The retailer whose Black Friday was decided in 2009

The situation

Retail in the UAE has compressed delivery expectations faster than most markets. A customer who can get groceries in twenty minutes does not accept three days for a pair of shoes, and the competitive set for a speciality retailer now includes marketplaces that treat logistics as the product.

Meeting that is not primarily a warehouse problem. It is an information problem: the order system has to be able to describe a promise the business wants to make.

The problem

This retailer's order system was written in-house in 2009 and had worked, in the narrow sense that it had not fallen over. It had also accumulated seventeen years of assumptions, and the load-bearing one was that an order belongs to a store.

That was correct in 2009, when every order originated in a shop. It made same-day delivery impossible in 2026, because a same-day order needs to be fulfilled from whichever location can actually reach the customer, which may not be the one that took the order.

Around this had grown an apparatus of manual work. About thirty-nine per cent of orders were being touched by hand at some point, usually by a supervisor reassigning an order the system had sent to the wrong place. The company had normalised this. It was described internally as "checking", not as failure.

Two previous proposals had recommended replacing the platform. Both were priced in millions and scheduled across two years, and neither survived contact with a board that had to keep trading through the replacement.

What we did

We started by disagreeing with the premise. A full replacement was a defensible recommendation, but it answered a question the business had not asked: it treated a seventeen-year-old system as uniformly bad, when in fact most of it worked and exactly one assumption was blocking the strategy.

Three weeks of analysis established which. Order-to-store binding was set at creation and threaded through inventory, picking, returns and reporting. It was deep, but it was one idea, not a hundred.

Rather than replace the system we put an order routing service in front of it. New orders are created against a fulfilment location decided at routing time from stock, distance and capacity, and the legacy system continues to receive an order already bound to a store, exactly as it expects. It never learns that anything changed.

This drew an objection worth recording: we were adding a component rather than removing debt, and somebody asked, reasonably, whether we were making the problem worse. Our view was that the debt is only worth paying down where it blocks something, and that a targeted bypass shipping in one season beats a correct rewrite arriving after two more peaks have been lost.

Returns and reporting followed, because an order fulfilled from a different location has to be returnable to one.

The outcome

Same-day delivery went live inside a single peak season rather than after two years.

Manual intervention fell from thirty-nine per cent of orders to about six. Supervisors who had spent their days reassigning orders were doing something else by the end of it, which is the outcome nobody puts in a business case and everybody notices.

The 2009 platform is still running. It will be replaced eventually, and that programme is now a planned piece of work rather than an emergency, because the thing that made it urgent is no longer blocking the business. Renovating around a load-bearing wall is less satisfying than demolition, but the shop stays open.