The gap between AI investment and actual deployment remains striking, and many organizations are still searching for effective ways to turn promising AI budgets into tangible results. But here's where it gets controversial: despite widespread enthusiasm and significant funding commitments, only a small fraction—just 18%—have fully implemented agentic AI solutions. This discrepancy raises crucial questions about what’s holding organizations back, and whether current strategies truly set them up for successful AI integration.
Qlik recently published its third annual report on Artificial Intelligence (AI), in collaboration with Enterprise Technology Research (ETR). The study surveyed over 200 decision-makers responsible for enterprise technology, aiming to understand how businesses are progressing from initial AI experiments to deploying smarter, more autonomous systems known as agentic AI. The findings reveal that a staggering 97% of large enterprises have allocated budget towards agentic AI projects, with 39% planning investments exceeding one million dollars. Yet, despite this high level of commitment, only 18% have actual operational deployments.
James Fisher, Qlik’s Chief Strategy Officer, candidly shares his surprise at the persistent gap: “We see a lot of talk about AI and agentic AI, but actual adoption at scale remains limited. Turning these promising ideas into measurable benefits is still a challenge, and that’s something we need to address.”
Looking back at Qlik’s 2024 research, the progress is significant—fewer organizations, about 37%, had formal AI strategies then. Today, that percentage has nearly doubled to 69%. Still, most organizations believe it will take three to five years before they can confidently operationalize AI at scale, highlighting the slow yet steady pace of evolution.
Budget constraints and fragmented funding sources are major hurdles in accelerating AI deployment. Fisher explains that enterprise IT budgets are often under constant pressure, and allocating substantial new funds to AI—especially agentic AI—becomes a tough decision. While 56% of organizations report having dedicated budgets for AI innovation, a majority still rely heavily on existing IT funds (60%) or line-of-business budgets (42%). Interestingly, 79% of respondents consider agentic AI critical to their strategic plans over the next three to five years.
Yet, foundational data issues and skills shortages pose significant barriers. More than half of the respondents (56%) cite data quality, access, and availability as primary obstacles. Paradoxically, although 77% express confidence in distinguishing agentic AI from other tools, only 42% believe their organization possesses sufficient internal expertise to design and deploy such systems independently—highlighting a skills gap. Fisher notes that data literacy and quality have long been challenges, impacting not only traditional AI and generative AI but now the deployment of agentic solutions as well.
When it comes to integrating AI with existing systems, nearly half (49%) cite system compatibility issues, and a similar number (48%) point to a lack of internal expertise. Interestingly, only 13% mentioned multi-agent systems specifically when discussing agentic AI; most focus on autonomous decision-making (61%) and task automation (49%).
Security and governance concerns dominate deployment reservations. Cybersecurity vulnerabilities are the top concern, with 61% citing them as barriers, especially around deploying AI in IT operations—the primary target for 72% of organizations. Legal and compliance issues are also significant, with 51% citing risks related to governance and regulation, along with 47% citing challenges related to explainability and auditability of AI decisions.
Fisher emphasizes that stakeholder involvement is evolving: “While cybersecurity and governance teams have traditionally been engaged, legal teams are now increasingly taking a seat at the table, which is vital for responsible AI deployment.” He underscores that systematic policy development and clear governance frameworks are crucial starting points.
Despite these challenges, there are promising signs of success when organizations take a pragmatic approach. Fisher shared several inspiring real-world examples:
- A North American specialty chemicals distributor built a generative AI assistant to support sales and customer service within just two months by connecting existing data repositories in Qlik and SharePoint. About 40 employees now use it daily for complex product queries, enabling the company to onboard new staff more effectively.
- An industrial manufacturer with 3,500 workers set up knowledge bases for technical manuals and project documents in roughly 15 minutes per base, demonstrating how quick setup can be when leveraging existing data rather than undertaking large, costly data-engineering projects.
- An entertainment enterprise operating cinemas and theme parks in Asia Pacific improved their attendance forecasts from around 70% accuracy to over 90% by layering predictive analytics onto their existing Qlik Cloud platform. These improved forecasts support near real-time scheduling and operational decisions.
- A European food producer developed AI-driven demand forecasts for organic meat products, reducing forecast deviations to approximately 1%. This precision helped avoid overproduction, minimized product downgrades, and reduced storage costs—all while supporting sustainability objectives.
Fisher sums it up: “Today, we have everything needed to demonstrate AI value. The real challenge is understanding how to deploy these solutions across different parts of an enterprise without having to overhaul entire systems or resolve every data quality issue beforehand. Connecting AI to existing foundational data and starting small can produce significant early wins.”
So, what’s the key takeaway? Despite大的 investment commitments, the impressive early examples suggest that success often comes from simple, focused applications that connect AI to existing data and processes—without multi-year transformations. The ambiguity around what defines agentic AI, especially with only 13% mentioning multi-agent systems, may contribute to organizations underestimating the scope and potential of these technologies.
Policy, governance, and clearly scoped projects seem to be the most effective starting points. Should organizations continue to chase large-scale, enterprise-wide AI initiatives, or prioritize smaller, targeted deployments? Based on these insights, the latter might be the smarter strategy for practical success—and perhaps, for turning AI promises into real-world benefits.
