As we enter the season for predictions and trend analyses, data science and artificial intelligence are becoming increasingly vital to the global economy. It’s crucial for leaders to stay informed about emerging AI trends.
While AI is not being used to make these predictions (and we won’t use it either), we’ll highlight the AI trends expected to shape 2025, incorporating the latest research wherever possible. Randy has just completed his annual survey of data, analytics, and AI executives through the 2025 AI & Data Leadership Executive Benchmark Survey, organized by his educational firm, Data & AI Leadership Exchange. Meanwhile, Tom has worked on several surveys focused on generative AI, data, technology leadership structures, and the latest development in agentic AI.
Here are the AI trends for 2025 that leaders should be aware of and closely monitor.
1.Leaders will face the challenge of navigating both the potential and the hype surrounding agentic AI.
Let’s address agentic AI, the type of AI that operates independently, right away: It’s poised to be the top AI trend of 2025. Agentic AI appears to be on an upward trajectory, with excitement across the tech vendor and analyst communities about AI programs working together to perform tasks rather than just generating content, though there’s still uncertainty about how it will all come together. Some IT leaders believe they already have it (37%, according to an upcoming UiPath-sponsored survey of 252 U.S. IT leaders); most are expecting it soon and are ready to invest (68% within the next six months); and a few skeptics, particularly those we’ve encountered in interviews, feel it’s mostly vendor-driven hype.
The majority of technology executives believe these autonomous, collaborative AI systems will be based on targeted generative AI bots that handle specific tasks. It’s widely expected that there will be a network of such agents, with many hoping that these ecosystems will require less human involvement than AI systems in the past. Some envision robotic process automation tools orchestrating the technology; others suggest that agents will be integrated into enterprise transaction systems; and some even foresee the rise of an “uber agent” that will manage everything.
Here’s what we believe: There will be, and in some cases already are, generative AI bots designed to handle specific content creation tasks. However, it will take more than one of these agentic AI tools to complete more significant actions, such as making a travel reservation or performing a banking transaction. These systems still function by predicting the next word, which can sometimes lead to errors or inaccuracies. As a result, humans will still need to check in on them from time to time.
The first agents will be used for small, structured internal tasks with minimal financial involvement, such as assisting with password changes in IT or reserving vacation time in HR systems. We don’t expect companies to fully rely on these agents for handling real customers and financial transactions anytime soon, unless there’s an option for human review or transaction reversal. Therefore, we don’t anticipate a major impact on the human workforce from this technology in 2025, except for the creation of new roles writing blog posts about agentic AI. (Wait, can agents do that?)
2. It’s time to measure the outcomes of generative AI experiments.
One of the reasons behind the excitement around agents is that, as of 2024, demonstrating the economic value of generative AI has proven challenging. In last year’s AI trends article, we pointed out that the value of GenAI still needed to be proven. However, data and AI leaders in Randy’s 2025 AI & Data Leadership Executive Benchmark Survey expressed confidence that value is being generated. Fifty-eight percent reported that their organization has seen significant productivity or efficiency gains from AI, likely driven by generative AI. Another 16% mentioned that GenAI tools have helped “liberate knowledge workers from mundane tasks.” Let’s hope these optimistic views hold true.
However, companies shouldn’t rely solely on such confidence without verifying it. Very few companies are actively measuring productivity improvements or analyzing what knowledge workers are doing with their freed-up time. While some academic studies have measured GenAI’s productivity benefits, they have generally found improvements, but not exponential ones. Goldman Sachs, for example, is one of the few companies that has tracked productivity gains in programming, with developers reporting a 20% increase in output. Other similar studies have shown that less experienced workers gain more (as in customer service and consulting), while more experienced workers see better results (such as in code generation).
To accurately measure productivity gains, controlled experiments may be the best approach. For instance, a company could have one group of marketers use generative AI for content creation without human review, another group use it with human review, and a control group refrain from using it entirely. However, few companies are conducting such experiments, and that needs to change. Since GenAI is primarily focused on content generation for many organizations, evaluating content quality will also be necessary. This is difficult to measure, especially with knowledge work output. For example, if GenAI allows blog posts to be written faster but the posts are dull and inaccurate, that’s something that should be measured—there will be little value in that case.
The unfortunate reality is that if many organizations are to achieve exponential productivity improvements, those gains may be reflected in large-scale layoffs. However, current employment statistics do not show signs of such mass layoffs. Additionally, Nobel Prize-winning economist Daron Acemoglu from MIT has stated that we haven’t seen substantial productivity gains from AI yet and doesn’t expect any dramatic changes over the next several years, with perhaps only a 0.5% increase in productivity over the next decade. In any case, for companies to truly benefit from GenAI, they will need to measure and experiment to understand its value.
3. The reality of a data-driven culture begins to emerge.
It seems that while generative AI is impressive, it doesn’t have the power to revolutionize everything, particularly long-term cultural changes. In our trend article from last year, we mentioned that Randy’s survey revealed a sharp increase in the percentage of companies claiming to have “created a data and AI-driven organization” and “established a data and AI-driven organizational culture.” These figures doubled from the previous year, with organizations reporting growth from 24% to 48% for having a data- and AI-driven organization, and from 21% to 43% for fostering a data-driven culture. We were surprised by this rapid reported improvement, attributing much of it to the widespread and fast adoption of generative AI.
Our long-term prediction, however, is that generative AI alone won’t be sufficient to create truly data-driven organizations and cultures.
This year, the numbers have returned to a more realistic level. Thirty-seven percent of respondents said they work in a data- and AI-driven organization, and 33% said they have a data- and AI-driven culture. While it’s still encouraging that data and AI leaders feel their organizations have made progress over the past few years, our long-term outlook remains the same: generative AI on its own will not drive data-driven transformations in organizations or cultures.
In the same survey, 92% of respondents identified cultural and change management challenges as the main barriers to becoming data- and AI-driven. This indicates that technology alone isn’t enough. It’s also worth noting that most of the respondents came from legacy organizations, many of which were founded decades ago and have traditionally evolved slowly. These companies made significant strides in executing their digital strategies during the pandemic, more than in the previous twenty years.
4. Unstructured data gains significance once again.
Generative AI has influenced organizations in another way: it’s bringing unstructured data back into the spotlight. According to the 2025 AI & Data Leadership Executive Benchmark Survey, 94% of data and AI leaders believe that the growing interest in AI is driving a stronger focus on data. Since traditional AI has been in use for many years, it’s clear that the impact they’re referring to is from generative AI. In another survey mentioned in last year’s AI trends article, there was strong evidence showing that most companies hadn’t yet begun to properly manage their data in preparation for generative AI.
Much of the data that generative AI works with is unstructured, such as text, images, video, and similar forms. A leader at a major insurance company recently shared with Randy that 97% of their data was unstructured. Many organizations are now looking to leverage GenAI to help manage and make sense of their data and documents, often through a method called retrieval-augmented generation (RAG). However, some businesses haven’t paid much attention to unstructured data since the days of knowledge management more than 20 years ago, focusing instead on structured data—usually rows and columns of numbers from transactional systems.
To get unstructured data organized, companies need to select representative examples of each document type, tag or graph the content, and input it into the system. (This is where the complex world of embeddings, vector databases, and similarity search algorithms comes in.) While these methods offer significant knowledge access benefits for employees, they are still resource-intensive. In the future, perhaps, we’ll be able to upload large amounts of internal documents into a generative AI prompt window, but 2025 likely won’t be that year. Even when this becomes possible, human curation will still be necessary after all, ChatGPT can’t decide which of 20 different sales proposals is the best.
5.The Struggle for Data and AI Leadership Continues
It’s perhaps not surprising that, despite the increasing attention and investment in data and AI within organizations, the leadership function for these areas continues to face challenges. The role is still in its early stages only 12% of organizations had appointed a chief data officer when Randy conducted his first annual executive survey in 2012. However, progress has been made: 85% of organizations now have a chief data officer, and a growing number of these leaders are focusing on growth, innovation, and transformation, rather than just managing risks or regulatory concerns. Additionally, 33% of organizations have introduced a chief AI officer role, which is a notable increase.
Despite the evolution of these roles, organizations are still grappling with defining their responsibilities, mandates, and reporting structures. Less than half of data leaders (mostly chief data officers) in Randy’s AI & Data Leadership Executive Benchmark Survey felt their function was well-established and successful, and only 51% believed their role was clearly understood within their organizations. There is still uncertainty over whether the roles of chief AI officer and chief data (and analytics/AI) officer should remain separate, though some companies, like Capital One and Cleveland Clinic, have made the chief AI officer a peer to the chief data officer.
One thing is certain: the demand for data and AI leadership will continue to grow, regardless of the structure or form it takes.
Randy has a clear perspective on the future of the chief data and AI officer role. He strongly believes that the chief data and AI officer should be a business leader, reporting directly to business leadership. He points out that 36% of data and AI leaders in his survey this year reported to the CEO, president, or COO. Randy insists that data and AI leaders must demonstrate measurable business value and be able to communicate effectively with business stakeholders.
Conclusion
2025 is poised to be a pivotal year for AI and data science, with several key trends set to shape the industry. As businesses continue to explore the potential of agentic AI, the challenge lies in navigating both its possibilities and the hype surrounding it. Generative AI, while still in its experimental phase, will see a stronger focus on measuring its tangible outcomes, particularly in terms of productivity and efficiency.
At the same time, the promise of a fully data-driven culture is beginning to take shape, although cultural and change management barriers remain significant. Furthermore, the increased emphasis on unstructured data highlights the need for organizations to rethink how they manage and leverage this vital asset.