The channel that stayed open

In March, retail's slow decade of digital transformation was compressed into a weekend. Stores closed; the website, the app, and the partner marketplaces became not a channel but the business. Six months on, with shops reopened and the numbers in, it is possible to say what separated the retailers who bent from the ones who broke.

The honest version of the story is less about heroics during the crisis than about decisions made quietly in the years before it. By the time the stores closed, most of what determined the outcome had already been built or left unbuilt. The spring did not create the difference between retailers; it revealed it.

It was rarely the storefront

The public failures of the spring looked like website crashes, but the sites themselves usually held. What buckled was behind them: the inventory service asked for real-time truth by five channels at once, the order APIs that suddenly fed lockers and curbside flows invented in a week, the integrations between store systems and e-commerce that had always been reconciled overnight and were now expected to agree by the minute.

The pandemic did not test anyone's website. It tested the seams between systems that had never been asked to work this hard.

When batch became real-time overnight

One shift did more damage than any raw traffic spike: the collapse of the overnight window. Retail had spent decades building on a comfortable assumption that systems could disagree during the day and reconcile at night, that the store's view of stock and the website's view of stock needed only to converge by morning. Curbside pickup and click-and-collect destroyed that assumption in a week, because a customer standing in a car park needs the inventory to be true now, not by the next batch run.

Retailers whose architecture already treated inventory and orders as live, shared services absorbed this without noticing it was hard. Those whose systems relied on nightly reconciliation found themselves trying to convert batch processes into real-time ones under maximum load, with customers watching. It was the least forgiving possible moment to discover that "eventually consistent" and "consistent enough to sell against" are not the same thing.

What the resilient ones had

• Inventory as an API product: one governed source of stock truth that every channel consumed, instead of five copies agreeing eventually.
• Room to improvise safely: click-and-collect in a fortnight was possible where new consumers could be onboarded onto existing APIs, with quotas protecting the core.
• Visibility by consumer: when everything surged, they could see which channel was straining what, and shed the right load instead of all of it.

None of that was built in March. It was built in the unglamorous years before, by teams treating integration as infrastructure rather than plumbing. The retailers now planning for an uncertain winter might take the lesson while it is fresh: the next disruption will also arrive without notice, and it will also be won or lost at the seams.