Automating regulatory diligence for cosmetic brands: 75% saved

Twistag built a regulatory and supplier-diligence platform for a European RegTech startup, cutting the time spent on a customer regulatory inquiry by 75% and reaching production at three of the top ten European cosmetic brands. The client's customers were assembling compliance evidence — REACH, CLP, food-contact rules, allergen disclosures, supplier provenance — by hand from documents scattered across email, supplier portals and ERP systems, with sales reps losing whole afternoons to it. Twistag designed the platform's data model first, then built four composable AI agents that reason over it. Revenue grew roughly 3x in the year following launch.

European regulatory obligations on cosmetic brands and their ingredient suppliers have grown faster than the teams handling them. A single product launch can require evidence spanning REACH, CLP, food-contact regulations, allergen disclosures, supplier provenance, and dozens of customer-specific compliance forms — each assembled by hand from documents scattered across email, supplier portals and ERP systems. Sales teams spent whole afternoons answering customer regulatory inquiries instead of selling. R&D waited weeks on supplier data to finalise formulations. Procurement chased the same vendor information every quarter. The cost was not only inefficiency; it was deals lost to faster competitors and launches delayed by paperwork. The client is a European RegTech startup serving the cosmetic industry and is not named publicly.

Twistag built the platform as an operating system for the cosmetic supply chain rather than as a tool bolted onto an existing one. That meant designing the data model first — a single source of truth covering raw materials, suppliers, formulations, regulatory documents and customer-specific compliance requirements — so the agents had something coherent to reason over.

Four agents ship in production against that shared model. A regulatory inquiry agent reads inbound customer questions and assembles answers from the document store and the supplier data attached to each ingredient, citing sources and flagging anything that needs a human. A supplier diligence agent collects, verifies and refreshes vendor documentation. A formulation agent surfaces compliant raw-material combinations for R&D. A procurement agent handles routine purchase-order workflows. Because all four reason over one schema, they compose: the inquiry agent can use supplier data the diligence agent maintains.

The compliance rep is not asked to trust a generated paragraph. She is shown where each claim came from and told which parts need her. Models sit behind a provider abstraction and were swapped twice during the engagement without the product changing shape. The data model was the first deliverable rather than preparation for one. Everything the agents do — citing a source, refreshing a supplier document, surfacing a compliant raw-material combination — resolves against the same single source of truth covering raw materials, suppliers, formulations, regulatory documents and customer-specific compliance requirements. A firm that thinks of itself as a model shop builds around the model; a product company builds so the model can be replaced.

Stack: OpenAI, AWS Lambda, PostgreSQL, Next.js. Services: AI agents and multiagent systems, AI-ready data platforms, product engineering.

Customer regulatory inquiries that took a sales or compliance rep three to four hours now resolve in under an hour — 75% less time per inquiry — and a meaningful share never reach a human at all. The platform is in production at three of the top ten European cosmetic brands, with several more in pilot. Revenue grew roughly 3x in the year following launch. The inquiry agent's source citations are what made the compliance team willing to send an answer without opening the underlying documents themselves, and the platform now carries the diligence, formulation and procurement workflows on the same schema.

The agents are independently useful and dramatically more useful together — which is the architecture decision that turned a feature roadmap into a product.