How Unilever Scales AI, From Predictive Models to Agentic Workflows
Unilever turned customer operations into its own function just four years ago, and it has already taken on a vastly different shape. The department touches all areas related to customer experience, with Graham Sommer at the helm as head of customer operations.
Much of that change stems from advancing AI capabilities that allow the company to better react to shifting customer needs through end-to-end intelligent planning. This impacts how Unilever prepares in advance for demands related to materials, factories, distribution and customer fulfillment.
Because the company processes about 23 million orders per year for more than 200,000 customers — managing roughly 70,000 products across 200 warehouses — a more dependable, agile customer operations effort capable of navigating Unilever's complex network has become essential.
As a result, Sommer’s team of 6,000 people — split evenly between Unilever employees and partner resources — works together in an AI- and agent-enabled, end-to-end operating system the company built over time.
"We've done predictive AI in our forecasting so we start to understand how our product is going to sell into the future," Sommer said. "And with that, we built a regression model — a machine learning model — which we piloted and is now running across about [45 billion euros] of our turnover."
He recently shared that the next stage of this journey is agentic AI, having already rolled out eight agents worldwide. As part of this push, Unilever is scaling an agent called Nova, which will soon go live in North America, the company's largest market.
Nova helps Unilever ingest large volumes of orders. The agent receives emails, interprets their context and inputs that data into Unilever's SAP operating system. It also helps navigate the claims process by validating orders to determine whether a customer is at fault, using agentic capabilities to close the loop on these workflows.
Behind the Scenes: Sommer Discusses the Importance of Strong Data Foundations for AI Scalability
Determining the Value of New Products
Before AI implementations and agents like Nova get the green light to scale, Unilever uses a strategic approach to validation and value monitoring to ensure a tech use case should be expanded, revised or scrapped.
"We can tell anywhere in the world if this process is following the global process design or not," Sommer said. "And with that information, we're able to then improve it because normally there's a good reason why somebody is not following a process. And when you start to dissect that, you start to understand the real drivers of why that process isn't being followed and you then have areas where you can improve."
The next phase revolves around the technology itself and how it is organized. Unilever determines which roles personas should have in the market and how partners such as Genpact and Capgemini will support and improve process design to drive product standardization globally.
"For example, in planning, our global partner is Kinaxis," Sommer said. "It's a Canadian-based or Ottawa-based planning software company, but they work with us around the world on our planning system, and that technology is then standard across the world following that global process design."
Also: How Unilever's FIFA Product Plan Came to Life With AI, Data, Digital Twin Tech
Last, Unilever measures these products to determine if they are truly "no touch" or fully automated. For example, the forecasting capabilities Sommer referenced are about 60% no touch, leaving room for improvement. However, some of its planning processes with Kinaxis are closer to 90%.
"That means we are really following that global process design and it's working with us and delivering the business outcomes without people needing to go in and change the outputs," he said.
Enabling Growth With Change Management, Knowledge Sharing and Governance
As companies adopt AI, roles shift. Sommer noted that planners who previously worked within spreadsheets and specific portfolios now collaborate directly with automated AI solutions.
Their roles are evolving into end-to-end flow orchestrators, allowing them to look with broader perspective across the value chain while automation handles individual calculations.
Sommer emphasized the importance of ensuring employees understand and feel comfortable with these changes. In the early stages of AI integration, employees were often hesitant to rely on machine-generated forecasts over manual calculations.
To bridge this gap, Unilever established a DigiOps program to build digital fluency across the workforce. The program provides employees with nine months of on-the-job training totaling roughly 150 hours of education. It is work-level agnostic, making it accessible to anyone in the organization.
"There is an absolute direct correlation between the number of DigiOps learners and the [increased] adoption rates of those markets that they work in," Sommer said.
Governance is also integrated into upskilling, though flexibility varies between enterprise projects like Nova and smaller utility agents designed for personal productivity. Sommer said the organization is training nearly 2,000 people to build their own AI agents within a secure cybersecurity environment to safeguard data.
"We want to have an environment where people are experimenting, sandboxing and trying because the technology is advancing so fast," he said. "But this is not something which we believe we can control centrally and say every new release needs to be governed and owned by the product team."
Unilever aims to strike a balance between necessary governance and operational flexibility, recognizing that evolving technology will require strategic pivots along the way.
"If it's something which is a bigger project, which is really scalable ... we embed that into the global product design," Sommer said. "And then it goes into the governance — a full rollout plan around the world where we work with our partners to help and build this together with us."
