The week everything goes wrong at once

Every trading business has a peak. For some it is the run-up to Christmas, for a garden wholesaler it is the first warm weeks of spring, for others it is the start of the school year or the first frost. Whatever the date, there is one week a year when volume is highest, people are most stretched, and the systems are properly tested for the first time since last year.

What makes that week hard is rarely a single failure. It is that the things which quietly coped all year stop coping at the same time, because they were never really automated. People with time were holding them together, and in the peak week nobody has time.

What breaks first

Synchronisation turns into overselling. A stock figure copied between systems every few minutes is fine when an item sells twice a day. When it sells twice a minute, the gap between updates is where the same last unit is sold to three people.

The reconciliation falls behind. The spreadsheet that ties the shop, the warehouse and the invoices together is kept by someone who, this week, is helping to pack.

Exceptions pile up. Each missing item, address change and duplicate costs someone a few minutes. At normal volume they are absorbed along the way; at peak they form a queue, and a queue is where orders get forgotten.

Temporary staff do not know the product. Extra hands at the packing stations and on the phones are essential, and everything that lives only in experienced people’s heads is out of their reach exactly when it is most needed.

The buffers are wrong. Safety margins set for an average week are too small for the products everyone wants and too large for the ones nobody does.

A cancellation costs more than the sale. Every oversold order means an apology, a refund and, on some marketplaces, a cancellation counted against the seller.

And then the returns

The second wave arrives after the peak is over. Returns from the busiest weeks come back together, into what was already the least automated part of the business, while stock counts are still recovering. For a December peak that means January, and a business that treats returns as an afterthought finds out then how far its numbers drifted during the busiest weeks.

What to settle before it arrives

Find the real limit. Not the advertised capacity of any system, but the point where your process breaks: how many orders an hour before stock lags, the picking queue grows or the order desk falls behind. Rehearse it against last year’s busiest day.

Freeze changes. No new integrations, price structures or warehouse layouts in the weeks before peak. Every change made in a hurry is an exception waiting for the worst possible moment.

Set buffers product by product. Deliberately, for the items that matter, instead of one global margin that is wrong in both directions.

Write the exception rules down. What happens when an item is missing (wait, split, substitute or cancel) decided once and in advance, so that a temporary colleague can apply it without having to ask.

Put product knowledge within reach of temporary staff. What fits what, what replaces what, what customers usually get wrong. Structured product data is an onboarding tool before it is anything else.

Plan for the returns. Space, people and a clear state for goods that have been received but not yet inspected, so that the weeks after peak do not quietly corrupt the stock figure.

The honest test

Peak is where the difference between one record and several synchronised ones stops being an architecture discussion. Less to synchronise means less that can fall behind when everything moves at once.

It is also the test any system should be judged by, including ours. No business runs on Mercantir yet, and the first ones will be watched through a full season, this week included, because software that has not survived it should say so before it says anything else.