How to Measure AI ROI: A Practical Method for SMEs

What AI ROI Is – and Why Many SMEs Get It Wrong

The return on investment (ROI) of an AI initiative measures how much value it creates relative to its total cost. The base formula is simple: ROI = (value − cost) / cost, expressed as a percentage. An ROI of 100 percent means the investment paid for itself once and generated the same amount again on top.

The error rarely lies in the formula, but in the numbers you put into it. Many SMEs count only the licence fees, ignore setup and maintenance, and value the benefits too optimistically. Measuring ROI honestly requires a complete cost base, a documented starting point (baseline) and conservatively quantified benefits.

The Cost Structure: Setup, Licences and Maintenance

AI costs fall into three blocks. Only by capturing all three do you get a reliable ROI. Use a twelve-month horizon (total cost of ownership) so that one-off and recurring costs become comparable.

• Setup (one-off): needs analysis, data preparation, integration with existing systems, configuration, testing and training – often the most underestimated item.
• Licences (recurring): per-user subscriptions, usage-based fees or API costs by token consumption. With generative AI, costs can fluctuate strongly with usage.
• Maintenance (recurring): monitoring, prompt and model upkeep, updates, retraining, quality control and governance.
• Internal time: the hours your staff spend on rollout, support and change management – valued at a realistic fully loaded cost rate.
• Risk and compliance costs: data-protection review, documentation and any adjustments required under the revDSG.

The Value Drivers: Time, Revenue, Errors and Risk

Value arises in four ways. Separate hard, cash-effective impacts (lower spend, higher revenue) from soft impacts (satisfaction, relief). Both matter, but only hard impacts belong in the ROI calculation without a discount.

• Time saved: automated or accelerated tasks. Value = hours saved × fully loaded cost rate. Only time genuinely reallocated or removed is cash-effective.
• Revenue captured: faster quotes, higher close rates, better service availability, fewer lost enquiries.
• Fewer errors: reduced rework, complaints, returns or fines thanks to higher quality and consistency.
• Reduced risk: better traceability, fewer compliance breaches, less dependence on individual people.

A Simple ROI Method in Five Steps

• 1. Define the use case and baseline: what exactly should the AI improve? Measure the current state (time, cost, error rate) before you start.
• 2. Estimate total cost over twelve months: setup + licences + maintenance + internal time + compliance.
• 3. Quantify value conservatively: convert to francs, report soft effects separately, factor in the adoption rate (not everyone uses the tool 100 percent).
• 4. Calculate ROI and payback period: ROI = (value − cost) / cost. Payback = cost / monthly net benefit.
• 5. Track actuals: after three to six months, replace estimates with real measurements and decide whether to scale up or stop.

Common Mistakes When Measuring AI ROI

• Counting only licence costs while ignoring setup and maintenance.
• No baseline: without a starting value you cannot prove any improvement.
• Booking soft savings as cash: '20 percent faster' is only money if the time is genuinely reused or removed.
• Overestimating adoption: a pilot with ideal usage rarely scales unchanged.
• Extrapolating a one-off pilot value to permanent operation.
• Ignoring risk and compliance – especially with personal data.

The Swiss Perspective: revDSG, Francs and Governance

For Swiss SMEs, compliance belongs in the cost calculation – not as a brake, but as part of an honest TCO. If your AI processes personal data, the revDSG (revised Data Protection Act, in force since 1 September 2023) requires, among other things, transparency, a record of processing activities and, depending on risk, a data-protection impact assessment. The EDÖB is the supervisory authority.

Calculate every figure in francs and factor in data location, data processing agreements and the providers' contractual terms. These points are not pure cost: good governance lowers risk and strengthens the trust of customers and staff – a benefit that pays off over the long term.