How to measure AI ROI in your company

To measure AI ROI you need three things: a baseline of costs before implementation, clear metrics (hours saved, errors avoided, response speed), and a comparison period of at least 30 days. The average ROI for Italian SMEs that automate customer service with AI is 250%, with payback in under 5 months.

The measurement problem

Many Italian SMEs find themselves in a paradoxical situation: they have implemented AI solutions, they perceive that they work, but they cannot precisely quantify the value generated. This creates problems on two fronts: it makes it difficult to justify additional investments within the company, and it prevents optimizing what has already been built.

The problem is almost always in the same place: lack of a baseline.

First rule: measure before

AI ROI is measured like any other investment: by comparing the state after with the state before. If you did not measure the state before, you cannot measure the improvement.

This seems obvious, but in practice it is the step most often skipped. The AI solution is implemented, perceptive improvements are noticed, and then the business case is constructed after the fact — with incomplete data and selective memory.

The rule at IL DOGE DI VENEZIA is simple: before any implementation, we spend at least 2-4 weeks measuring the existing process. Not estimating it. Measuring it.

The metrics that truly matter

Operational efficiency metrics

• Person-hours per unit of output: The most direct KPI for repetitive processes. If it previously took 3 hours to process 100 orders and now it takes 0.5, the improvement is measurable and unequivocal.
• Error rate: Fundamental for processes where accuracy matters. Data entry, quality control, accounting reconciliation.
• Cycle time: From the moment a process starts to when it concludes. Relevant for customer service, order management, approvals.

Commercial impact metrics

• Conversion rate: If you have automated commercial follow-up, measure whether the conversion rate has changed.
• Customer satisfaction (CSAT/NPS): If you have implemented AI in customer service, this is the metric that matters.
• Customer response time: A proxy indicator of service quality that is easily measurable.

Financial metrics

• Cost per transaction: Divide the total process cost (staff + technology) by the number of transactions processed.
• Absolute annual savings: (Hours saved x average hourly cost) + (Errors avoided x average cost per error)
• Incremental revenue: For implementations that impact the commercial front-end.

The ROI calculation framework

The basic formula is simple:

ROI = (Total annual benefits - Total solution cost) / Total solution cost x 100

The "total solution cost" must include:

• Implementation cost (development, configuration, integration)
• Annual license/SaaS cost
• Maintenance and update costs
• Initial training cost
• Cost of internal time spent on the project

The "total annual benefits" must include:

• Labor cost savings (hours saved x hourly cost)
• Reduction in error costs
• Incremental revenue (if applicable)
• Value of scalability (additional capacity without additional costs)

Market benchmarks

In AI implementations for Italian SMEs, the benchmarks we use as reference are:

• 12-month ROI: 150-400% for standard process automations
• Payback: 3-8 months for Tier 1 solutions (high-volume processes, clear rules)
• Average annual savings: 80,000-250,000 euros for 50-200 employee SMEs with their first project

These numbers vary significantly based on process complexity, available data quality, and speed of internal adoption.

When ROI is hard to measure

There are AI implementations where direct ROI is difficult to isolate: adoption of generative AI platforms for general staff use, improvement of managerial decision quality, risk reduction.

For these cases, our advice is not to try to force precise quantification. Instead, use reasonable proxies and be transparent about the uncertainty of the estimate. An honest business case with declared uncertainties is more credible — and useful — than a precise number built on fragile assumptions.

If you want to build a solid business case for an AI project in your company, contact us — it is one of the services we offer as part of our assessment process.