The most treacherous technology pilot is not the one that ends in failure, but the one that works and still gives the company no reason to do anything with it.
In recent years, proof of concept has become the default response of businesses to almost every new technology. Generative AI, agents, automation, digital twins and new data platforms are first ‘tested’, often on a small group of users and outside core processes. Whilst this helps to mitigate risk, it has also created a whole portfolio of experiments that can prove the technology works, without answering the question of whether it is worth implementing.
In 2026, this distinction begins to take on greater significance. IDC estimates that spending on AI in the EMEA region will reach $319 billion, 19.2 per cent more than the previous year. At the same time, the research firm points to a marked increase in budgetary selectivity driven by economic uncertainty, regulation and geopolitical tensions. Funding for technology is on the rise, but so is the pressure to demonstrate exactly what that funding is intended to buy.
A PoC may work and still be a failure
Technical success is a poor investment criterion. A model may generate responses correctly, an agent may perform tasks, and an automation system may streamline a single operation. None of these things guarantees that the solution will withstand integration with production systems, meet security requirements, handle real-world data volumes or cope with maintenance costs.
Gartner reported in January that at least half of generative AI projects were abandoned after the proof-of-concept phase. The reasons included data quality, inadequate risk control mechanisms, rising costs and unclear business value. By July, the firm was already speaking explicitly of ‘AI pilot fatigue’ and the problem of metrics that look good in an isolated test but cease to hold true once the solution is transferred to a real-world operational environment.
This is one of the main pitfalls of modern innovation programmes. A pilot is designed to test the technology, although the focus of the investment should be on the business outcome.
The difference is fundamental. “Let’s test AI agents in customer service” may produce an interesting demonstration. “Let’s see if the agent can handle 30 per cent of a specific type of enquiry whilst maintaining current quality and reducing transaction costs” provides the basis for a decision.
A project without an owner easily loses its purpose
A second warning sign is blurred accountability between the innovation department, IT, the supplier and the business. A project may have a sponsor, a steering committee and a budget, yet still lack a person whose financial performance will be affected by its implementation.
BCG’s data clearly illustrates the scale of the problem. In a survey published in July 2026, 64 per cent of CEOs stated that they were conducting AI pilot schemes, but only 26 per cent of companies had integrated AI into a broader corporate transformation. More than half of company leaders pointed to a lack of a clear link between AI and the profit and loss account, whilst only 14 per cent of organisations identified an impact on the P&L for all AI initiatives.
This is not an argument against experimentation. It is an argument against experiments that lack a clear business objective.
The best-performing companies do not simply run a greater number of projects. According to BCG, they are around seven times more likely to overhaul entire workflows. PwC reaches similar conclusions: 20 per cent of the companies surveyed capture 74 per cent of the economic value generated by AI, and leaders are twice as likely to overhaul processes rather than simply adding new tools to them.
McKinsey identifies the same pattern in Central Europe. AI is already being used by 88 per cent of organisations surveyed worldwide in at least one area, but 94 per cent have not yet achieved a significant impact on EBIT. The firm estimates the economic potential of the technology in Central Europe at between €280 billion and €700 billion; however, realising this potential requires process re-engineering, rather than simply multiplying individual applications.
Innovation should culminate in a decision
The value of an experiment therefore does not depend on whether it culminated in implementation. A successful pilot may demonstrate that the technology is too expensive, the data too poor, integration too difficult, or the market not yet ready. In such a case, the rational outcome is to close the project.
An experiment that leads to no decision whatsoever is far more costly.
The line between a strategic pilot and a trendy project lies precisely here. The former has a clearly defined change it is intended to prove, a measurable benchmark and an owner of the outcome. The latter primarily demonstrates that the company has also taken an interest in the technology everyone is talking about.
In an environment of rapidly growing technology budgets, the mere ability to experiment is no longer a competitive advantage. What is becoming increasingly valuable is the ability to quickly identify which experiments deserve the capital needed to scale up, and which should be concluded once the results have been presented.
