In the corporate space, the topic of artificial intelligence has dominated the narrative. Stakeholder expectations are huge and marketing messages suggest that the technological revolution has already taken place. However, an analysis of the actual state of deployments reveals a different picture: most companies, despite hype declarations, are still operating in the realm of testing. Instead of strategically scaling innovation, the market is seeing the phenomenon of ‘perpetual piloting’, where the pressure to demonstrate modernity obscures a lack of operational readiness.
Despite the prevalence of discussions about the impact of AI on business models, the current discourse in many organisations is slipping away from the technological reality. Although rarely discussed officially, a significant number of companies are stuck in the early experimentation phase. Rather than consistently implementing productive solutions, boards often focus on demonstrating ‘innovation courage’, but this is not accompanied by the confidence needed to deeply integrate technology into the business.
Failure statistics versus market pressure
The scale of the problem is illustrated by hard data. An investigation by MIT’s NANDA initiative revealed that as many as 95% of AI pilot programmes fail to deliver the expected business results or fail completely. This is a rate that in any other investment area would be considered unacceptable.
The high failure rate is largely due to the fact that companies try to implement AI solutions under pressure from the environment, without adequate preparation. Pilot programmes are often treated as an end in themselves – proof of an organisation’s modernity – rather than as a prelude to real transformation. The result is projects that, although technologically advanced, do not generate a return on investment (ROI) and have no chance of moving beyond the test environment.
For IT and business decision-makers, this means a paradigm shift in how success is assessed. In the current reality, simply launching an AI initiative is no longer a market differentiator. The real challenge – and measure of success – is becoming the ability to move solutions from a secure ‘sandbox’ to a production environment.
The data barrier: Information architecture as a foundation
When analysing the reasons for failure, it is important to look at the foundation, which is data. Generative artificial intelligence, LLM models or predictive analytics are entirely dependent on the quality and availability of the data on which they operate. Meanwhile, managing this resource in the era of AI is becoming a challenge beyond previous standards.
Global data volumes are estimated to reach 181 zettabytes this year. Organisations are struggling with information overwhelm, and the problem is exacerbated by the structure of these resources. According to Gartner analysts, 80% of business data is unstructured. Before the era of AI, these resources were typically archived and secured, with no attempt at deep analysis. Now that technology is making it possible to extract value from them, the lack of proper categorisation and governance is revealing.
Introducing AI algorithms into an unstructured data environment is one of the main reasons why pilot projects fail. Without first ensuring data visibility and resilience (data resilience), organisations risk building innovation on unstable ground.
Investment in ‘data hygiene’ has ceased to be a purely technical issue and has become strategic. Existing security measures often prove insufficient when confronted with the demands of modern AI models. Without structuring and validating data, any attempt to implement advanced analytics is doomed to remain in the realm of theory.
Shadow IT: The risk of apparent control
Delays in official implementations and the failure of pilot programmes have serious security implications, known as the Shadow IT phenomenon. Employees, recognising the potential of AI tools to streamline their daily work, often do not wait for authorisation from the head office.
When official innovation paths are blocked or inefficient, teams start experimenting with publicly available tools on their own, outside the control of security departments. This creates the illusion that the organisation has an airtight AI policy, when in reality, data flows through unauthorised channels. Until companies manage to get their resources in order and provide secure, in-house alternatives, this phenomenon will increase, generating the risk of sensitive information leaks.
Shadow IT in the context of artificial intelligence is a wake-up call for boards. It indicates that there is a ‘demand’ for innovation within the organisation that official structures are unable to meet. The role of leaders is to redirect this grassroots energy to safe tracks, rather than ignoring it.
Evolutionary strategy: From order to innovation
Treating AI as a ‘new era’ does not absolve the obligation to take care of the foundations of the previous era. Experts point out that the key to success is not to reject existing procedures, but to adapt them.
A recommended approach is to change the vector of initial deployments. Instead of focusing on impactful client applications, it is worth harnessing the potential of AI for clean-up work. AI excels at data classification processes, mapping data flows and enhancing digital resilience. A company’s first AI project should therefore serve to tidy up the ‘data landscape’. Only when the algorithms help bring the information chaos under control will it become possible to safely scale more advanced solutions.
Sustainability instead of revolution
An all-or-nothing approach rarely works in digital transformation. The solution to implementation stagnation is a strategy of small steps. It is not necessary for an organisation to immediately become a market pioneer in every aspect of AI. The key is to demonstrate the ability to generate value while maintaining full control over processes.
It is recommended to start with precisely defined initiatives where AI can safely and measurably improve processes. Successes on a smaller scale build the organisation’s confidence (trust) and provide the proof of concept needed for subsequent implementation of transformational solutions. At each stage, verification of the model’s compliance with cost, performance and security requirements is essential.
The gradual build-up of competencies helps to negate the paralysing fear of failure that blocks the decision-making of many boards. Maintaining a balance between control and innovation, with operational resilience, appears to be the only effective way out of the ‘perpetual pilot’ phase.
