AI in 2027: CIOs Bet on Fewer High-Impact Projects

Facing faster and cheaper competitors, CIOs are planning to focus AI on a few high-impact processes. Feedback from experience shows concrete gains, while the selection and adjustment of models become key decisions. Meanwhile, the architecture for supervising agents and the involvement of financial departments frame the ambition.

Measurable Results but Rare Architecture

Travelers has developed TravelersLLM, a proprietary insurance model adjusted from an open-weight model, and trained it on internal documents before using it in underwriting. Mojgan Lefebvre indicates that data scientists and business experts conducted thousands of tests to evaluate this system. Meanwhile, only 24% of respondents to the Infotech 2027 Global IT Leadership survey report having an architecture that allows for the supervision of agentic workflows. When these workflows are executed at scale, differences in accuracy, speed, or cost can become decisive.

Competition Puts CX First and Forces Process Simplification

AI-focused startups are bringing to market services that are either faster or cheaper, making customer experience a critical priority by 2027. Manish Jain, founder of Strategic Horizon Research, notes that AI startups are achieving valuations of $1 billion in nearly three and a half years while employing about half as many employees as before. At OneDigital, Mike Sullivan predicts that competition will directly target current customers, views AI as a systemic risk, and fears the emergence of rival proposals at half the price. When asked about neobanks, Mojgan Lefebvre acknowledges the quality of digital experiences offered by some insurers and believes this competition encourages Travelers to continue technological improvements to remain competitive in both technology and insurance. Enhancing CX also involves revisiting internal processes. Startups can design services around AI from the outset, while established companies must deal with more complex procedures, validations, and systems. At JPMorganChase, Vrinda Menon's team analyzes workflows, particularly in lending, to identify necessary actions and validations and eliminate steps. Menon emphasizes that some processes date back to the 1990s or early 2000s, and her team is now examining end-to-end journeys, even questioning the necessity of certain tasks.

Choosing and Adjusting Models According to the Task

Integrating AI agents into more workflows will make decisions about model selection more frequent by 2027. Leaders will need to arbitrate between advanced models and specialized models tailored for specific tasks. Vrinda Menon cites examples such as information classification or identifying the nature of a customer request, where smaller models, adapted to the language and categories of the business, are relevant. She believes it is almost essential to adjust models and reminds that cutting-edge intelligence is not necessary for all products. According to her, agents must also be able to access the accumulated know-how of employees and industry practices when handling a task, such as a loan request.

From a Proliferation of Pilots to a Few Major Framed Bets

CIOs plan to reduce the dispersion of experiments in 2027 to concentrate their efforts on a limited number of critical processes. Vrinda Menon describes a landscape saturated with pilots and recommends refocusing on two or three major bets per organization, a position expressed during a panel with Mojgan Lefebvre at the GAI World 2026 conference. However, expanding the role of AI agents requires having supervisory systems, while 42% of financial leaders surveyed by Deloitte believe their ambitions for agentic AI exceed their organization's current capabilities. This gap could slow expansion and encourage more limited scope choices. The selection of projects will involve reducing supported experiments. Travelers illustrates the possible gains by having reduced the time to register submissions from two hours to two minutes and targeting between five and ten priority areas, according to its Chief Technology and Operations Officer, Mojgan Lefebvre.

Budgets and Change Management Impact 2027 Priorities

The involvement of financial and commercial departments in spending decisions related to AI is among the priorities put forward for 2027. Networks of AI champions are being considered to support the evolution of employee practices. Overall, CIOs aim for greater gains with more selective investments, focused on well-identified processes and task-adapted models. Gill Haus, CIO of consumer and community banking at JPMorganChase, advocates for using AI to reinvent work rather than simply automating existing methods.