CIOs shift from AI pilots to fewer, bigger 2027 bets

More than 85,000 Travelers claim calls were eligible for a generative AI assistant, while only 24% of surveyed IT leaders had architecture to monitor agentic workflows—a readiness divide CIOs face as 2027 spending shifts from pilots to production.

Technology leaders expect fewer, larger AI bets. JPMorganChase consumer and community banking CIO Gill Haus said: “The job is to use AI to reimagine the work, not just automate the old way of doing it.” Deloitte’s “Finance Trends 2027” found 42% of finance leaders’ agentic AI ambitions exceeded current capabilities.

From pilots to production

After task-level experiments, CIOs plan to concentrate investment on five to 10 end-to-end processes. Travelers technology and operations chief Mojgan Lefebvre said the company must consider five to 10 areas “end-to-end to drive value,” warning against “demo love.”

At GAI World 2026 in September, JPMorganChase Private Bank CTO Vrinda Menon urged organizations to choose two or three big bets over thousands of pilots. Travelers’ Bond & Specialty unit cut new-business submission registration from two hours to two minutes.

Only 24% of Infotech’s 2027 Global IT Leadership Survey respondents had agentic-workflow monitoring architecture. Research director Brian Jackson said 2026 centered on automating tasks and augmenting staff; 2027 will orchestrate workflows with agentic AI.

Some CIOs will assign autonomous agents several workflow steps. Samsara, a fleet-management provider, will expand agents across selected high-value workflows in 2027. Its shared platform lets non-IT functions build and run agents; each owns outcomes and a quantified roadmap. Low-cost experiments will continue beside larger investments.

At First Student, Chief Information and AI Officer Sean McCormack reported a fivefold increase in software development throughput and an 80% cost reduction after adopting development agents. In 2027, he plans less outsourcing and an internal team to oversee agents.

Customer experience under competitive pressure

AI-native startups are reaching $1 billion valuations in about three and a half years with roughly half the staff startups once needed, according to Manish Jain, founder of Strategic Horizon Research. OneDigital co-founder Mike Sullivan calls AI a systemic risk and expects rivals to offer existing clients better service at half the price.

A GAI World attendee raised digital insurers; Travelers said they can deliver strong experiences. Menon is redesigning JPMorganChase lending workflows, some dating from the 1990s or early 2000s, by removing tasks and approvals. One process required seven people, 40 screens and seven systems; redesign and automation cut seven-day completion to a few minutes.

Forrester’s April 2026 report “Accelerate Your AI Voyage,” based on 1,500 AI decision-makers, found 52% of high AI adopters focused on customer experience, versus 44% of low adopters. At Travelers, the generative assistant targets initial loss-reporting calls; Lefebvre said more than 85,000 calls were eligible.

Lefebvre’s 2027 conclusion: improve controlled capabilities and never assume competitors cannot displace the business.

Matching models to tasks

As agents enter workflows, model selection becomes operational: frontier models for broad reasoning, narrower tuned models for classifications and defined tasks. JPMorganChase uses smaller models for categorizing information and identifying customer-request types because accuracy, speed and cost compound across thousands of runs.

Menon calls fine-tuning nearly imperative and says frontier intelligence is unnecessary for every product. Agents also need employees’ accumulated know-how, such as lending practices, available during tasks.

Travelers built TravelersLLM, a proprietary insurance model, by fine-tuning an open-weight model on company documents and deploying it in underwriting. Data scientists and subject experts ran thousands of tests; Lefebvre said answers for company-specific tasks were more accurate, 90% cheaper and five times faster. Travelers still uses frontier models for broader reasoning.

At First Student, McCormack wants tools to switch models without rebuilding because capabilities can change weekly.

Funding beyond IT

AI funding decisions will increasingly involve CIOs, finance and business leaders. Boston Consulting Group’s 2026 Applied AI Index put 80% of AI spending outside enterprise IT budgets.

Deloitte’s “Finance Trends 2027” found 54% of finance leaders lead enterprise AI and technology investment decisions. Its U.S. finance services leader Ed Hardy described decision-making as a partnership among finance, IT and business users.

That matters because consumption-priced pilots can become expensive at scale. IT and finance must estimate workflow cost, while business teams define the improvement that justifies it. Sedgwick CIO Sean Safieh says the company now identifies who owns expected business value and how outcomes will be measured before committing resources.

Post-launch, tracking can combine cost per customer resolution with service quality. At Samsara, Franchetti said the company is shifting from budgeting for licenses to tracking cost per unit of work so money follows results.

AI champion networks

Nearly four years after ChatGPT’s public debut, adoption still depends on changing routines. Samsara CIO Stephen Franchetti identifies people and process, not technology, as the biggest obstacle; mandates alone do not change how employees work.

U.S. Navy CTO Justin Fanelli wants experienced AI users across departments and teams to form a champions network for sharing notes. Samsara runs a bottom-up network drawn from every function, pairing executive sponsorship with employees closest to the work.

Champions can share where AI saved time, where human review remained necessary and what colleagues should consider. The network should reach employees not already eager to experimenters and connect AI to their jobs.

What changes for enterprises in 2027?

Choose one process with a measurable difference. Business defines the customer or employee improvement, IT assesses models, oversight and systems, and finance tests whether scaled costs justify expansion. Involve employees early. The aim is fewer deployments producing repeatable enterprise results, not more pilots.