Dreen Yang is Capgemini Group’s Global Consumer Products and Retail, focused on AI transformation and driving enterprise value.gettyWhen Coca-Cola relaunched its iconic "Share a Coke" campaign in 2025, it was lauded as an AI success story.Building on the groundbreaking 2011 personalization campaign, the relaunch used AI to refine and localize names, pick up culturally relevant slang like "bro" and "bestie" and extend the experience into digital platforms through consumer-generated content.The result was strong. The company reported a 2% sales increase in the U.S. alone and a near-perfect social sentiment score across Asia and the South Pacific.AI played a real role, but it was not the whole story. The creative bet, the strategic decision to revive a 14-year-old campaign and the campaign adaptation to resonate with Gen Z were human calls. AI simply accelerated the supporting work around them.That balance is the point. AI can be a source of value, but layering it into existing workflows often drives complexity without bringing a competitive advantage. To generate real impact from AI, leaders must first take a counterintuitive step: unbundling work to understand where AI can add the most value.The Deeper Shift Hiding Underneath The AI DebateTwo thinkers, decades apart, asked the same question in different ways.In 1991, Ronald Coase won the Nobel Memorial Prize in Economic Sciences for his work on transaction costs and property rights, including his foundational question of why companies exist at all. His insight was, essentially, that a company draws its walls around the work it can do better itself than it can buy from others.​Gino Wickman echoes a similar principle in Traction, arguing that strong companies identify and consistently execute a small set of core processes that drive their business. This discipline helps organizations reduce complexity and focus their resources on the activities that matter most.​​Two thinkers, decades apart, made the same point in different ways: Successful companies protect the work that makes them win and delegate everything else to whoever can do it best.AI now changes who can do “everything else” best. For decades, utility work such as regulatory documentation, demand forecasting and media buying stayed inside the company because no outside partner could match the scale or institutional knowledge required. AI has changed that. In many cases, an external partner can do utility work at a higher scale and with a deeper level of insight than an internal team.A Simple Way To Draw The WallsEvery activity inside a company can be sorted along two dimensions: how much it contributes to competitive advantage, and the maturity of the AI capability needed to perform it. The 2x2 Framework that I developed can be helpful here:• Zone 1: Human-led, high advantage, limited AI. Activities include innovation direction, brand strategy and major platform bets.• Zone 2: AI-augmented, high advantage, mature AI. Functions here include insight synthesis, claims testing and account management.• Zone 3: AI-executed, low advantage, mature AI. Activities include forecasting, optimization, documentation and media buying.• Zone 4: Simplify or eliminate, low advantage, limited AI. Legacy processes and redundant approvals should be streamlined or eliminated because they add little strategic value.The strategic question facing every CEO is not "where can we use AI?" It is "which zone does each activity belong in?" Most AI failures are not technology failures. They are zone-assignment failures. Using R&D as an example: Strategy sits in Zone 1; in Zone 2, AI mining a million social posts becomes a force multiplier for in-house teams; and in Zone 3, AI partners handle formula optimization and compliance, fast and cheap.What stays inside remains the same: the strategic bets, the customer relationships, the creative intuition that defines the brand. The great unbundling is not about doing less. It is about redrawing the walls in the right place.How One Company Got It RightAt first glance, Unilever’s AI-enabled freezers seem like a Zone 3 activity, since inventory monitoring is utility work. But the company, intentionally or not, treated the work differently. With 3 million freezers placed in shops, airports and supermarkets worldwide, the data flowing back gave Unilever a real-time picture of impulse-purchase behavior that is not easily to replicate.The company built and now operates the capability in-house, treating it as Zone 2. As a result, retail orders and sales have increased by up to 30% in some markets.In short, zone assignment is not determined by whether an activity feels operational or strategic in the abstract, but rather by whether the AI capability creates competitive advantage.​Where Zone 3 Dollars Should GoZone 3 is broader than most CEOs realize. It is not just the supply chain, but rather the entire shared-services backbone: demand forecasting, trade promotion, marketing analytics, procurement, finance close, HR transactions and regulatory submissions.Reimagining this GBS layer with agentic AI is where the largest pool of measurable return sits in the typical CPR business. In fact, data from the Capgemini Research Institute reveals that AI is delivering cost savings of 26% to 31% across these core business functions.The catch is that Zone 3 cannot be handed to generic AI. A demand-forecasting agent that does not know the difference between a promotional lift and a true demand signal optimizes for the wrong thing. A trade-promotion agent that does not understand bottler economics or shopper-marketing dynamics recommends tactics that destroy ROI.Domain expertise is what separates Zone 3 work that delivers from work that creates expensive new problems.Making The Case For Focused AITo succeed, CEOs must focus on the handful of processes their business actually runs on. Once those are named, the AI strategy becomes obvious. Double down on AI in Zones 1 and 2, augmenting the people whose judgment, creativity and customer relationships are the source of advantage. For Zone 3, lean on domain-first external partners to focus internal resources where they matter most (Zones 1 and 2).The companies that win will not be the ones with the most AI. They will be the ones that double down on AI to elevate their people in the few areas that truly differentiate the business—and have the discipline to let partners handle the rest.​Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?