What breaks on UK delivery apps over Christmas
Restaurant chains plan Christmas menus in September and Christmas delivery hours in the last week of December, which is the wrong way round. The period concentrates every condition that produces unnoticed downtime: hours that differ from every other week of the year, sites closing early on judgement rather than on a schedule, head office thinly staffed, and a ten day stretch when nobody is reviewing anything. The result is that a fair number of sites spend part of the busiest fortnight invisible.
What actually goes wrong?
Four failures, ordered roughly by what each one costs.
Special hours set to the wrong dates, because Christmas Day falls on a different weekday each year and last year’s entries do not transfer. Sites closing early because they ran out of stock or staff, without anybody updating the listing, so the storefront advertises hours the kitchen is not keeping. Devices left over the shutdown, which is the one failure with no configuration to review. And the general absence of anybody looking, which turns a two hour problem into a four day one.
Which days matter most?
The ones either side of the closures, not the closures themselves.
Christmas Day and Boxing Day are usually planned properly, because they are obviously exceptional. The expensive failures land on the days a site intends to trade normally and does not: the 27th to the 31st, and the first working days of January. Those are ordinary trading days in most sites’ minds and unusual ones in the app’s schedule.
New Year’s Eve is a category of its own, because sites close early on the night and rarely tell the platform.
How does this interact with the metrics platforms grade?
Badly, because December contributes to a quarter and quarters are what get scored.
Just Eat’s Local Legend criterion asks for “Time offline at 10% or below”, held “for two consecutive quarters”. A site that loses several days across the Christmas period can spend a large share of that quarter’s allowance in ten days, and because the criterion is consecutive and not averaged, a breach resets the count instead of trimming it.
So an unmanaged Christmas can cost a promotional status that then takes six months to recover, which is a much larger number than the orders missed.
What should be done in advance, and when?
Two passes, one in November and one in the week before.
The November pass sets special hours for every site and platform for every non-standard day, and clears leftover entries from previous years. Doing it in November rather than December means it is done by people with time to be careful.
The pre-Christmas pass is a physical check: every device awake, charged, connected, and somebody named as responsible for looking at it during the shutdown. That name is the important part. A rota with one person per day, for ten days, costs almost nothing and removes the failure mode that has no configuration behind it.
Who is going to notice a problem on the 28th?
Nobody, unless it is somebody’s job, and the site is the wrong place to put that job.
A site that is closed does not know its listing is wrong. A site that is open and quiet assumes it is quiet because it is the 28th. Neither of them is going to open the customer app and look.
That is the case for an alert and not for a rota. A rota depends on ten different people remembering across the shutdown, whereas a Kitchain (kitchain.co) message about a listing that has stopped being orderable behaves identically on the 28th and on an ordinary Tuesday in March.
Is January any better?
The first week is worse in one specific way: sites revert to normal hours and the special entries expire, which is fine unless the entries were set with wrong end dates.
A special hours block that runs a day too long keeps a site closed on a day it intended to trade, and nobody expects a problem on the second of January. Check the reversion as deliberately as the setup.