Thor Olavsrud
Senior Writer

Unilever leverages GPT API to deliver business value

Feature
Mar 10, 20236 mins
Artificial IntelligenceDigital Transformation

The CPG titan has created AI tools using neural networks to help it respond to messages from customers, generate product listings, and even minimize food waste.

Alessandro Ventura stylized
Credit: Unilever

The past several years have thrown numerous challenges at consumer packaged goods (CPG) companies. The pandemic has led to shifting consumer channel preferences, a supply chain crunch, and cost pressure, to name just a few. CPG titan Unilever has been answering the challenge with analytics and artificial intelligence (AI).

The 93-year-old, London-based CPG company is the world’s largest soap producer. Its products include food and condiments, toothpaste, beauty products and much more, including brands like Dove, Hellmann’s, and Ben & Jerry’s ice cream.

Alessandro Ventura, CIO and vice president of analytics and business services for North America at Unilever, has been at the forefront of helping the company apply AI to its businesses for years. While originally in the role of IT director, he has since added analytics and people services to his portfolio.

“That’s everything from facility management, fleet management, employee and facilities services, and people data, and that kind of stuff,” Ventura explains.

Unilever believes AI is not a technology of tomorrow. It’s already being widely used, and Ventura feels all industries will need to adapt to it.

In recent months, Unilever has developed a number of new technology applications to help its lines of business in the markets of tomorrow. One of the most important is “Alex,” short for Alexander the Great. Alex, powered by GPT API, filters emails in Unilever’s Consumer Engagement Center, sorting spam from real consumer messages. For the legitimate messages, it then recommends responses to Unilever’s human agents.

“Although Alex is good at what it does, it may lack a bit of a personal touch that instead our consumer engagement center agents have in big quantities,” Ventura says. “So, we let them decide whether they want to respond to our consumer as Alex suggested, or they want to add some personal recommendation; if the answer suggested by Alex is wrong or doesn’t have an answer, they can flag it so Alex can learn it the following time.” 

Generative AI in action

Alex was created using a system of neural networks, with GPT API for content generation. Ventura says the tool can understand what a consumer is asking and even capture the tone. It can then store the answer and sentiment in Salesforce. Importantly, he says, the tool does the heavy lifting on those tasks, giving the human agents more time to dedicate to what they do best. To date, Ventura says Alex has helped Unilever reduce the amount of time agents spend drafting an answer by more than 90%.

Another Unilever tool, called Homer, leverages GPT API to generate content. It’s a neural network that takes a few details about a product and generates an Amazon product listing, with a short description and long description that matches the brand tone.

“We want to ensure we captured the voice of the brand so, for example, that we differentiate between a TRESemmé and a Dove shampoo, and the system got it absolutely nailed,” Ventura says. 

Another AI-based tool that Unilever launched on the week of US Thanksgiving supports the Hellmann’s mayonnaise brand. Its purpose is to reduce food waste.

“It links up with the recipe management system that we have at Hellmann’s, so somebody can go in and select two or three ingredients that they have in the fridge and get in exchange recipes for what they can do with those ingredients,” Ventura says.

In the first week, the tool got 80,000 users who reported loving it.

For Ventura, that’s the magic of analytics and AI in the CPG space: It enables personalization at scale.

“In CPG, we rely more and more on analytics and AI for different things,” he says. “Consumers are more and more specific about what they want. It’s a bit of a cliché, but they really do want personalized products and experiences. Analytics helps CPG to understand the context they’re navigating through and what the consumer wants, and then, with AI, we can scale that one-to-one relationship across all the multitude of consumers that we have.”

Co-creation key to AI success

Beyond the consumer relationship, analytics and AI are also key to making CPG companies more sustainable. Ventura points to examples like ingredient traceability and using machine learning (ML) to automate forecasting, which in turn helps the company minimize waste. Unilever is also applying analytics and AI to logistics, including tracking inventory and optimizing routes.

“The old interpretation of elasticity, we threw it out the window,” Ventura says of operations in the wake of the inflation crisis. “We had to come up with new calculations because the traditional ones were giving us very different scenarios from what we were seeing happening at the shelves. Going forward, we will continue to see that pressure from all the different challenges coming from the geopolitical situation around the world.” 

To support its innovation around analytics and AI, Unilever has adopted a hybrid model. It has a global center of excellence, but also keeps some data scientists embedded with business units.

“It’s basically a two-gear system,” Ventura says. “The local team can be activated very quickly, ingest the data very quickly, and then create a statistical model and analytics model together with the business, sitting next to each other. Then, if that model can be leveraged across and scaled, we pass it on to the global team so they can move data sets in the global data lake that we have and can start creating and maintaining that model at a global level.” 

Ventura believes co-creation and co-ownership of analytics and AI capabilities with the business function is essential to success.

“Whether it is machine learning for automating the forecast or Alex with the Consumer Engagement Center, if we show up with a black box and say, ‘Hey, follow whatever the machine tells you,’ it will take a long time and probably will never get to 100% trust in the machine,” Ventura says. “With co-creation and co-ownership, I feel like we get to start with the right foot, with the human and the machine working alongside each other in partnership, almost as colleagues. Also, you get a much less biased system in the end because you’re able to introduce a much more diverse angle in your algorithms, both from a business perspective and a technology perspective.”