Unlocking Real Value: Moving Beyond AI Experimentation in the Enterprise
Generative AI (gen AI) has undeniably permeated boardrooms and businesses globally. According to a McKinsey Global Survey on AI, more than 78% of companies are now using gen AI in at least one business function. Yet, despite this widespread adoption and significant investment, a striking "gen AI paradox" exists: roughly the same percentage of companies report no significant bottom-line impact or material contribution to earnings from their gen AI initiatives. As noted in a report on AI in the Enterprise, companies are still navigating this "new paradigm" and learning how to leverage AI most effectively. Why is this happening, and what can enterprises do to truly unlock value?
Current AI Use Cases: A Mixed Bag of Progress
Today, companies are leveraging gen AI in various ways, often seeing improvements in specific areas:
Workforce Performance & Automation: AI is being used to help people deliver higher-quality outputs in shorter timeframes and to free them from repetitive tasks, allowing them to focus on higher-value work. For example, Klarna's new AI assistant now handles two-thirds of customer service chats, reducing average resolution times from 11 minutes to just 2, with a projected $40 million in profit improvement. OpenAI itself uses an internal automation platform built on top of existing workflows to automate rote work for its support teams, handling hundreds of thousands of tasks monthly and freeing people for high-impact work. Morgan Stanley's financial advisors use AI daily, increasing access to documents from 20% to 80% and dramatically reducing search time, allowing more focus on client relationships.
Powering Products & Customer Experiences: Companies are embedding AI into their products to create more relevant and responsive customer experiences. A leading job site uses GPT-4o mini to provide personalized job recommendations and explain why a job is a good fit, leading to a 20% increase in job applications started and a 13% uplift in downstream success. Lowe's improved product tagging accuracy by 20% and error detection by 60% by fine-tuning models for their e-commerce search, addressing incomplete or inconsistent product data from thousands of suppliers.
Empowering Experts & Developers: AI is being put into the hands of domain experts to solve unique problems. BBVA, a global banking leader, rolled out ChatGPT Enterprise globally, enabling employees to create over 2,900 custom GPTs in five months, reducing project timelines from weeks to hours in areas like credit risk, legal, and customer service. Mercado Libre, Latin America’s largest e-commerce and fintech company, built an AI-powered platform called Verdi, powered by GPT-4o and GPT-4o mini, to help its 17,000 developers accelerate AI application builds. This platform has improved inventory capacity, fraud detection, personalized notifications, and more.
While these advancements are significant, most current deployments, especially of "horizontal" (enterprise-wide) tools like copilots and chatbots, deliver diffuse, hard-to-measure gains, even as they scale quickly. More transformative "vertical" (function-specific) use cases, despite their higher potential, frequently get stuck in pilot mode (about 90% of them) due to technical, organizational, data, and cultural barriers. The prevailing issue is that AI is often "bolted on" or used in a "shallow way"—as an assistant that sits alongside existing workflows, rather than being deeply integrated.
The Path to Real Value: Embracing Agentic AI and Reinvention
The limited bottom-line impact largely stems from this "gen AI paradox" where broad adoption doesn't equate to material economic results. To move beyond this and achieve scalable impact, companies need to make a fundamental shift:
Embrace Agentic AI: AI agents represent a major evolution, moving gen AI from reactive content generation to autonomous, goal-driven execution. Unlike traditional LLMs that are passive and require prompting, agents combine LLMs with components for memory, planning, orchestration, and integration, allowing them to understand complex goals, break them into subtasks, interact with both humans and systems, execute actions, and adapt in real time with minimal human intervention. This enables the automation of complex business workflows that were previously beyond first-generation gen AI tools. For example, in a customer call center, agents can proactively detect issues, diagnose them, initiate resolutions automatically, and even communicate directly with customers, potentially resolving up to 80% of common incidents autonomously and reducing resolution time by 60% to 90%.
Reimagine Workflows from the Ground Up: The key is not simply plugging agents into existing processes but reinventing entire workflows with agents at their core. This involves reordering steps, reallocating responsibilities between humans and agents, and designing processes to fully leverage AI's strengths like parallel execution, real-time adaptability, and elastic capacity. For instance, a retail bank reimagined its credit-risk memo process by using AI agents to extract data, draft memo sections, and generate confidence scores, shifting the human analyst's role from manual drafting to strategic oversight and exception handling, leading to a potential 20-60% increase in productivity.
Shift to Strategic Transformation: Organizations must move from scattered, bottom-up initiatives to strategic programs directly aligned with critical business priorities. This means defining transformation around end-to-end business processes, not just isolated use cases, asking "What would this function look like if agents ran 60% of it?" Delivery models need to shift from siloed AI teams to cross-functional transformation squads, integrating business experts, AI engineers, and IT architects. The focus should move from experimentation to industrialized, scalable delivery, anticipating technical prerequisites and managing recurring costs.
Build a Robust Foundation: To scale agents effectively, companies must equip their workforce for human-agent collaboration, fostering a "human + agent" mindset and introducing new roles like agent orchestrators and human-in-the-loop designers. Strong governance frameworks are essential to establish agent autonomy levels, manage risks like "sprawl containment" (uncontrolled proliferation of agents), and ensure compliance. Furthermore, the underlying technology architecture needs to evolve from LLM-centric setups to an "agentic AI mesh" that allows multiple agents to reason, collaborate, and act autonomously across various systems, tools, and language models securely and at scale. Data accessibility and quality, especially for unstructured data, are also critical enablers.
Ultimately, unlocking the full potential of AI requires a strategic pivot led by the CEO. This means concluding the experimentation phase, realigning AI priorities, redesigning governance and operating models, and launching high-impact transformation projects while simultaneously building the necessary technological foundations. The time for exploration is ending; the time for transformative action is now.
