As corporate adoption matures, enterprise AI implementation is entering a more demanding phase where financial accountability takes center stage. Organizations that once deployed digital assistants and copilots freely are now establishing strict frameworks to track tangible returns. Across banking, insurance, and semiconductor manufacturing, leadership teams are discovering that scaling artificial intelligence requires more than just distributing software tools. Proving what these technologies earn has become the central challenge for executives balancing growing operational costs against elusive productivity gains.
What Changed in Corporate Deployment
For years, corporations distributed generative tools and digital assistants across their workforces, leading to widespread productivity claims that remained difficult to verify or record in corporate books. According to Marco Bressan of Banco Santander, organizations frequently encounter the ‘adoption trap.’ While individual workers may experience productivity gains from AI assistants, those gains are often difficult to justify financially because employees end up shifting to other tasks or generating extra work for colleagues. This dynamic contributes to heavy token consumption without clear bottom-line improvements.
In response, financial leaders are tightening controls. At AMD, the AI token bill grew significantly during agent deployment, prompting the chief financial officer to condition ongoing funding on explicit return on investment proof. Leadership explicitly stated that funding would continue only when the organization is sure of getting the return on investment.
Key Benefits and Financial Frameworks
To counter unverified productivity claims, major financial institutions and enterprises are building rigorous measurement strategies. ABN Amro now requires every live AI use case to feature a formal business case built on five specific measures: lower costs, business growth, higher customer satisfaction, better use of capital, or reduced risk. To maintain structural integrity, ABN Amro separates its AI agents into conversational, reasoning, and task layers to ensure proper policy checks are enforced.
Financial targets are also scaling up across the financial sector. Banco Santander committed to generating more than €1bn in business value from artificial intelligence between 2026 and 2028 through a combination of extra revenue and lower costs. The bank reported €35m in value during the first quarter and expects to exceed €200m by year-end. Similarly, insurance provider Axa presented a strategic plan targeting €500m to €700m in annual recurring pre-tax benefits from scaling AI by 2029.
Practical deployments are demonstrating verifiable value when carefully targeted. Pernod Ricard deployed an AI recommendation tool for sales representatives across more than 28 countries, utilizing A/B testing to attribute a few million euros in direct benefit. Meanwhile, Albert Heijn tested a staff app assistant concept by having human teams manually answer questions in a store back room before building the actual AI assistant, which now handles approximately 80,000 conversations a week.
Sector Impact and Operational Transformation
Beyond customer-facing chatbots, companies are redesigning core workflows. ING Belgium rolled out an AI tool designed to extract company information from the Belgian Official Gazette and is building a second agent to check incoming business loan applications for missing documents. Decathlon utilized a framework weighing revenue potential, risk, capacity, and measurability before constructing its e-commerce search tools.
In hardware manufacturing, AMD applied agentic AI across its chip design workflow by chaining specialized sub-agents together, successfully saving months on an effort that typically spans 18 months to two years. However, industry observers note that agentic transformation demands fundamental operational shifts. Many corporations fail by attempting to maintain their exact legacy operating models while simply inserting artificial intelligence into old processes.
Risks, Limitations, and Governance
Pinpointing whether specific business results stem directly from AI deployments or from concurrent company initiatives remains difficult. Furthermore, human oversight models present ongoing operational challenges. As code generation and automated processes scale, human review can degrade if reviewers act merely as stamping machines. Executives note that human oversight of code generation or processes may become ineffective over time under heavy workloads. Even before the advent of artificial intelligence, accountable professionals were required to test work and stand behind it, a fundamental standard that corporate leaders emphasize has not changed.
What to Watch Next
As organizations navigate these operational hurdles, the focus will shift toward refined governance and granular verification. Enterprises will continue to evaluate whether human-in-the-loop or on-the-loop models provide the most effective risk mitigation. With firms like ABN Amro, Santander, and Axa tying multi-million and multi-billion-euro projections to their AI roadmaps, the ability to isolate and prove financial earnings will dictate which projects survive budget scrutiny in the coming years.