Systems measured in gift seasons and best-before dates

Ochugen in summer and oseibo in December bring us about 28 percent of annual revenue in a few weeks. Orders arrive from department-store gift catalogues, our own shops, our online store and corporate buyers, each in its own format. I design the intake so that every order reaches one queue with a recipient, a delivery promise and a noshi instruction checked against the sender's own wording. Capacity is planned from the shipping centre outwards: carrier pickups, packing hours and label printing set the daily limit, and every order channel is told that limit in advance. Recipient addresses are personal information that the recipients never gave us themselves, so we use them for delivery and nothing else, in line with Japan's personal-information law. I review each season's error log myself before the next catalogue goes to print.

Our sweets carry best-before periods from one day for fresh namagashi to several months for yokan and fruit jellies. Shop stock is counted at closing; department-store counters run on consumption-based purchasing, so their sales reach us the next morning; online orders arrive through the night. I will not let a screen promise stock that exists only in yesterday's count. Where items are short-dated or limited, we reserve production for online and collection orders at the planning stage and let shops sell their own shelves. Longer-dated products can be pooled across the shipping centre and the two factory shops, with a safety margin set by merchandising. My team owns the data and the rules; merchandising owns the quantities. The test is simple: a customer collecting a reserved box at five in the evening should find it waiting.

Our 1998 system held merchandising, recipe costing, production plans and gift orders together, and it was written and maintained by Toyasumi Data Service, a small vendor in Utsunomiya. A single cutover would have been cheaper on paper and far riskier in December. I divide a replacement into stages that can each survive a gift season on their own: gift orders first, then merchandising and production planning, with interfaces tested against last year's peak data. I keep the incumbent vendor engaged and paid during the transition rather than treating it as an obstacle, and I make documentation of its data definitions a contracted deliverable. The board receives each stage with its parallel-running cost, its exit condition and the date after which the old component is switched off. Legacy systems rarely fail all at once; they become impossible to change.

Japanese consumption-tax and invoice rules have changed our systems twice in seven years. From October 2019, takeaway sweets stayed at the reduced rate while café service moved to the standard rate, and gift sets combining sweets with a non-food item needed a rate decided by our tax advisers. From October 2023, the qualified invoice system required registration numbers and tax by rate on invoices to corporate buyers, simplified invoices on shop receipts, and confirmation of the settlement statements that department stores prepare for consumption-based sales. My approach is to hold the rate and invoice treatment as an attribute of each product and channel, set by finance and checked by advisers, rather than as a rule buried in till programs. Our systems then apply decisions that finance has made; they do not make those decisions themselves.

Labour, not hardware, now limits most of our plans. Many of our shops open with two or three people, often part-time, and seasonal packers for December are harder to recruit each year. Before approving a system change, I ask the shop operations team how many minutes it adds to or removes from a closing routine. We print packing labels in delivery-route order so that a new packer can work from the sheet without memorising codes, and we ended address retyping by taking department-store gift files directly. I also avoid features that depend on staff noticing exceptions under pressure. When a design needs a skilled person at the busiest hour, I treat that as a defect in the design rather than a training problem. The aim is fewer decisions at the counter, not more screens behind it.