What each decision gave up, and why

Postponing code payments until refunds reconciled under two tax rates

In 2019 the consumption tax was due to rise to 10 per cent on 1 October, with most food remaining at a reduced 8 per cent. Nearly half our sales were food, so almost every basket in our then 196 stores would carry both rates. Payment companies were promoting code payments heavily that autumn, and customers had begun asking for them at the counter. Our sales director and the president wanted code payments to start on the day of the tax change.

A six-store pilot in July used standalone tablets beside the tills. Sales worked. Partial refunds did not: in about one refund in forty, the tablet returned an amount without knowing which tax rate it belonged to, and store accounting had to correct the record by hand. Across 196 stores, that meant daily hand corrections by staff already short of hours.

I recommended delivering the tax change on 1 October and adding code payments only when they ran through the tills. The sales director disagreed strongly; competitors would accept code payments from October, and we would lose four months of goodwill. The president accepted my recommendation on condition that I committed to a fixed date. We gave up those months and paid ¥340 million for till integration instead of ¥95 million for tablets.

My team rebuilt refunds so that every line carried its original rate, tested 3,200 refund cases with store accounting and trained about 1,900 checkout staff by store cluster. Code payments started in all stores on 3 February 2020. Cashless transactions rose from 23 per cent of the total in September 2019 to 37 per cent in June 2020. In the first six months, 3 of about 41,000 code-payment refunds needed correction. We never measured reliably how many customers shopped elsewhere in the meantime.

The till has since been the only place a payment method may be added. I would start the tax-rate work in spring rather than summer, and I would announce the launch date in August instead of leaving checkout staff to explain the delay daily.

Leaving store staff the final word on daily-food orders for a year

By 2023 daily foods, meaning bread, milk, tofu, eggs and prepared dishes, were about ¥31 billion of our sales, and disposal had reached 3.6 per cent of them. Ordering took an average of 74 minutes a store each day, usually by a part-time leader at the end of a shift. A forecasting supplier proposed fully automatic ordering across the chain and projected disposal of 2.0 per cent. We were short of about 260 part-time staff, and Ryota Kanezumi, our merchandising director, wanted automatic ordering in every store from October 2023 to release hours before winter.

I did not think a model trained on our sales history yet knew enough about each town. I proposed that the system suggest every order, that staff confirm it, and that any change beyond a set margin carry one of nine reasons. An item family would move to fully automatic ordering only after eight consecutive weeks within an agreed error band. Ryota disagreed, and the supplier warned that overrides would dilute its model. The reason screens cost ¥120 million and delayed full automation by about six months, so the year's labour saving was smaller than planned.

We piloted in 24 stores from July 2023 and reached every store by January 2024. The reasons revealed school events, local festivals, funerals in small towns and, in Kagoshima city, ash-fall forecasts, which we then added to the model.

In the year to February 2025, disposal fell from 3.6 to 2.6 per cent and ordering time from 74 to 31 minutes a day. Overrides fell from 34 per cent of order lines in the first month to 11 per cent. By February 2025, 58 per cent of daily-food items were ordered automatically; prepared dishes still needed confirmation. We did not reach the supplier's 2.0 per cent.

Merchandising now reads the reason report every week. I would separate prepared dishes from packaged foods at the start, and I would agree a disposal goal with store managers instead of adopting the supplier's figure.

Keeping medicines and sensitive purchases out of a supplier data service

In 2024 our marketing division proposed a supplier data service. Manufacturers would buy aggregated analysis of points-card purchases and fund offers to selected members in our app. The plan projected ¥640 million of income in the first year, of which ¥270 million came from medicines and health categories, the data manufacturers valued most.

Our members' purchases include cold remedies, pregnancy tests, incontinence care and hair-loss treatments. Dispensing records were never in scope. Our lawyers advised that aggregated analysis could be designed lawfully. They could not tell us how a member would feel on receiving an offer that revealed what the store had inferred from such a purchase, and neither could the projection.

I recommended excluding all non-prescription medicines and 41 sensitive product groups from both analysis and offers, and requiring a separate opt-in before any supplier offer reached a member. The marketing director disagreed, and two directors asked whether we were surrendering the most valuable part of the service to avoid a hypothetical complaint. The board approved the exclusion in March 2025. We gave up most of the ¥270 million, and two medicine makers declined to take part.

Pharmacists and category buyers tagged 6,300 product codes into the excluded groups. My team built the exclusion into the data pipeline itself, so that excluded purchases never reach the environment suppliers' analysts use, rather than hiding them in reports.

The service started in September 2025 and earned ¥352 million in its first twelve months. Of app users, 46 per cent opted in to supplier offers. We received 19 complaints about offers in the year, none about a health inference, though a low count proves little on its own. Part of the shortfall against plan came from slower take-up rather than the exclusion.

The boundary is now a standing rule reviewed each year by pharmacy operations and legal. I would settle it before signing letters of intent with suppliers; three manufacturers had planned around categories we later withdrew, and that cost us trust we are still rebuilding.