How Procter & Gamble accelerates AI innovation by prioritizing data transparency and security

Problem

The P&G R&D team collects large amounts of unstructured data throughout their prototyping, testing, and manufacturing process. This data can be used to train various AI models when refined, tagged, and classified. However, because of its proprietary nature, the team needed to ensure stringent privacy standards and full transparency throughout their labeling process.

Solution

P&G’s R&D team chose Labelbox after a rigorous analysis of all the data annotation platforms in the marketplace. P&G chose Labelbox for its ease of use, compatibility with their existing data infrastructure, and data security and transparency standards. With Labelbox, the team was able to bring in both internal and external labelers as required for their projects, recalibrate their labeling processes quickly as needs evolve, and accelerate the pace of their model iterations and time to market.

Result

The R&D team first used Labelbox for their computer vision projects for defect detection on their product lines. Over the course of three years, they’ve expanded their use of the platform for more AI projects. Today, Labelbox is an enterprise-wide solution at Procter & Gamble, used to produce training data for AI across all their business units around the globe.

Note: This story is a recap based on the P&G panel discussion, “Partnering on AI Capability Today to Accelerate Innovation Tomorrow,” at CES 2023.


Procter & Gamble is a Fortune 500 consumer goods giant and makes many well-known household and personal care products under a number of popular brands, including Tide, Charmin, Pampers, Braun, Old Spice, and many more. The process for testing, developing, and manufacturing their products involves a large amount of research and analysis — from studying public behaviors and sentiments around certain brands, ingredients, and even scents to identifying defects in the production line — which in turn involves collecting large amounts of unstructured data in the form of images, videos, and text. 


“There's a lot of work that goes into every little business decision that we make. And for a large company that works at scale to serve billions of consumers, the decisions that we make in terms of what we're going to manufacture, put into the market, and support, are very critical. Getting high-quality trustworthy data that we own and that we are fully confident in is critical for us,” said Kelly Anderson, Director of R&D Data Science and AI at Procter & Gamble during a panel at CES 2023.


The team needed a central platform with which to label and improve their AI data. “We did a side-by-side comparison of all the different data annotation platforms in the marketplace, and we really focused on things such as the user experience, compatibility with our existing solutions and infrastructure, as well as the cost model. Labelbox really stood out to us because [they] had a very compelling technical solution, we were strategically aligned on where we wanted to go with the data, and [they] also had a very clear and sustainable price structure,” said Mercy Chang, Senior Purchasing Manager at Procter & Gamble.


While refining unstructured data into high-quality structured data is a challenging enough task on its own, this team also had to prioritize privacy and security in their search for a data labeling platform. Data entrusted to the company by consumers needs to remain private. Data generated during the manufacturing and prototyping process is often considered IP. “Data is the lifeblood and one of the foundation blocks of how we're able to deliver superior products to consumers. So it's incredibly important that we safeguard it. When it comes to data privacy and information security, it's non-negotiable, because our consumers, customers, and partners put so much trust in us,” said Chang.


“One of the things that we loved about Labelbox from the very beginning is their belief that the data is ours,” said Anderson. This emphasis on data security and full transparency throughout the labeling process ended up offering them insights into how their data was being enriched that they would not have had with a typical black box labeling service or product. “Having a transparent platform that allows us to keep our data in our cloud instance, link that data to the Labelbox user interface, have an external high-quality labeling workforce enrich that data with tags….we could actually see how the data is maturing,” said Anderson.


With Labelbox, the team was also able to bring whichever labelers they needed — whether internal or external — onto a project within the platform, and maintain full transparency and ownership of their data no matter who was labeling it, unlike previously used labeling partners that would take data and give it back labeled, with no insights into their actual process. “Having a transparent agile platform that’s fluid to work with….has been a transformation in how we do our work, and has been very powerful in increasing the speed of iteration, which leads to faster time to market with better insights,” said Anderson. 


After three years of iterating on and expanding their use of Labelbox on bigger and more significant projects in the R&D department, Procter & Gamble is now leveraging the platform as an enterprise solution. “We're packaging the Labelbox solution and redistributing it across all our business units around the globe. So this is no longer just an R&D solution, it's a corporate solution,” said Chang.