Unilever, the global consumer goods company behind well-loved brands such as Dove, Vaseline and Domestos, processes around 20 million orders a year for customers ranging from global retailers to local distributors. At this scale, the opportunity is not simply to serve customers more efficiently, but to build a more connected, intelligent enterprise. So Unilever has launched NOVA – a system of AI agents developed with Microsoft – to unlock value faster and create a better customer experience.
Every day, 3.7 billion people use Unilever products. Ensuring those products reach the shelves means serving about 300,000 direct customers, including supermarket chains, wholesalers, distributors and e-commerce partners, who in turn supply millions of smaller shops.
The result is a vast, complex flow of orders, deliveries, invoices, payments, and claims for trade promotions.
Francis Castro, Head of Technology for Customer Operations at Unilever, says the majority of these orders are already digitised and pass directly between computer systems, but around two million still arrive in a variety of formats, from emails to PDFs, spreadsheets to photos of paper forms. These require a lot of manual, time-consuming processing by colleagues in Unilever service hubs around the world.
“They are crunching data, trying to input orders, when that time could be spent talking to customers,” says Deepalok Roy Chowdhury, Unilever’s Global Director for ‘Order-to-Cash’ – as this part of the business is called.
“At the end of the day, at the heart of our operations is excellence and service at scale – providing the best possible experience for our customers and our people.”
For Unilever, the priority is to make ordering as easy as possible for customers. A retailer working with many different businesses cannot be expected to convert documents to every required standard. As Roy Chowdhury puts it: “Customers shouldn’t have to adapt to our technology. However they choose to work with us, our job is to make it simple.”
Five agents, big impact
So, Unilever developed a network of AI agents – NOVA – built on Microsoft Foundry, which now read, check and enter those orders in a much more efficient way. They are also applying the same intelligent technology to claims, extending the impact across the value chain.
NOVA involves five agents, four reading the email, interpreting its context and extracting the order; a fifth checking the work. Missing information and exceptions go to a person for review.
For claims validation, Microsoft engineers joined Unilever’s team to build and test the process in less than eight weeks.
The impact has been immediate.
Across the 11 pilot markets, NOVA is reducing the first stage of the order-to-cash process from 15-to-20 minutes to around 90 seconds, says Castro.
Orders are now 95% touchless, with teams stepping in where exceptions require review, and eight times faster than before. Refund claims are processed twice as fast, and new customer onboarding has been reduced by 90%, he says.
These time savings are freeing up Unilever’s people to spend more time resolving unusual cases and helping customers, says Castro.
“NOVA is not just giving them more time back,” he says. “It’s supercharging the abilities they have.”
Consultation key
The project’s success can be attributed in part to the work carried out by Castro and Roy Chowdhury’s teams, who consulted customer operations analysts to identify which repetitive checking and data-entry processes were taking up the most time and causing the most friction.
NOVA was specifically designed to remove those pain points so colleagues could do less repetitive work and more analysis and customer conversations.
The feedback from colleagues has been overwhelmingly positive, says Roy Chowdhury.
“There was apprehension when we started. Usually with technology, there’s a push: ‘Can you please adopt this?’ Now it’s the other way around. Countries are putting up their hands and saying: ‘We want to go next’”, he says.
While AI carries out a significant number of steps within guardrails, people still supervise and decide where judgement is required.
The next frontier
NOVA is Unilever’s first multi-agent supply-chain system at this scale, and the early months involved finding the right architecture, says Castro.
“One of my dreams was to compress the cycle time of programmes and the cycle time of the operation, and we are showing that we can do that,” says Castro.
“You need to put yourself in a learning spirit and ask how we can do things differently with new technology,” he says.
NOVA has already processed thousands of orders in production, and Unilever plans to expand it to more than 50 markets.
Castro is proud of these results, but believes NOVA is just the first step towards a more intelligent enterprise for Unilever. The wider goal is increasingly connected operations across forecasting, planning, logistics and order-to-cash, he says.
For the teams in Unilever’s service hubs, the litmus test is simpler: less time moving information between spreadsheets and more time using their experience to help customers.