Modernising a Platform the Business Depends On Every Day
Modernising a Platform the Business Depends On Every Day
At Tesco I own the national ordering engine, the system that decides what gets ordered, in what quantity, from which supplier and to which depot. I led its move from a legacy monolith to an API-first architecture. Modernising a system like that is different from building something new, because the business cannot stop while you work on it.
Start with what the system decides
It is tempting to describe a modernisation in technical terms: services, interfaces, deployments. The more useful starting point is the decisions the system makes. Once the team agrees on those, it becomes clear which parts must be preserved exactly and which can change.
Make the pieces separately releasable
In a monolith, a small change to one algorithm can mean a large release of everything. Moving to an API-first design let us release algorithms as modules. That shortened deployment cycles from weeks to days, and it also changed how product and engineering plan. Smaller releases mean smaller risks and faster learning.
Run old and new side by side
When the output of a system becomes real orders, confidence matters more than speed. Comparing the new path against the old one before switching over gives the business evidence instead of promises. It also surfaces the hidden rules that the legacy system had been applying for years without anyone writing them down.
Keep the users close
Supply chain teams know the patterns of their depots and suppliers better than any document. Involving them in reviewing outputs during the migration catches problems early and builds the trust needed for the switch.
Modernisation is what makes the next step possible
The value of an API-first platform is not only the cleaner design. It is that new capabilities, such as better forecasting, demand signals and inventory optimisation, can be added without another rewrite. That is the real reason to do the hard work of modernising.
Where AI fits
Building AI products in the Curious PM cohort made me think about how much of supply chain work is judgement on top of data. Models can help a team understand why the engine made a decision. The decision rules themselves should stay clear and owned by people.
Sunil skipped presentations and built real AI products.
Sunil S Sakri was part of the August 2026 cohort at Curious PM, alongside 24 other talented participants.