Just Eat Takeaway.com
Rebuilding a 17-year-old engine room
Design leadership for the refactor of TMS, the back-office platform behind a global food marketplace: 16 development teams, one design system, and an AI assistant that helped save 1,000+ FTEs.
Problem
Everything at Takeaway ran through TMS: menus, orders, delivery areas, customer service. Seventeen years of accretion had made it powerful and brittle, and every operational team had learned to work around it rather than with it. Refactoring it meant redesigning the tools a whole company stood on, without stopping the company, with 5 designers and 1 researcher across 16 development teams.
Outcome
The Saffron design system, and a module-by-module rebuild into a high-performing microservices platform. An AI assistant that contributed to 1,000+ FTEs saved. Customer service consolidated from three platforms into one, saving roughly 10,000 minutes a day and cutting ticket resolution time by 85%. Restaurants got self-service menus and delivery areas that adjust themselves to order load.
Snapshot
- Client
- Just Eat Takeaway.com
- Years
- 2017-2022
- FTEs saved
- 1,000+
- Resolution time
- 85% faster
- CS minutes
- ~10,000 / day
- Dev teams
- 16 directed
Context
The platform behind the platform
Customers see a menu and a courier. Behind them sits TMS, the back-office platform that ran operations for the whole marketplace, and had been accreting logic for seventeen years. Every order, menu change, delivery area, and support ticket passed through it. It was the kind of system nobody gets to rewrite from scratch, because the company cannot stop running while you do.
Systems
Saffron, and a rebuild you could ship
The refactor worked module by module, moving TMS into a microservices platform without a big-bang cutover. The Saffron design system carried the consistency: with 5 designers and 1 researcher serving 16 development teams, reusable patterns were not a nicety, they were the only way the ratio could work. Design leverage came from the system, so the small team could direct the rebuild instead of chasing it.
Operations
Three platforms into one
Customer service agents worked across three separate platforms, and the seams between them cost minutes on every ticket. Consolidating them into one saved roughly 10,000 customer-service minutes a day, about 24 FTE at MVP, and cut ticket resolution time by 85%. An AI assistant layered on top contributed to more than 1,000 FTEs saved. In an operation that size, interface decisions are staffing decisions.
Self-service
Letting restaurants run their own shop
The same rebuild pushed control outward. Menu management gained EU additives and allergens compliance, AI-powered OCR input, add-on groups and variants, and bulk pricing and stock editing, so restaurants could manage their own catalogue instead of filing tickets. Delivery areas became self-service polygons that adjust dynamically to order load, keeping promises to customers honest when the kitchen is drowning.
Lesson
Trust is designed in small units
The strongest product moments were often modest: a clearer label, a better default, or a status that arrived exactly when a person needed it. Marketplace products live and die on those small promises, and a back-office refactor is where most of them are kept.