KT and Amorepacific Cut Cosmetics R&D Review From 12 Days to 5 Minutes With AI

South Korean telecom operator KT and cosmetics conglomerate Amorepacific announced on Tuesday the launch of a jointly developed AI assistant called Lemon, built to search and analyze seven decades of the company's research and development data. In beta tests run with researchers from Amorepacific's R&I center, the time needed to review research materials and regulatory requirements during new product planning dropped from about 12 days to five minutes.

What Lemon Actually Does

Lemon lets Amorepacific's researchers ask questions in natural language and get integrated answers drawn from databases of ingredients, formulas, experiment reports, and regulatory documentation. Instead of manually digging through archives and binders built up over decades, a researcher simply describes the query in a plain sentence, and the system pulls together data that had previously been scattered across the company's many systems.

The project rests on earlier work called "Data Highway," which organized and converted raw, often unformatted research data into a form AI models can read. Without that step the assistant would have had nothing to draw on, since Amorepacific's R&D data has been piling up for decades across a patchwork of formats and systems.

The Scale of the Time Savings

The drop from 12 days to five minutes applies to one specific stage of the process: reviewing scientific literature and regulatory requirements when planning a new cosmetic product. That task recurs with every new formula project at a company this size, so the cumulative time savings for researchers could add up, even though the finished product still needs lab testing and approvals that AI does not replace.

KT stresses that structuring the data lets the AI system understand and analyze research information without extra on-the-fly processing, which translates into faster and more consistent answers than searching through unorganized archives.

Why Corporate AI Starts With Data, Not the Model

The Lemon project fits a broader pattern in which large industrial companies treat preparing their own data as the first and most expensive step of an AI rollout, while the language model itself becomes a swappable component. KT says it will apply what it learned from the Amorepacific project to similar deployments in other industries, including manufacturing, retail, and services.

We will use our experience from this project to expand data-ready AI transformation initiatives across various industries - Noh Hyung-rae, head of Enterprise Business at KT

What Comes Next

Both companies say they plan to keep expanding their AI-ready data environment and to lay the groundwork for more autonomous AI agents within Amorepacific's R&D divisions. The Lemon platform is meant to serve as a foundation for further tools, not as a finished, closed product.

For readers in Poland, the Amorepacific and KT case stands out as concrete and measurable, unlike many vague reports of corporate "AI transformation." Here the companies stated plainly how many workdays were saved and at which stage of the process. It's the kind of deployment that Polish manufacturing and research firms, in pharmaceuticals, chemicals, or food, could also consider, provided they have a comparably large and well-organized R&D data archive.

The story also shows where the real cost of such projects lies: not in the language model itself, but in years of work standardizing formats and metadata. Without the Data Highway stage, no AI assistant would have had anything to work with, no matter how advanced the model behind the interface.