How Procter & Gamble uses AI to derive new insights from data

Few organizations can legitimately claim We have been conducting analytical research for over a century. But in 1924, William Cooper Procter, CEO of Procter & Gamble, asked Paul “Doc” Smelser, an economist at P&G for 34 years, to study how many customers were using Ivory soap to wash themselves rather than wash dishes or clothes. 12% answered for washing dishes, 31% for washing face and hands, 40% for bathing, and 17% for other uses (perhaps washing clothes). Although ivory soap has become more of a personal care product than a household item, Internet commentators suggest it is still effective for washing clothes.

P&G was also a pioneer in creating common data across its global organization, establishing a large group of analytical experts and incorporating them into business units. Continuing our tradition of leveraging customer and market insights to this day, we now use analytical, generative, and agentic AI to address critical business problems.

To find out what P&G is doing with artificial intelligence, we spoke to Jeff Goldman, the company's vice president of enterprise data science and leader of its business AI initiatives. We first met Goldman more than a decade ago when he was developing an innovative approach to analytical visualization at P&G through an initiative called Business Sphere. Since then, he has assembled a group of hundreds of data scientists and AI engineers to build and deploy AI algorithms at scale across the company's marketing, digital commerce, supply chain, and sales organizations. Most data scientists are embedded directly into business units or AI product teams.

As Goldman noted, a 100-year-old orientation toward data and analytics remains prevalent. “The historical analytical spirit continues in our current culture,” he said. “We have always had a desire to perform analytics and understand the dynamics behind our business. However, the introduction of AI has dramatically expanded the nature of the questions we can ask and the depth of answers we can provide, as well as our ability to leverage AI to optimize our business processes.”

P&G's “AI Factory”

Around 2021, Goldman and his colleagues realized that AI was playing an increasingly strategic role in P&G's business. The complexity of algorithms in production was increasing, and delays in moving from prototype to scale with AI algorithms were having significant financial implications. The algorithms were developed in a bespoke manner, with each algorithm requiring custom deployment by the company's AI engineers.

To speed up the creation of AI models, P&G has developed a feature it calls the “AI Factory.” It provides the means to rapidly develop, test, deploy, and monitor algorithms in production, including data sources, software tools, methods, and defined security protocols. P&G CIO Seth Cohen explained the project in a podcast episode earlier this year: “The great thing about the AI ​​factory approach is that it's a platform where people not only have instant access to the data in the data repository, but also instant access to the AI ​​algorithms and generative models. So developers spend a lot less time worrying about, 'How do I scale this?' — because it's something that works right out of the box.”

Goldman says AI Factory will reduce model deployment time by about six months. He also pointed out that as technology evolves, so do factories. For example, it includes agent AI capabilities such as monitoring large agent systems, registering agents, and applying the Agent2Agent and model context protocols needed to connect multiple agents and tools. Goldman said there are essentially two sides to the company's AI engineering organization. One is responsible for building and continuously updating the factory, while the other focuses on expanding and operating the algorithms developed within the factory.

The AI ​​Factory feature makes it easy to test different versions of models to meet specific business requirements. For example, one of the use cases developed at the factory is the Pampers My Perfect Fit application. The application uses the results of an AI-driven survey to provide parents with diaper fit recommendations with 90% accuracy to prevent leakage, the most common problem on the market. In another example, an analytical AI use case is being realized in Brazil. The system divides and schedules customer orders into truck-sized loads and prioritizes them according to shelf needs. This system has reduced the occurrence of stockouts in the country by 15%.

P&G's generative and agentic AI products

Goldman pointed to use cases around supply chain management and media decision-making, noting that a large share of the value has and will continue to come from analytical AI. But P&G was also an early adopter of generative AI, with a range of internal products that incorporate the technology. The company developed imagePG and askPG after providing its employees (through a product it named chatPG) with secure access to a variety of underlying language models that they could choose from based on the business problem they were addressing. ImagePG supports the generation and analysis of images and videos, including for corporate advertising. askPG, on the other hand, incorporates internal unstructured data that is curated for use by employees.

While P&G's generative AI products are focused on personal productivity, the company is equally interested in use cases that can be extended directly to business processes. For example, the company's Great Idea Generator tool creates product and advertising concepts based on consumer trends and previous test results. From concept to advertising to shelf, you can significantly accelerate the progression of new ideas. Another AI tool, Project Genie, integrates information from articles and help documentation to inform more than 800 customer service representatives, significantly reducing question processing time.

P&G has long focused on providing easy access to data about the performance of various aspects of its business. Business Sphere, an analytics visualization environment, and a desktop version called Decision Cockpit were some of the early approaches to this problem, and now the company offers a business data generation front end called insightsPG. As CIO Cohen said in an article earlier this year, “Why do you need a dashboard when you can manipulate data?'' Goldman said that although insightsPG is relatively new, it is transforming the way businesses work with data by bringing advanced analytics and inference capabilities to a broader range of people within the company.

P&G's experiments with agent AI have focused on a series of pilots to determine where the technology can be used most effectively. Goldman commented that the agency is already seeing early success in advertising, supply chain and consumer relations areas. He said it remains important to keep humans in the loop on these and other processes.

Improving R&D with AI

Goldman's data science and AI organization has a strong partnership with P&G's research and development organization. The company's research and development has a long history of using data and quantitative analysis to understand the chemistry and physics of products and how best to manufacture them. Recently, our research and development team has used AI algorithms to complement the work of laboratory technicians and enable faster molecular discovery. One example is P&G's Perfume Development Digital Suite. It's an ecosystem of digital tools that leverages AI and advanced data processing to create new fragrances five times faster than before. AI algorithms analyze millions of data points and help create a perfume character model based on consumer insights into what makes a fragrance smell good. The model integrates perfume character models to identify formulations with a high probability of success, which are then tested through rapid prototyping and experimentation.

Research and development organizations are collaborating with IT to test new ways of working using generative AI. P&G partnered with Harvard Business School's Institute for Digital Data Design to conduct a large-scale field experiment on how AI can function as a “cybernetic teammate.” It involved 776 commercial and R&D professionals who were assigned to use generative AI, either individually or in teams, to address real-world consumer problems across four business units, with or without generative AI. The results of the experiment showed that individuals using chatPG achieved the same performance levels as teams not using the AI ​​tool, but teams that partnered with chatPG consistently produced the best results. The use of AI has also helped break down functional silos between R&D and commercial professionals. Experts from both backgrounds using AI were able to produce more balanced solutions.

Build human and AI capabilities

P&G has long focused on building human capabilities in data, analytics and AI. The company is partnering with Harvard Business School and Boston Consulting Group to improve AI skills across the P&G workforce. To date, more than 4,000 executives have taken an intensive eight-week AI upskilling program focused on AI's strategic impact on the industry. This is complemented by an internal upskilling program that focuses on the day-to-day use of generated AI in employees' work. Together, these programs are rapidly piloting new AI capabilities and creating a growing group of leaders who collaborate with Goldman teams to develop the firm's next-generation AI algorithms.

Additionally, several years ago P&G began offering the ongoing Friends of Data Science certification program. The company has had a large quantitative analyst community for many years, and the purpose of creating the certification program was to improve the skills of these individuals in analytical AI. The program requires 15 weeks of intensive study and focuses on not only how to build models, but more importantly, how they can go wrong. This content is continually updated for new cohorts and currently includes content on Transformer models behind generative AI and graph machine learning for uncovering increasingly complex data signals.

“Organizations' reluctance to digitize makes it difficult to introduce new capabilities,” says Cohen. “We’re changing that.” Given the company’s long-term focus on data and analytics, and now on various forms of AI, P&G appears to be one of the most “digitally reluctant” companies we’ve come across.