen

Value in AI comes from a small number of sources. Being specific about which ones apply to your case — and how much — is the foundation of a credible analysis:

• Labour cost reduction: Time saved on specific tasks multiplied by the fully-loaded cost of the people performing them. Be specific: which tasks, how many FTE, what percentage of their time?
• Error cost reduction: If the AI reduces error rates, what is the current cost of those errors? Include rework, customer impact, and compliance exposure.
• Revenue enablement: If AI allows faster decisions, more personalised offerings, or better customer service, what is the attributable revenue impact? This is harder to quantify and should be presented conservatively.
• Risk and compliance value: Reduced regulatory risk has real financial value. Quantify the cost of incidents the AI would prevent.
• Speed and capacity: If AI allows the same team to handle more volume, what is the cost of adding equivalent headcount? This is the opportunity cost of not implementing.

Step 2: Build a Realistic Cost Model

Stap 2: Bouw een realistisch kostenmodel

Most AI business cases significantly undercount costs. A complete cost model includes:

• Build/procurement cost: Development or licensing fees, integration work, data preparation.
• Change management and training: Typically 15–25% of total project cost, and almost always underestimated.
• Infrastructure and operations: Cloud compute, data storage, monitoring tools.
• Internal time: The time your own team spends on the project is a real cost. At senior rates, this adds up quickly.
• Annual maintenance: Budget 20–35% of initial build cost for ongoing maintenance, retraining, and improvement.
• Risk provision: A realistic contingency of 20–30% on the total cost estimate to account for delays and scope expansion.

A typical mid-complexity AI implementation for a 200-person organisation might look like: €180,000 build, €40,000 change management, €30,000 annual maintenance, €35,000 contingency. Total Year 1: approximately €285,000. This is a real number — not a headline figure.

Step 3: Model the Returns Conservatively

Stap 3: Modelleer de opbrengsten conservatief

Having established your cost model, apply conservative assumptions to your value levers. For a document processing automation example:

• Current state: 3 FTE spending 60% of their time on manual document processing, fully-loaded cost €75,000 per FTE = €135,000 annual cost of task.
• AI reduces manual processing time by 70%, saving approximately €94,500 annually.
• Error rate reduction saves an estimated €18,000 annually in rework and exception handling.
• Total annual benefit: approximately €112,500.
• Payback period at €285,000 Year 1 cost: approximately 2.5 years.
• 3-year NPV (assuming 10% discount rate): approximately €10,000 positive.

This is a modest but credible case. If your numbers look similar, the question is whether 2.5-year payback with positive NPV meets your investment threshold — and whether there are strategic benefits (capability building, future optionality) that justify the investment even if they don't show up in the NPV.

What Makes a Business Case Credible to a CFO

Wat een businesscase geloofwaardig maakt voor een CFO

Finance directors reviewing AI business cases look for the same things they look for in any capital investment proposal. They want to see that the team understands the risk, has been conservative in their assumptions, and can explain the mechanism by which value is created — not just assert that it will be.

Specific things that build credibility: named individuals responsible for delivering the benefits, a phased value realisation schedule, clear assumptions documented alongside the numbers, sensitivity analysis showing how the case changes if key assumptions shift by 20%, and a defined go/no-go decision point at which the project will be re-evaluated.

What destroys credibility: round numbers, vendor-sourced ROI estimates, no contingency, and benefit projections that assume 100% adoption on day one.