Three choices that slowed someone down

Forecasting route-days instead of 23,500 separate shops

In early 2023 our commercial team proposed a shop-level ordering module from a software vendor. It would forecast each of the 23,500 outlets our vans visit and place a suggested order on the salesperson's handheld. The licence and integration would cost RM2.4 million over three years, and the vendor's case assumed a six per cent increase in volume per call.

We piloted it for ten weeks in the Klang, Ipoh and Kuantan depots. At shop level, weekly forecast error exceeded sixty per cent. Orders moved with paydays, school holidays, a wholesaler's weekend offer and whether an owner had cash that morning. Salespeople overrode most suggestions. Aggregated to the route-day, the same history produced errors near fourteen per cent, which was useful for loading vans and for planning chilled soya, whose unsold bottles we write off.

The commercial director wanted to keep the module and train salespeople to follow it. I recommended declining it. We would forecast route-days and depot volumes, give salespeople a four-visit history at each shop and flag chilled lines where returns were rising. The trade-off was explicit: no suggested-order lever for sales management, and vans carrying a planned buffer of about twelve per cent on ambient lines. The group managing director accepted the recommendation, and the commercial director's disagreement was recorded in the paper.

In 2024 chilled soya returned unsold fell from 4.1 to 2.3 per cent of loaded volume, about RM3.2 million on the year's chilled sales of RM176 million. Routes on which a van ran out of a product before its last call fell from nine to five per cent. I cannot attribute all of that to forecasting; we also changed loading times at two depots. Volume per call did not rise, and the vendor's six per cent was never tested.

What I would do differently is ask salespeople why they were overriding the suggestions in the first fortnight, not the tenth week. Their explanations would have shortened the pilot considerably.

Coding every bonus carton at the counter, at the cost of two calls a day

By 2023 our traditional-trade schemes cost about RM47 million a year, much of it as bonus cartons given with qualifying orders. Nine depots recorded a bonus carton as a zero-price invoice line; seventeen deducted its value from the order. Salespeople could also match a nearby wholesaler's price on the spot. Our reports of net price per carton therefore mixed three conventions, and the commercial team was comparing them with field observations of competitors' shelf prices. Payment type was entered at the depot at night, so cash shortfalls were often settled without an explanation.

I proposed that the handheld require a scheme code for every bonus carton, a reason for any concession above three per cent and the payment type before an order could close. Hairul Azmi Sulaiman, our head of van sales, opposed it. Calls already averaged six minutes and routes were planned at thirty-eight calls a day. Our trials showed the new steps added about thirty-five seconds a call. We agreed to reduce planned calls to thirty-six a day for the first quarter, accepting fewer visits to marginal outlets.

The change went live across 26 depots in March 2024. Within six months it showed RM6.8 million of bonus stock given outside any approved scheme. Most had been authorised verbally by area managers to clear slow stock. Approved schemes were reduced from 214 to 61. Net price per carton in our reports now agrees with finance's ledger to within half a per cent each month. Unexplained cash variances fell from RM1.1 million in 2023 to RM0.3 million in 2025. Calls returned to thirty-seven a day once salespeople knew the screens.

My error was in how I first showed the out-of-scheme figure. I listed it by salesperson, and it read as an accusation against people who had followed instructions. Hairul made me redraw it by approval chain, which was both fairer and more useful to the commercial director. I would start there next time.

Moving 52,000 households' delivery records off distributors' phones

Outside the Klang Valley, Penang and Johor Bahru, eleven independent distributors deliver our 19-litre water to about 52,000 households. Customers sign up with us, through our app or hotline, so we are responsible for their personal data. In 2024 my team mapped where it was kept. Most distributors held names, addresses and telephone numbers in spreadsheets and messaging groups on drivers' personal phones. Some notes recorded gate codes, which family member would be home and which elderly customers lived alone.

Malaysia's personal data law had been amended that year to introduce breach notification and data protection officers, and to give processors security obligations of their own. I did not need the amendments to see the problem, but they sharpened the timetable. I recommended that every distributor use our delivery application on devices we supplied, sign processor terms and delete local copies within ninety days. The chief operating officer preferred a two-year phase, fearing that distributors would leave. We compromised on twelve months and RM0.6 million for devices and training.

Nine distributors accepted. Two, in Pahang and Terengganu, declined, and we took over their 7,800 households with our own crews at about RM2.10 more per bottle delivered. In the transfer quarter those households cancelled at 6.4 per cent, against our usual 3.1 per cent. By December 2025 all 52,000 records were on one system, access notes had become a short structured field, and our data protection officer had reviewed deletion with four distributors, with their consent, on their premises.

The cost was real: lost distributor relationships, a higher delivery cost in two states and some unhappy customers. What I would do differently concerns the access notes. We removed free text without asking households what they wanted retained. For several weeks, crews arrived at locked gates they had previously opened with an instruction we had deleted. A short message to each household before migration would have kept the useful notes and let customers remove the rest.