Chief Product Officer / United States
Gains reported beside the price paid
- Annual company revenue
- US$1.62bn
- Year to 31 January 2026; 47 per cent from trade contractors
- Home centres and lumber yards
- 58
- Washington, Oregon, Idaho and Montana
- Product and engineering colleagues
- 172
- In my organisation; store infrastructure sits with the CIO
- Contractor app orders placed ahead for pickup
- 26%
- Fiscal 2025, three years after launch
Decisions & consequences
01Delaying contractor order-ahead until the yard crews could see the orders
In January 2022 our contractor app had about 9,400 active trade accounts. Their most frequent request was order-ahead pickup: order the night before and collect at the yard gate next morning. The customer screens were nearly finished, but our 58 yards still worked from pick tickets printed at the pro desk. An overnight order would sit in a queue no yard associate could see until the contractor arrived.
Delphine Arkwrenne, our vice-president of pro sales, wanted the feature live for the spring building season. A national chain had launched a similar service locally, and two of our larger framing contractors were trying it. She argued that an imperfect launch was better than none. I disagreed: a contractor kept waiting twenty minutes for an order placed the night before would trust the app less than one never offered the feature.
I recommended a five-month delay to our chief executive and moved two of our eight teams from loyalty work to build a handheld pick list for yard crews, showing bay location, staging area and a photograph of the completed load. Delphine put her case to the chief executive directly, with my support, and the chief executive accepted the delay. It postponed the loyalty redesign by two quarters and added US$1.9 million for rugged handhelds and printers.
We piloted in six yards in August and reached all 58 by November 2022. By fiscal 2025, order-ahead carried 26 per cent of contractor app orders. The median wait for a staged order was seven minutes, against twenty-one for an unstaged desk order, and mis-picked lines fell from 3.8 to 1.4 per cent. The delay had a cost: pro sales traced about US$2.3 million of annual purchases from two framing contractors to that competitor in the spring, and only one returned.
I would now trial staged pickup on paper in two yards before writing software. That would have shown the yard's needs within weeks and given pro sales a firm date.
Taking our largest trade accounts out of the household points programme
Since 2016 trade accounts had been able to join the same points programme as household customers, earning two per cent of spending. By 2023 the programme cost US$15.8 million a year, and 61 per cent of that cost went to about 4,200 trade accounts buying more than US$60,000 a year. Their purchases followed job schedules and competing quotes. In a five-month test across two regions, removing points from a matched group of large accounts produced no measurable change in their spending.
Our chief marketing officer wanted one programme under one name for every customer. Delphine Arkwrenne worried that larger contractors would read any change as a cut and move their accounts. Our chief financial officer argued for ending points for all trade accounts. I proposed a middle course: large accounts would move to delivery credits, a dedicated order-desk line and quarterly purchase reviews, while smaller contractors and household customers stayed in points at 1.5 per cent.
The trade-off was explicit in the paper we took to the executive committee. About 1,100 mid-sized contractors, buying between US$60,000 and US$150,000 a year, would receive less value from the new benefits than from points. Marketing gave up a single message. We gave ninety days' notice and wrote to every affected account with its own comparison.
Programme cost fell to US$10.9 million in fiscal 2024. Retention among large accounts was 94 per cent, against 95 per cent the year before, and 71 per cent of them used their delivery credits. About 380 accounts complained, and pro sales identified 140 whose spending fell noticeably. In mid-2024 we added a bridging tier for accounts just above the threshold.
Next time I would give six months' notice and build the bridging tier from the start. Our data had shown that mid-sized accounts would lose value; I underestimated how strongly a short notice period would read as a decision made without them.
Replacing guessed home projects with a list customers write themselves
In 2024 our data science team built a model that inferred a household customer's project from recent purchases and browsing: a bathroom remodel, a fence repair, a deck. Emails and app prompts then suggested the next stage, such as grout after tile. Measured against a holdout group, the programme was credited with US$6.4 million of incremental sales in its pilot year.
The complaints were few in number and serious in content. About 1,900 customers wrote or called within twelve months. Several shared an email account with other members of the household, and the prompts disclosed purchases intended as gifts or not yet discussed. Others were uneasy that we had guessed a project they had never mentioned. Unsubscribe rates on those emails reached 0.47 per cent, against 0.21 for ordinary promotions. Oregon and Montana had also brought comprehensive consumer-privacy laws into force during 2024, and our counsel advised a fresh review of how inferences were used.
I recommended switching off inferred-project prompts and replacing them with a project list that customers create, name and choose whether to receive reminders about. Our chief marketing officer and the data science lead disagreed; the attributed revenue was real, and they proposed gentler wording instead. I argued that wording would not change what we had decided to know about a customer without asking.
In the first twelve months, 214,000 rewards members created 302,000 projects. Attributed incremental sales were US$3.9 million, US$2.5 million less than the inferred programme. Complaints fell below 100 and the unsubscribe rate returned to 0.23 per cent. Service-desk associates now open a customer's project list when the customer chooses to share it.
I would ask the question before building the model. Every personalisation proposal now passes a written test: would a reasonable customer expect us to know this, and could they see and change it? Applying that test first would have saved a year.