Most companies struggle to scale AI despite rapid adoption, says SparkOptimus
AI adoption has surged across organizations worldwide, but the gap between testing and scaling remains a stubborn challenge for most businesses, according to a new industry benchmark published by SparkOptimus.
The report, which draws on interviews with more than 50 decision-makers across finance, energy, consumer goods, industrial, and professional services sectors, paints a picture of an industry at a turning point.
Agentic AI adoption has increased dramatically, rising from 45% of organizations surveyed in 2025 to nearly all of them in 2026, according to the report. Daily AI use among employees also rose sharply, with 40% now using AI tools every day, a trend that has been steadily increasing in the last few years. More companies also reported having a clear view of how AI affects their core business proposition.
Yet despite this momentum, the path from idea to scaled deployment remains steep. According to the benchmark, just 15% of AI initiatives ever manage to truly scale in the end. In many cases, projects end up getting stuck in what some call pilot purgatory’.
SparkOptimus attributes this to a number of stubborn barriers, including insufficient AI literacy among management and a near-total absence of quantified ambitions at most companies. Meanwhile, there is a clear lack of standard scaling governance and most surveyed leaders also point to a skills gap in their teams.
“AI developments are moving fast due to increasing use of Agentic AI, more mature models, and model democratization,” said Matti van Engelen, associate partner at SparkOptimus.
“All organizations are focusing on productivity gains, while front-runners are reinventing their proposition to disrupt industries. There are concrete actions that organizations can take to make sure they don’t get left behind.”
Agentic AI and customer proposition
The report identifies the rise of Agentic AI as one of three forces reshaping the landscape in 2026, alongside the growth of function-specific models and the democratization of AI access through mainstream software platforms. The report notes that while most organizations have already begun experimenting with basic agents, very few have progressed to testing true multi-agent systems, in which several agents collaborate across an entire process.
In contrast with traditional AI tools, Agentic AI stands out for its ability to make autonomous decisions and actions, like setting goals and completing tasks with very little human intervention. Multi-agent systems, in which multiple AI agents interact and work as a team, have a lot to offer.
The vast majority of companies noted the impact AI has had on the customer proposition. Despite that, awareness has not always translated smoothly into action, with many companies still anticipating significant disruptions. Few companies are tracking how customer expectations are evolving and disruption risk is often assessed against today’s customers, not future customers.
A widening gap between leaders and laggards
SparkOptimus highlights a growing divide between organizations that have embedded AI into their core strategy and those still running isolated experiments. Companies that have established a formal data strategy move five times as many use cases from pilot to scale compared to those without one.
Those with a centralized AI technology stack scale twice as many use cases. Customer-focused use cases deliver more impact and reach scale more rapidly as those focused solely on internal efficiency.
SparkOptimus also found that the AI skills gap, while improving, remains the single largest barrier to scaling. The report suggests more targeted hiring, internal upskilling, and easier-to-use tools as potential solutions to a lack of skills preventing teams from keeping up.
In order to succeed with AI scaling, the report suggests companies test smaller use cases, allowing themselves to fail and learn going forward, then scale the projects that work. The study also noted that when companies built their own agentic and Gen AI solutions, more use cases made it from idea to pilot. The same was true of those with their own agentic and Gen AI tech stacks.
The companies leading the way in AI tend to anchor initiatives in business needs rather than IT departments, with the majority of top-scaling companies starting pilots from a clear business problem. The study notes business leaders should avoid chasing shiny use cases or blindly copying competition without taking into account the company’s strengths and unique selling proposition. Companies must also understand the limitations of AI and the challenges they will face adopting it.
Organizations leading in AI also tend to test in short, structured cycles and they ensure leadership is genuinely AI-literate, since companies with fully AI-literate management focus twice as much on transforming their customer proposition rather than chasing internal efficiency alone.
The report warns that first movers are already beginning to reshape customer expectations across industries, and that organizations slow to act risk being disrupted rather than empowered by the shift.
“At SparkOptimus, we’ve seen in the past years how Gen AI Agentic AI can unlock meaningful business value, driving efficiency, and enabling new ways of working,” said Van Engelen.
“There has been clear progress this year: More areas are seeing use cases realizing impact at scale, and we’re seeing first signs of AI threatening markets and reshaping propositions. Yet scaling remains difficult for many, especially in Agentic AI. Impact often trails ambitions, and large – scale adoption is challenging for most.”
