The headlines promise transformation. CEOs announce ambitious AI strategies. Tech vendors showcase impressive demos. But behind the hype, most organizations are hitting walls that have nothing to do with the technology itself. McKinsey’s latest research shows that while 88% of companies now use AI regularly, only 39% report any measurable impact on enterprise earnings. The gap between adoption and value has never been wider.
As Ivo Bozukov observes, “What’s striking isn’t how many companies are using AI, but how few are seeing real returns from it. The technology often works exactly as promised, yet value stalls because organizations aren’t ready to change how decisions, incentives, and accountability actually work. That gap between adoption and impact is where most AI strategies break down.”
The ROI Problem No One Talks About
Executives greenlight AI projects expecting quick returns. But what they often get are pilots that work brilliantly in controlled environments but collapse when scaled across the business. Klarna offers a cautionary tale. The Swedish buy-now-pay-later company partnered with OpenAI in 2024 and publicly declared that its AI chatbot could do the work of 700 customer service agents. CEO Sebastian Siemiatkowski went further, stating that “AI can already do all of the jobs that we, as humans, do.” But by mid-2025, the company was backtracking. In a Bloomberg interview, Siemiatkowski admitted the company “went too far” and that “cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality.” Klarna is now actively recruiting human customer service agents again, with Siemiatkowski acknowledging that “from a brand perspective, a company perspective, I just think it’s so critical that you are clear to your customer that there will always be a human if you want.” The company that once touted AI-driven cost savings is now piloting an Uber-style remote work model to rebuild its human support team.
BCG finds that 74% of companies struggle to turn AI investments into measurable value. Part of the problem is that traditional financial metrics don’t capture AI’s experimental nature. A chatbot might handle thousands of customer queries, but if it can’t quantify cost savings or revenue impact, it is hard to justify continued investment.
MIT research reveals that 95% of generative AI pilots at companies are failing to move beyond the testing phase. In capital-intensive environments where investment decisions demand clear justification, uncertainty around AI returns often slows momentum. Ivaylo Bozoukov encounters this hesitation even among technically sophisticated organizations.
People Problems Trump Technical Ones
BCG’s analysis shows that roughly 70% of AI implementation challenges stem from people and process issues, with only 10% involving the algorithms themselves. Yet organizations spend disproportionate time on technical problems while ignoring the harder work of changing how teams operate.
40% of enterprises report lacking adequate AI expertise internally. Rather than a shortage of data scientists, Ivo Bozukov finds that what is missing are people who understand both the business problem and what AI can actually do about it.
Employee resistance compounds the challenge. Workers fear job displacement. Managers worry about losing control to automated systems. Without transparent communication about AI’s role, these fears create passive resistance that stalls implementation regardless of technical readiness.
Infrastructure Reality Meets Legacy Systems
The data problem hits after projects start. Companies commit to an AI initiative, then discover their customer data lives in three different formats across five systems. The cleanup work can take longer than building the actual model.
This is a pattern that Ivaylo Bozoukov encounters frequently as companies modernize infrastructure to support continuous data and AI workloads. Existing systems were designed for batch processing. AI needs continuous data streams and heavy computation. Most IT architectures can’t handle both without major overhauls.
Governance and Trust Gaps
When an algorithm denies someone a loan, who’s accountable for that decision? Most organizations don’t have answers because the frameworks don’t exist yet.
Banks face regulatory pressure to explain algorithmic decisions. Healthcare systems need doctors to be able to verify what AI recommends. Until governance catches up, AI stays in applications where mistakes don’t matter much.
Trust matters to customers. A chatbot that can’t answer basic questions or recommendations that make no sense damages credibility quickly. Trust issues often slow AI adoption just as much as technical barriers. Ivaylo Bozoukov runs into this repeatedly in enterprise deployments, where accountability lags the technology.
The Path Forward Exists
These struggles don’t prove AI is overhyped. Most organizations are still figuring out how to work with something fundamentally different from traditional software.
Companies making progress stopped treating this as purely a technology problem. Training happens across departments. Broken processes get fixed before automation starts. Timelines reflect reality instead of vendor promises.
The organizations that solve these problems gain advantages that competitors can’t replicate quickly. The gap between companies that make AI work and those stuck running pilots widens every year.
