Promises we stopped making, and the ones we kept

Taking 'in stock at your store' off four in ten product pages

When I arrived in March 2021, our website showed store stock for every product at every store. Click-and-collect made up 17 per cent of online orders, and 11.2 per cent of those orders were cancelled after payment because the store could not find the item. Spring cycle counts put record accuracy at 74 per cent by product and store. Customers who drove to a store and found nothing complained more than any other group.

Collection cost a third as much to fulfil as delivery and brought customers into stores. I proposed a rule: show 'in stock at your store' only where that store's counts for the product category had been at least 95 per cent accurate over eight weeks and at least three units were recorded. Elsewhere, offer delivery to the store in three to five days. On day one, 41 per cent of store-and-product combinations would lose the claim.

Our vice-president of stores argued that stores would look empty online and lose customers who check before visiting. The chief commercial officer estimated C$12 million of lost orders. Our chief technology officer wanted to wait for a two-year RFID programme. I argued that two more years of 11 per cent cancellations would cost more trust than any orders we forwent. The chief executive backed the rule on condition that we reported lost orders weekly.

We counted the 600 best-selling styles weekly, fixed the receiving step where most errors began and tagged outerwear and footwear with RFID in 2023 and 2024 at a cost of C$13.5 million. Forgone orders came to an estimated C$9.8 million in 2022–23, and conversion on affected pages fell about 6 per cent. Cancellations fell to 4.9 per cent that year and 2.4 per cent by 2025–26. Record accuracy reached 93 per cent across the chain and 98 per cent in tagged categories, 88 per cent of combinations carry the claim again, and collection has grown to 24 per cent of online orders.

Stores, supply chain and product now review the rule quarterly. I would change two things. I would explain the rule on the page, because customers read 'ask the store' as 'out of stock'. And I would have tagged outerwear, our worst category, in year one rather than awaiting a chain-wide programme.

Writing a bilingual grading language before a single resale listing

By the autumn of 2022 about 9,000 traded-in jackets, packs and boots sat in storage in Vancouver. We had accepted trade-ins for store credit since 2021 but had no way to sell them: every item was unique, and nobody buys a used jacket undescribed.

Our merchandising and marketing teams proposed launching online resale in April 2023 with free-text descriptions written by graders, telling each item's story, and launching first outside Québec. Translating free text had been quoted at about C$7 a listing and two days' delay, on items averaging C$118. Our general counsel was clear that a Québec launch could not simply be in English, and I was not willing to buy trade-ins from Québec customers, who sent a large share of them to our Montréal centre, and then sell them only to the rest of the country.

I proposed putting the substance of each description into structure. With graders from both centres, our content team wrote five condition grades and 64 codes for defects, repairs and wear, each written once in English and once in French by writers working in their first language. A listing became a set of codes, photographs at fixed angles, measured dimensions and an optional grader's note. The trade-off was real. Launch slipped eleven weeks to July 2023, the story-led campaign was dropped, and graders were slower for their first month. A marketing director told me, fairly, that we had made something charming bureaucratic.

Once graders knew the codes, listing time fell from 14 minutes to about six. In 2025–26 we resold 38,000 items for C$4.5 million; Québec customers bought 23 per cent of them, slightly above their share of new online sales. Resale returns ran at 7.8 per cent, against 16.5 per cent for new clothing bought online, because customers knew what they were getting. The same codes now classify returns and repairs, and two products have been redesigned after the codes showed a recurring seam failure.

I would do one thing differently. The first list was drafted in Vancouver and checked in Montréal, and within two months our Montréal graders showed us eleven codes that French-speaking customers read differently. I would write the vocabulary with both centres in the same room from the first day, and test it with French-speaking customers before launch rather than after.

Letting members leave in two steps and losing 61,000 of them

Sablemere launched a paid care membership in 2019: C$49 a year for free repair labour, free return shipping and early access to new products. By January 2022 it had 212,000 members and earned C$10.4 million in fees. Much of that growth came from a checkbox. First-time buyers of jackets over C$250 found membership pre-selected at checkout, a growth experiment that had quietly become permanent. Renewal was automatic, with no reminder, and cancellation was possible only by telephone during Pacific business hours, which was awkward for anyone in eastern Canada. We received about 1,900 card disputes a year and 3,400 cancellation calls a month.

In October 2022 I proposed removing the pre-selection, sending a renewal reminder thirty days ahead, putting cancellation online in two steps from the account page and offering a pro-rated refund on request in the month after renewal. The chief commercial officer disagreed. Members spent 2.3 times as much as other customers, and he read that as the programme's effect. The chief financial officer asked me to phase the changes over two years to protect the margin plan. My view was that much of the spending gap reflected who had been enrolled, buyers of expensive jackets, rather than anything the membership did, and that consumer-protection expectations on recurring charges were moving in one direction only. The board chair backed acting at once.

Membership fell to 151,000 by January 2024, and fee income fell by about C$3 million a year. Card disputes dropped to 140 a year and cancellation contacts by 88 per cent. Repair bookings per member rose 40 per cent, because the people who stayed had chosen the programme, and 71 per cent of deliberate joiners renewed. Membership has since recovered to 166,000. Any default selection of a paid product now needs my sign-off and our general counsel's.

What I would do differently is harder to write. Six months after the change, the committee chair asked what we had done for the people enrolled through the pre-ticked box. We had done nothing proactive. We then wrote to the 18,400 members enrolled that way in the previous twelve months and offered full refunds; 6,100 accepted, about C$0.3 million. I should have done that on the day we removed the box.

Holding a new checkout through the holiday season

In 2024 we replaced our web storefront: product pages, search and checkout. Launch was planned for September, before our busiest trading weeks. In July, paid testers from a Vancouver co-operative of disabled people found 47 failures against WCAG 2.1 Level AA in the new checkout. Nineteen stopped a screen-reader user from paying: focus vanished in the address look-up, error messages were never announced, the payment provider's card fields were unlabelled and a session expired without warning.

The old checkout ran on a platform whose licence ended in October, and extending it through January, with engineers maintaining both systems, would cost C$2.1 million. Marketing had projected that the new checkout would add C$5.5 million of sales over the season. Our chief technology officer proposed launching everything in September and fixing the failures within eight weeks; the chief marketing officer agreed with him. I argued that we would be knowingly releasing a checkout some customers could not use, at the moment they most needed it, and that we have more than fifty employees in Ontario, where accessibility standards apply to our public website. The chief executive agreed to hold checkout.

We launched the new product pages and search in September as planned and kept the old checkout behind them. By December we had fixed 44 of the 47 failures. The remaining three sat in the payment provider's hosted card fields; the provider fixed them in April 2025, after we made conformance a condition of renewing its contract. The new checkout went live in January 2025.

We spent the C$2.1 million and forwent the projected uplift. In moderated sessions after the fixes, 22 of 24 disabled testers completed a purchase, against 9 before. Checkout conversion for all customers in 2025 ran 3.8 per cent above the old checkout, and accessibility complaints fell from 140 in 2024 to 22 in 2025. Accessibility criteria now sit in every team's definition of done, and every technology contract we sign includes a conformance clause.

I would change the timing of the testing. We found the failures in July because we tested in July. Had we paid testers to use prototypes in January, fixing the focus order would have cost a designer an afternoon rather than an engineering team a month. I would also have written the conformance clause into the payment contract before we signed it.