Five disciplines for data gathered in a doorway

A sundry shop's order depends on cash in the drawer, the owner's credit with us, a nearby wholesaler's offer and whether the van arrived before or after payday. Forecasting each of 23,500 shops every week produced errors above sixty per cent and suggestions that salespeople ignored. I forecast at the levels where errors partly cancel and decisions are actually made: route-days for van loading, depot and product for replenishment, and plant lines for production. At the shop, the salesperson sees a short history and decides. For chilled soya, where unsold bottles are written off, we add flags based on recent returns. I judge forecasts by the cost of their errors rather than a single accuracy percentage, because a missing carton of barley drink and an expired bottle of fresh soya do not cost the same.

Our list price rarely describes what a shopkeeper paid. Bonus cartons, on-the-spot concessions and scheme rebates change the net price, and our depots once recorded each of them differently. I define net price per carton from the transaction, with every bonus carton linked to an approved scheme and every concession above a set limit given a reason. That lets commercial colleagues see which schemes changed buying and which simply moved stock between depots. Field observations of competing products are useful, but I keep them to what our salespeople see openly on shelves and price boards. I do not seek competitors' prices or volumes from distributors who also carry them, and our legal counsel reviews any arrangement through which commercial information could pass between businesses that compete.

More than half of what our vans collect is cash, about three payments in ten arrive by QR, and approved outlets buy on short credit. Each mode tells us something different about a shop and carries its own risk. I make sure payment type is captured at the counter with the order, so that a depot reconciliation explains variances rather than absorbing them. Cash shortfalls are investigated with finance and depot managers as process questions first; my team supplies the pattern, not a verdict on an individual. Credit limits use the payment history an outlet has built with us, not data bought about its owner. As QR use grows, I watch for a quieter risk: assuming that shops which still prefer cash are less valuable or less reliable. Often they are simply older businesses.

Our water customers give us names, addresses, telephone numbers and access instructions, and some households also join a rewards scheme in our app. That data sits with our own crews, eleven independent distributors and several service providers. I work with our data protection officer to keep one record of who holds what, under written terms, on systems we can audit. Marketing use depends on the choice a customer made when joining, and the default is no. Malaysia's personal data law, amended in 2024 to bring in breach notification and data protection officers, and Singapore's own law for our customers there set the floor rather than the design. I would rather hold less personal data well than collect access notes and family details we cannot protect.

Supermarket and convenience chains offer us their point-of-sale data, and distributors report sell-out from their own systems. Both help forecasting. Both also open a channel through which details about other suppliers could reach us, or ours could reach them. I set the terms in advance: we receive data about our own products and categories at aggregated levels the retailer is entitled to share, and we do not supply our traditional-trade prices or outlet lists to chains that compete with those shops. Data received from distributors who carry rival brands is limited to our products. These boundaries cost us analytical depth and, occasionally, a joint promotion. I record each exception, who approved it and when it ends, so the board can see the boundary rather than trust that one exists.