"Enterprise AI implementation seen from the 2025 New York AI Leader Summit: Multi-cloud strategy and small models becoming mainstream choices."

Deutsche Bank recently released a research report stating that after attending the 2025 New York Artificial Intelligence Leaders Summit, the bank is more convinced that companies are still in the early stages of developing their AI transformation roadmaps. Deutsche Bank pointed out that the summit attracted over 50 technology business leaders and practitioners from a wide range of vertical sectors to participate in group discussions and breakout sessions, which revealed: 1) a lack of consensus on measuring return on investment; 2) data readiness remains key and is a major limiting factor in whether companies can fully leverage AI benefits; 3) regulatory and governance policies are still a focus compared to the operationalization of AI. Although specific vendors were rarely discussed at the summit, the bank expressed that it did feel that packaged software could play a role in future architectures, as many organizations seem unprepared or lack the expertise to take a DIY approach. Here is an overview of Deutsche Bank's key takeaways from the summit: 1) Return on investment (ROI) remains a moving target across the entire enterprise, with business leaders choosing/defining their own key metrics. Legacy systems are expected to deliver the greatest benefits but also face the biggest adoption challenges as they lag behind several generations. 2) While quantifying time savings is part of a broader ROI discussion, optimists believe that a significant portion of the value lies in the entirely new capabilities unlocked by AI. 3) From an adoption perspective, the majority of customers (about 80%) are still in the stage of optimizing existing business processes, while the remainder (about 20%) are more willing to experiment. 4) Data readiness remains key for enterprises, with the fundamental issue being the management team's understanding of their data and the location of data storage. Interestingly, it is claimed that only 10%-20% of total time is actually spent on training models, with the rest focused on data preparation, indicating that the quality of the model depends on its input data. 5) Data cataloging remains at the core of AI interpretability. Cataloging is cited as a common blind spot for companies in preparing AI processes, as if cataloging fails to keep up with the evolving data over time, there will be a continuous inconsistency in the data fed into large language models. 6) Many leaders believe that maintaining human involvement in agent processes is still crucial. Deutsche Bank understands that the most important thing is the need for sanity checks, making human involvement key in agent workflows. 7) Regulatory uncertainty and governance policies are identified as barriers to the speed of enterprise-wide adoption of AI. From a security perspective, the focus remains on improving disaster recovery policies and reducing shadow AI. 8) Deutsche Bank noted that there is a preference for Small Language Models (SLMs) over Large Language Models (LLMs), as they allow for full control of model operation location and increased efficiency. It is claimed that models trained by vendors have too many parameters and improper context, leading to higher total ownership costs and lower contextuality responsiveness. 9) Multi-cloud seems to be the preferred strategy for enterprises, with business leaders leaning towards an "optimal breed" approach. Deutsche Bank did not find a one-size-fits-all solution, as the bank understands that analyzing the enterprise's infrastructure mix requires a case-by-case analysis and involves other decision criteria. 10) Low-risk and repetitive workloads are said to be the areas leading the way in AI utilization, with the most common applications being AI search and analysis. While companies focus on customer-facing projects, business leaders believe there is still significant disruptive potential in back-end functions, with great opportunities for AI to create value. Furthermore, some key points from the voting surveys conducted at the summit include: 1) 73% of participants believe that their organizations are at varying stages in their AI application journey, with only a very few systems entering production systems (18%) and early pilot phases (9%). 2) 70% of participants prioritize balancing AI innovation with security as their top solution priority, with the rest evenly divided between rapid deployment, risk management, and regulatory compliance. 3) The biggest obstacle to creating a seamless AI-driven customer experience is still legacy system integration (56%), followed by unclear return on investment (33%) and data silos (11%).