Companies are convinced that artificial intelligence brings benefits, although most are unable to quantify them. This means that the next stage in the development of AI will not be the implementation of more tools, but rather aligning them with specific financial and operational outcomes.
As many as 91 per cent of companies in the telecoms, media, semiconductors and technology sectors say they are satisfied with their AI investments. At the same time, only a third have a framework in place to measure their commercial value. This is according to the Economist Enterprise survey, prepared with the support of HCLTech based on responses from 202 senior executives in the US and Europe.
This discrepancy is more significant than the level of declared enthusiasm itself. It shows that many companies regard the launch of a tool, an increase in its usage, or positive feedback from staff as a success. However, none of these indicators determines whether the investment improves margins, boosts sales, reduces customer churn or lowers the cost of process management.
Without such data, the board does not know which projects to develop and which to discontinue. Nor can it compare investment in AI with infrastructure modernisation, product development or the recruitment of new staff. The technology may function correctly, yet still be economically unjustified.
The cost of the model becomes part of the business case
The problem becomes more acute as the scale increases. The cost of a pilot project is relatively easy to control. In regular use, however, there are additional expenses for computing power, integration, data, licences, security, testing and human oversight. A company may therefore record a rise in productivity which, when total costs are taken into account, does not yield a real return.
The latest KPMG survey from the second quarter of 2026 indicates that only 26 per cent of organisations have full, real-time visibility of their AI operating costs. Companies with this insight are five times more likely to report a proven return on investment than those that do not monitor costs in a similar way.
It is impossible to manage the value of AI without measuring its full cost. The number of users, queries or models deployed should be treated as technical data, not as proof of success.
Every major implementation requires a baseline from before the AI was deployed. This could be the cost per transaction, customer service response time, error rate, sales conversion rate or time to market. Only by comparing these figures with the system’s costs can one determine whether value has been created, or whether it is merely an additional layer of technology.
Companies are scaling up their tools faster than their ability to manage them
A lack of measurement is not a standalone analytical problem. It usually points to a weakness in the entire operational model. In an Economist Enterprise survey, 62 per cent of companies admitted that AI development is being slowed down by integration difficulties and technological debt. Without a consistent flow of data, it is difficult both to automate a process and to evaluate its outcome.
A similar gap exists when it comes to people and accountability. Only 20 per cent of the organisations surveyed have a strategy for upskilling or recruitment related to AI. A mere 17 per cent state that the oversight policies they have adopted actually influence how the systems operate.
This means that companies often purchase the technology before determining who is to redesign the process, evaluate the results and take responsibility for erroneous decisions. Within such a structure, AI remains an add-on to existing work. It may speed up individual tasks, but it does not change the outcome for the organisation as a whole.
Oversight is not, however, solely a matter of compliance. From 2 August 2026, further provisions of the AI Act will come into force in the European Union, including requirements regarding the transparency of certain systems and AI-generated content. Monitoring how models operate therefore becomes a prerequisite for scaling, risk management and regulatory compliance.
It is the ability to choose, rather than the number of deployments, that will provide a competitive advantage
AI is blurring the boundaries between telecoms, media, chip manufacturers and software providers. Companies now compete not only on the basis of their products, but also on access to data, customer relationships and control over the digital ecosystem. This opens up new sources of revenue, but also increases the risk of funding projects whose value cannot be demonstrated.
The most important task for boards of directors is therefore not to accelerate every implementation. It is to create a mechanism that allows them to quickly distinguish between projects that improve results and costly experiments.
The industrialisation of AI begins not with a model, but with a measurable business problem, an owner accountable for the outcome, and a decision on when to scale, modify or discontinue the solution. Companies that master this process will be able to invest more boldly. The rest may develop AI for years without knowing whether they are building a competitive advantage or merely increasing costs.
