Challenges in measuring return on investment in artificial intelligence

Investing in artificial intelligence has become more accessible than proving that the investment has generated financial returns. Companies have adopted AI tools in areas such as customer service, programming, marketing, finance, and operations, but many still struggle to translate these projects into concrete numbers.

A Deloitte survey of 1,854 executives in Europe and the Middle East found that 85% of organizations increased their investment in AI in the 12 months prior to the study. At the same time, most respondents said that a typical use case takes two to four years to achieve a satisfactory return.

Before evaluating the return on artificial intelligence, it is essential to understand how the technology is being applied and what results can be directly attributed to it.

How companies calculate the ROI of artificial intelligence

The traditional ROI calculation compares the gain obtained with the cost of the investment. However, in the case of AI, the calculation can involve less obvious expenses, such as infrastructure, system integration, employee training, security, governance, and model maintenance.

For this reason, companies monitor indicators such as cost reduction, revenue growth, productivity, task execution time, and process quality.

For example, a tool that reduces the time required to produce a report from two hours to 30 minutes generates a productivity gain. To convert this into ROI, the company needs to estimate the financial value of these saved hours and compare it with the cost of the solution.

PwC also distinguishes between so-called financial return, or hard ROI, and benefits that are harder to monetize, such as employee satisfaction, acquisition of new skills, and talent retention.

When the numbers don’t add up, the problem may not lie with the technology, but with the indicator chosen, which may not be connected to the objective that justified the investment.

Why many AI projects don’t show a return

One obstacle is measuring the result of a project in isolation. PwC points out that organizations need to consider AI performance on an ongoing basis, as models can lose efficiency and require new investments over time.

Another problem is the lack of reliable data. Systems scattered across different areas make it difficult to compare the situation before and after implementation. According to Deloitte, issues related to data and infrastructure are among the barriers to capturing value with AI.

There is also a difference between automating a task and redesigning a process. A company may provide a chatbot for employees and count the number of interactions, but this data alone does not show whether there has been a reduction in costs or an improvement in service.

The indicator needs to answer a business question. If the goal is to reduce expenses, the metric should show savings. If the goal is to increase sales, it should track revenue or conversion. If the goal is to speed up operations, time and productivity may be more relevant.

What to do when the return doesn’t appear

One of the first steps is to review the relationship between investment and expected outcome. This includes considering technology, integration, training, and maintenance costs, rather than just comparing the price of the tool with an estimate of productivity.

It is also necessary to establish a baseline before implementation. Without knowing how long a task took, how much it cost, or what its error rate was previously, it is more difficult to determine the impact caused by AI.

Another approach is to evaluate the portfolio of initiatives. A project may have limited financial returns individually, but contribute to a larger transformation when combined with other systems and processes.

A BCG survey of more than 280 finance executives highlighted this challenge. The median ROI reported for AI and generative AI initiatives in the area was 10%, while many organizations had a benchmark return of 20%. Nearly a third of respondents reported limited gains.

The return on AI starts before the technology

An analysis by PwC of 1,217 companies across 25 sectors found a strong concentration of gains: the top 20% in the index used by the study captured 74% of AI-related returns. The research linked this performance to factors such as strategy, data, technology, workforce, governance, and how AI is applied.

The practical conclusion is that more investment does not automatically mean more return. To know if the numbers add up, companies need to define the problem before the tool, establish metrics before implementation, and track results over time.