Hisense

Hisense is a global consumer electronics and home appliance group whose televisions, refrigerators, air conditioners and white goods reach consumers through an unusually broad mix of channels. In Southeast Asia the group sells through national retail chains, independent dealers, e-commerce marketplaces, direct-to-consumer storefronts and project channels supplying developers and hotel groups. Each of those channels expects a different fulfilment behaviour from the same regional inventory pool.

The regional distribution network covering Thailand and Malaysia is the operational heart of that business. It receives containerised imports from manufacturing sites, holds bulky finished goods that consume large amounts of cubic space, and ships everything from a full-truckload replenishment for a hypermarket to a single refrigerator delivered to a consumer's apartment with installation and old-unit takeback.

Volumes are also strongly seasonal and promotion-driven. A regional shopping festival can multiply daily order lines several times over within a week, and the same promotion often shifts demand between channels — pulling volume away from store replenishment and into direct-to-consumer parcels that need a completely different pick, pack and delivery pattern.

Warehouses were run largely on paper and on the knowledge of long-serving supervisors. Putaway locations were chosen by whoever was on the forklift, picking was list-based, and stock accuracy was maintained through periodic full counts that required halting operations. Transport was arranged by phone against a small pool of familiar carriers, with the delivery record captured as a signed paper note that returned to the office days later.

The model worked when volume was predictable and channels were few. It did not survive the shift to marketplace and direct-to-consumer volume, where order lines are numerous, small and time-sensitive, and where a missed delivery window is visible to the consumer immediately and to the platform's seller rating shortly afterwards.

Leadership framed the programme around a single test: could one regional inventory pool make a promise that every channel could rely on? Everything else — productivity, damage rates, delivery performance — was treated as a consequence of that question rather than as an independent target, which kept the scope coherent when individual functions pushed for their own priorities.

The team also set a deliberate constraint on the design. Bulky home appliances make automation-heavy warehouse concepts expensive and inflexible, so the programme aimed to raise the productivity of people rather than to replace them with equipment. That decision shaped every subsequent choice, from directed picking paths to how temporary peak staff are onboarded.

When stock accuracy sits in the low nineties, every channel commitment becomes a gamble. A marketplace listing shows availability that may not exist; a store replenishment order is confirmed against a quantity that turns out to be damaged or misplaced; a project shipment is promised against inventory that has already been allocated informally to another customer. The organisation absorbed those failures through expediting, substitution and apology, all of which cost money and none of which appeared as a line in the logistics budget.

Cycle counting was impossible at the required frequency because counts required stopping work. That created a slow feedback loop: an error introduced on Monday might not be discovered until the next full count weeks later, by which time it had propagated into planning, allocation and financial reporting.

Paper-based picking scales linearly with headcount, and during a festival peak there simply are not enough experienced pickers available to hire. Temporary staff needed days of shadowing before they were productive because the knowledge required — where things are, which SKU variants look identical, which items need two people — lived in people rather than in the system.

The result was a hard ceiling on throughput. Beyond a certain daily order line count, the warehouse could only respond by extending shifts, which raised cost per line and error rates simultaneously.

Home appliances are expensive to store badly. A poorly planned putaway wastes cubic space that cannot be recovered without re-handling, and re-handling large appliances is the single largest cause of in-warehouse damage. Without system-directed putaway and slotting, space utilisation depended on judgement under time pressure, which is exactly the condition in which judgement degrades.

Damage also carries a disproportionate downstream cost in this category. A scratched refrigerator is not a partial loss; it becomes a returns, refurbishment and disposal problem, and in a direct-to-consumer order it becomes a failed delivery with a redelivery cost attached.

Cross-border and domestic transport was arranged manually with a mixed carrier pool. Once a truck left the yard, the shipment was effectively invisible until a paper proof of delivery came back. Customer service could not answer a delivery question without calling a carrier dispatcher, and exception handling was reactive by definition — the company learned about a failed delivery from the customer.

For the project channel, where a hotel or developer expects a coordinated multi-item delivery to a construction site, the absence of reliable arrival information forced conservative scheduling and frequent on-site waiting.

The regional distribution centres were converted to system-directed operation on WMS Cloud. Receiving is scanned against the inbound ASN, putaway is directed by slotting rules that weigh velocity, cube and handling class, and picking is organised into waves that match the shape of the day's demand rather than the order in which orders arrived.

Crucially for a business with high staff turnover during peaks, the system now holds the operational knowledge. A new picker follows a directed path with scan verification at every step, which shortens ramp-up from days to hours and makes an inexperienced operator's error rate close to that of a veteran.

The same inventory pool now serves retail replenishment, marketplace orders, direct-to-consumer deliveries and project shipments under different rule sets. Allocation priority, packing specification, documentation and carrier selection all vary by channel automatically, so a promotional surge in one channel does not require a manual replanning of the whole warehouse.

Available-to-promise quantities are published from the live system record rather than from a periodic export, which closes the gap between what a marketplace shows and what the building can actually ship today.

Transport moved onto TMS Cloud in the same programme so that the outbound handoff is continuous rather than a rekeying step. Orders are consolidated into loads, carriers are selected against service and cost rules, drivers carry a mobile task list, and electronic proof of delivery with photographs returns to the record at the moment of delivery rather than days later.

In-transit tracking and predicted arrival times feed both customer service and the project scheduling conversation. Exceptions — a missed time window, a failed delivery attempt, a temperature or handling flag — are raised while there is still time to act.

Because both warehouse and transport events land in one model, AI workers can operate on complete data. They reconcile carrier status messages, predict which orders are at risk of missing their promised window, propose wave and slotting adjustments ahead of a promotion, and prepare the exception summaries that supervisors previously assembled by hand each morning.

The team treats these agents as a capacity multiplier rather than a headcount replacement: they absorb the surge of routine monitoring that a promotional peak generates, which is precisely when human attention is scarcest.

Inventory record accuracy rose from the low nineties to 99.6 percent and, more importantly, stays there because perpetual cycle counting corrects drift continuously instead of quarterly. Order accuracy improved in step, since scan verification catches the substitution and quantity errors that used to reach the customer.

Picking productivity improved by roughly 45 percent on comparable volume, and peak-season temporary staff reached target productivity in hours rather than days. That combination is what removed the throughput ceiling: the warehouse can now absorb a festival peak by adding people, because adding people actually adds output.

Directed putaway and slotting raised usable storage density materially without new construction, and the reduction in re-handling cut in-warehouse damage for bulky appliances. Both effects flow directly to cost per order line, which fell even as the channel mix shifted toward smaller, more labour-intensive direct-to-consumer orders.

On-time delivery across the two-country network improved to the high nineties, with electronic proof of delivery available in real time rather than after a paper return cycle. Customer service resolves a delivery enquiry from the record instead of from a phone call to a dispatcher, and the project channel plans site deliveries against predicted arrival windows rather than conservative buffers.

Home appliance delivery carries an additional service layer that most distribution networks do not: installation, packaging removal and old-unit takeback. Capturing those tasks as structured steps in the driver's task list, rather than as verbal instructions, removed a persistent source of failed visits where the crew arrived without the tools or the mandate to complete the job.

The operational metrics translate into commercial behaviour the business can act on. Trustworthy availability lets the marketplace team list deeper stock without hedging, which directly increases sellable assortment. Reliable delivery windows reduce the rescheduling and redelivery costs that are largely invisible in a transport budget but very visible in the cost of serving a direct-to-consumer order.

Peak seasons are now planned as a capacity exercise rather than a risk event. Because the productivity of an added worker is predictable, the regional team can model a festival week in advance and staff it deliberately, instead of committing to overtime and hoping the building absorbs the volume.