How CGs Like Unilever, P&G, Coca-Cola, Colgate-Palmolive Establish Strong Foundations for Scalable AI
As artificial intelligence capabilities evolve, so does the remit for consumer goods tech executives tasked with scaling efforts — moving beyond small trial phases and toward enterprise-wide adoption.
Moving past AI exploratory use cases to full-scale adoption relies on strategic alignment, operational readiness and baseline measurement. Executives must focus on wider business outcomes, establishing clear ROI metrics and governance standards while deploying technology at scale instead of within isolated lab experiments.
Yet, more than half of CPG companies currently do not track the ROI of their AI initiatives, according to Peri Edelstein, North America lead for Boston Consulting Group’s consumer work — a critical gap given that boardrooms are demanding strict accountability before greenlighting global rollouts.
[RELATED: Coca-Cola, Newell, Clorox: Three Distinct Roadmaps for the CPG AI Journey]
An example of enterprise scale comes from Unilever, which integrated predictive AI capabilities into customer operations through an end-to-end operating system. It is now testing an agent called Nova that analyzes 23 million orders and emails — determining order ownership, claim validity and fault — before ingesting data into Unilever’s SAP backbone.
"We start to understand how our product is going to sell into the future. We built a regression model which we've piloted and is now running across about [$52.4 billion] of our turnover," says Graham Sommer head of customer operations for Unilever. "It's really about how we get from pilots in certain markets to true scale, driving those business outcomes achieved in our pilots."
Procter & Gamble has been focused on AI scalability as well, using a platform-based approach, which CIO Seth Cohen calls "the AI factory," to create customized business capabilities from common workbenches. This functions by keeping agility at the forefront while connecting to data more quickly, ensuring scalability.
"AI can only create value when it is seamlessly integrated into the business, and not when it sits on the sidelines via a collection of pilots," Cohen previously said in a company blog.
Steve Daugherty, VP analyst with Gartner's supply chain practice, says the first step to transitioning from a pilot phase is linking AI initiatives back to operational objectives.
Meanwhile, Edelstein emphasizes that proving a pilot's value requires validating its economics and operating model in real-world conditions rather than isolated sandboxes. To achieve this, organizations must align initiatives with core commercial priorities, establish clear ROI metrics upfront and test tools within live, full-volume workflows.
"What we really find is that the tools are not hard to create, but changing the ways of working, the decision rights, the roles is really tough. So having a plan for that before you scale is important," says Edelstein.
Establishing Strong Data Cores & Multilayered Governance
When it comes to AI, the value is not in exposing more data, according to Balaji Balasubramanian, president and chief product officer of customer experience and consumer industries at SAP. While conversations often frame data access as the primary challenge, he notes that the real hurdle is access to trusted context so employees can make better decisions and execute effectively.
He warns that AI often acts as an execution layer that accelerates existing organizational errors if data isn't aligned: "Speed is only an advantage when underlying data, policies and workflows are aligned. Otherwise, organizations simply automate errors faster."
Also: Mars, PepsiCo Partner With Google Cloud on Enterprise-Wide AI Transformations
This is something The Coca-Cola Co. has addressed head on as it faces unique data consolidation challenges due to the complexity of its network and legacy workflows it is modernizing. The company said it focuses on precision at scale.
"Scaling generic content is not going to give you the greatest bang for your buck," president and CFO John Murphy said earlier this year. "The point is to be able to bring core content tailored appropriately but with efficiency to be more precise with our communication at the point of sale, whether it's in Haiti or in Germany."
As organizations implement these data practices, they must also ensure they are minimizing risk and monitoring how AI systems access and share information.
At Unilever, governance approaches differ based on the specific use case, according to Sommer:
- Utility & Personal Agents: For smaller utility agents that employees experiment with, a digital operations program built on a gamification platform educates associates at their own pace to ensure they understand cybersecurity and AI governance processes.
- Enterprise Global Agents: For major projects like Nova and eight other global agents, governance is embedded directly into the rollout plan alongside technology partners like Microsoft, Capgemini and Genpact.
"If it's something that is a bigger project, which is really scalable at a global scale, then we embed that into the global product design," Sommer explains.
Daugherty says high AI maturity firms are twice as likely to have cross-functional governance and also twice as likely to have continuous monitoring of their AI tools. And Claudia Clemens, senior director analyst for Gartner's supply chain practice says governance and controls are driving a lot of discussion right now.
"Global firms need to be thoughtful about AI sovereignty and concerns about model access from market to market."
Regional applications introduce operational variances. Edelstein recommends a federated product model over complete decentralization. Under this model, embedded regional or business pods handle local workflow redesigns and market-specific needs, while the central enterprise team retains ownership of the AI strategy, core architecture, orchestration platform, responsible AI guidelines, cybersecurity, vendor standards and reusable data assets.
Cohen has said that "data discipline isn't glamorous, but it's the difference between AI theatre and AI as a competitive advantage."
Strengthening Workforce Skills & Trust
Edelstein strongly advocates for functional AI literacy across organizations, noting that deep technical training alone yields limited ROI compared to practical application. Therefore, scalability requires upskilling staff to interact effectively with automated systems.
Organizations should prepare their employees to work alongside intelligent systems, P&G's Cohen has emphasized, freeing them up for higher-value work. But that requires first building trust.
At the start of Unilever's AI-enabled demand planning journey, Sommer recalls that employees expressed hesitation: "People were insecure and unsure." They'd say things like, "A machine-generated forecast is not my forecast. Do I really trust those numbers? What's under the hood?"
Over four years, Unilever focused on building comprehensive digital literacy, helping teams understand Python, Databricks, Azure data extraction, regression models in Copilot Studio and data storytelling. "People started to be much more comfortable working in this AI environment," says Sommer.
At Colgate-Palmolive, the organization is preparing for agentic scalability, asking employees to use large language models across their work to drive enhanced productivity.
As part of this, it has launched education programs to obtain enterprise-wide buy-in. The company has already trained and upskilled a significant part of its workforce. CEO and president Noel Wallace said they are about 70% done with advanced artificial intelligence training at the vice president level.
Across the CPG landscape, the takeaway is clear: enterprise scale isn't achieved by deploying tools faster in a vacuum, but by building the data foundations, operational governance and workforce literacy required to run them safely at high speeds.
