CONTEXT. JUDGEMENT. ACTION.
EDITIONBUSINESS.

Useful journalism to understand, manage and grow a business.

Search
Explore Edition Business
Analysis

How to Calculate the ROI of Generative AI for SMEs: A Practical Guide to Justifying the Investment

AI adoption among SMEs has outpaced their ability to measure its impact. This analysis offers a structured method for calculating return on investment, breaking down total costs and quantifiable benefits to turn spending into an investment that can be justified to management.

Illustration of a balance scale: servers, disks, and a cloud sit on one pan, while a potted plant sits on the other. A stylized circuit-brain appears in the background.
Editorial illustration · Edition Business

The measurement challenge: more adoption than visible returns

The integration of generative artificial intelligence into small and medium-sized enterprises (SMEs) is no longer a novelty; it has become an operational reality. However, there is a critical gap between the speed of adoption and the ability to measure its real impact. According to a national survey conducted by CEPE Di Tella and Fundar, with support from the IDB, 41.6% of SMEs already use at least one AI technology, but the vast majority focus on tools with a low technical threshold, and almost no one measures the impact.

The situation is similar in Spain. A McKinsey report indicates that only around one-third of companies have scaled their AI projects beyond the pilot phase, and just 39% attribute any impact on operating earnings (EBIT) to AI. For SMEs, limited budgets and lower risk tolerance make the problem worse. The result is a dangerous cycle: money is invested without defining key performance indicators (KPIs), the impact is not measured, management sees AI as an expense, and the project is abandoned before reaching its potential.

The ROI formula adapted to AI projects

Calculating the return on investment (ROI) of AI is not inherently more complex than calculating that of any other business investment, but it requires a rigorous framework before implementation. The classic formula is:

ROI (%) = [(Benefit generated by AI − Total AI cost) / Total AI cost] × 100

The denominator, total cost, is often underestimated. It includes more than just the tool license; it also includes:

  • Initial development or configuration: between 1,500 and 15,000 euros, depending on complexity (a customer service agent can cost between 3,000 and 6,000 euros; a complex integration can exceed 10,000 euros).
  • SaaS licenses: between 100 and 800 euros per month per tool.
  • Maintenance: budget 15–20% of the annual development cost.
  • Internal onboarding time: between 5 and 20 hours for each person involved in adoption.

The numerator, the benefit generated, should be broken down into layers that go beyond direct savings:

  1. Direct operational savings: hours saved multiplied by the employee’s hourly cost.
  2. Error reduction: multiplied by the average cost of an error in that process.
  3. Revenue improvement: an increase in the conversion rate or faster sales closures.
  4. Scalability with no marginal cost: the ability to handle more volume without hiring additional staff.
Illustration of a tall stack of papers beside a clock, a checklist on a clipboard, and, beyond an arrow, a smaller stack beside another clock.
Editorial illustration · Edition Business

Establishing the baseline: the critical step

The difficulty is not the calculation, but having an operational baseline before activating the tool. Without this starting point, any improvement will be invisible. The process should follow these steps:

  1. Document the current state: How many hours are spent on the task? What is the cost per error? How many leads are lost?
  2. Define specific KPIs: Avoid generic metrics such as “improve productivity.” Use indicators such as cost per task before and after, average response time, or the percentage reduction in manual errors.
  3. Calculate the actual total cost: Include all the components mentioned above.
  4. Apply the formula: Compare the quantified benefit against the total cost.
  5. Quarterly review: AI ROI is not static. Models improve and processes are adjusted. Regular reviews make it possible to identify deviations and correct them in real time.

Key metrics by project type

Not all metrics apply to every project. Below are five essential indicators that can be extracted from existing systems without the need for complex dashboards:

Indicator How it is calculated Data source Project that affects it
Hours freed up per month Hours spent before minus hours spent after Time records or team estimate Administrative task automation
Error reduction (Errors before − Errors after) / Errors before Incident logs or audits Data entry or reconciliation processes
Response time Average response time before and after CRM or support systems Customer service with AI agents
Conversion rate (Sales / Leads) before and after Sales and marketing data AI lead qualification
Cost per task Total process cost / Number of tasks Accounting and operational records Workflow optimization

Examples of application and returns

One illustrative case is a B2B services company with 12 employees that implemented an AI agent to qualify incoming leads through WhatsApp. Previously, a salesperson spent 2 hours a day answering initial inquiries, many of them from unqualified prospects.

  • Investment: 4,500 euros for development + 200 euros/month for maintenance.
  • Results at month 3:
    • Savings: 2 hours/day × 22 days × 35 euros/hour (salesperson’s cost) = 1,540 euros/month.
    • Improved conversion: from 12% to 18% among qualified leads, resulting in 2 additional customers per month.

This example shows that actual ROI includes every layer of benefit. In many projects, most of the return comes not from direct savings on hours, but from improved conversion and scalability.

Conclusion: from a tool to a business process

The next step forward is not buying the next fashionable tool, but choosing a specific process and redesigning it end to end with AI. A good candidate meets three conditions: it is repetitive, consumes measurable hours, and has clear rules or documents as inputs.

AI is measured not by what it promises, but by what it changes. Measuring ROI allows SMEs to allocate resources effectively, identify which projects align with strategic objectives, and confidently scale initiatives that generate real value. Without this measurement, AI becomes a scattered expense that is difficult to justify; with it, AI becomes a defensible and optimizable business infrastructure.

Sources and methodology

  1. ROI de la IA en tu pyme 2026: fórmula en 3 pasos ↗novantin.com
  2. ROI de la IA para Empresas: Cómo Calcularlo en 2026 ↗iaconsultora.com
  3. 5 métricas para medir el ROI de la IA en empresas ↗es.linkedin.com
  4. Cómo medir el ROI de las aplicaciones de IA Generativa ↗neuraltrust.ai
  5. ROI de la IA en PyMEs: cómo medir su impacto real ↗www.contarg.com
Report an error ↗

Continue exploring