A Vietnamese plant that assembles imported inputs has a supply chain whose longest and least predictable leg sits north of Lang Son, and the tariff treatment of its output depends on paperwork proving that leg is not the whole story. Customs settlement reporting ties consumption norms in the bill of material to declared exports, which makes bill of material accuracy a customs exposure rather than an engineering preference. Tet is a labour event with a return rate, the northern grid has failed in dry Junes, and the 2025 province merger rewrote the territory keys in your customer master. Time the Chinese input leg separately before you touch anything else.
The plan says the plant in Bac Ninh runs at eighty five per cent of nameplate through the quarter. It has not done that in eighteen months, and the shortfalls do not line up with anything in the demand file. Two of them start on a Tuesday in June. One runs for nine days after the lunar new year and then recovers to about ninety per cent of the pre-holiday rate and stays there for a month. Another follows a week when trucks stopped arriving from Lang Son. None of these is an efficiency problem, and none of them is visible in a capacity model that carries a single availability percentage.
Vietnam has attracted a great deal of assembly on the strength of labour cost and trade access, and the planning problems that follow are mostly about the things that were assumed to be constant.
The bill of material crosses a border you do not control
Most Vietnamese assembly runs on imported inputs. Electronics take components from China, Korea and Japan. Garment plants take fabric, mostly from China. The finished good is Vietnamese while much of its content is imported, and the inbound leg that matters runs overland through Lang Son at Huu Nghi or Tan Thanh, through Lao Cai, or through Mong Cai in Quang Ninh, with sea and air alternatives that cost differently and behave differently.
Those crossings queue. The closures during 2021 and 2022 produced truck lines measured in kilometres and holding times measured in weeks, and while that was extreme, the mechanism remains: limited lanes, inspection regimes on one side, and no alternative within the same lead time.
The planning move is unglamorous. Split the inbound lead time into an origin leg, a crossing leg and a domestic leg, and fit variance to the crossing separately. A single average lead time from a Shenzhen supplier to a Bac Ninh line combines a stable manufacturing process with a queue that has a long right tail, and the combined average describes neither. Where a component has both a road and a sea route, they are two supply options with different distributions rather than one option with a mode flag, and they should sit in the model as two.
Air is the release valve, and it splits the parts list in a useful way. A component small enough to fly can be recovered in three days at a price, while one that cannot fly economically needs the inventory, and that split is often a better basis for buffer policy than value or volume.
Origin has become a planning input
Vietnam's trade access is the reason many of these plants exist, and access is conditional on rules that reach into the bill of material.
For apparel the specific rules bite hardest. CPTPP applies a yarn forward rule, meaning the yarn and the fabric have to originate within the agreement for the garment to qualify. The EU agreement, in force since August 2020, applies a fabric forward rule with a cumulation provision that allows Korean fabric. Vietnam's own fabric capacity has not kept pace with its garment capacity, so the sourcing decision for cloth determines whether the finished garment gets preference in a given market. That makes fabric sourcing a commercial decision with a tariff attached, and it belongs in the sourcing model with the tariff priced per unit.
The July 2025 framework announced between the United States and Vietnam added a second layer, with a headline rate on Vietnamese goods and a substantially higher rate on goods judged to be transhipped. The operative definition of transhipment was left to be specified, which is precisely the sort of open question a planner cannot resolve and has to prepare for. The preparation is evidentiary: know what proportion of the value of each finished good is transformed in Vietnam, know it by product family, and be able to show the working. Companies that can produce that figure on request will have an easier time than companies that have to reconstruct it.
Underneath both sits a domestic mechanism that many foreign operators underestimate. Vietnam's duty exemption for materials imported to produce exports is settled after the fact, through an annual settlement report filed with customs that reconciles imported materials, the consumption norms declared for each product, and what was actually exported. The norm is the bill of material as customs understands it. Where the engineering bill of material and the declared norm drift apart, and they drift for the ordinary reasons of yield changes, substitutions and scrap rate revisions, the difference reads as material imported duty free and not exported.
That makes bill of material governance a customs matter with a cash consequence. The control is dull and effective: reconcile declared norms to the current engineering bill and to actual yield every quarter rather than at year end, and treat a norm change as an engineering change with a customs approver. The general practice around engineering change control is QQ2's subject; what is specific here is who else has to sign. Sites holding export processing enterprise status sit outside the customs territory altogether, which turns movements to and from the domestic market into declared imports and exports.
Tet is a labour event with a return rate
Every planner working Vietnam knows the plant closes for the lunar new year. The part that gets modelled badly is what happens afterwards.
Tet moves between late January and mid February, which alone breaks any model working on calendar months, and the general method for handling a moving holiday is D6's. The specific Vietnamese feature is the labour tail. A large share of the workforce in the industrial provinces has migrated from elsewhere, and the holiday is when they go home. Some do not come back, or come back late, or come back to a different employer offering a bonus. The result is a ramp rather than a step: capacity returns over two to four weeks at a rate that depends on how the previous year's bonus compared with the neighbouring park's.
Model this as a capacity profile rather than a shutdown window. Hold a daily availability curve for the six weeks around Tet, fitted from your own attendance data rather than from a holiday calendar, and let the supply plan see the ramp. Pre-build then gets sized against the area under that curve instead of against the closure days, which usually makes it larger than the first estimate, and the recruitment plan becomes a supply input worth putting in the sales and operations planning pack. The rest of the calendar has fixed points worth carrying: Hung Kings' commemoration on a lunar date in spring, and the pairing of Reunification Day on 30 April with Labour Day on 1 May, regularly bridged into a longer break.
The grid and the weather are capacity constraints
In May and June 2023 the northern provinces lost power on a rota. Drought had cut hydro output, demand was at a summer peak, and industrial parks in Bac Ninh, Bac Giang and Thai Nguyen took scheduled curtailment. In September 2024 Typhoon Yagi came ashore in the north and flooded and cut power to sites around Hai Phong and Quang Ninh.
Two different mechanisms, one planning implication: northern Vietnam has a correlated capacity risk that hits many suppliers at once. A dual sourcing strategy that puts both suppliers in Bac Giang is single sourcing with extra paperwork, and the correlation is invisible in a supplier risk model keyed on company rather than location. Carry the site coordinates and score the risk geographically; the general treatment of correlated failure points is HH5's.
The regulatory direction is worth tracking because it changes what a site can do about it. The direct power purchase agreement framework introduced by decree in July 2024 opened a route for large consumers to contract renewable generation directly, and the current power development plan is explicit about additions in the north. Neither removes the exposure within your planning horizon, so behind the grid question sits an ordinary one: how many hours of on site generation do you hold, and what does that cost against the margin of the hours it protects.
Typhoon exposure has a calendar. The northern season runs roughly July to October, and the useful response is to shape finished goods cover ahead of it for products whose customers hold no buffer of their own.
Getting product out, and the two ports that are not interchangeable
In the south, Cat Lai in Ho Chi Minh City handles the bulk of container volume and sits on a river with an approach road that congests. The deepwater complex at Cai Mep Thi Vai, roughly eighty kilometres downstream, takes the largest vessels and offers direct sailings to North America that Cat Lai cannot. The choice is a real trade: a longer road move or a barge leg against a shorter ocean leg with fewer transhipments.
That trade is worth computing rather than assuming, and the input most people are missing is the distribution of the road leg to Cai Mep by time of day. A move that takes ninety minutes at two in the morning and four hours at five in the afternoon is not one lead time, and cut off times at the terminal turn the difference into a missed sailing.
In the north, Lach Huyen gave Hai Phong deepwater capability from 2018 and shortened the northern export path considerably. The domestic link between the two halves of the country has not moved. The Reunification railway is a single track metre gauge line of about 1,726 kilometres carrying very little containerised freight relative to the road, and the north to south high speed line approved by the National Assembly in November 2024 is a decade away from carrying anything. For planning purposes Vietnam is two supply regions joined by Highway 1 and coastal shipping, and a national distribution centre in the middle is usually the wrong answer.
The domestic map changed underneath the customer master
Vietnam merged its provinces in 2025, going from sixty three to thirty four with effect from July. Sales territories, distributor coverage definitions, customer addresses and any forecasting hierarchy keyed on province all inherited a change they did not ask for.
The practical damage is to history. A hierarchy that reconciles a national forecast down to provinces has to be restated on the new structure to keep the history usable, and a naive mapping loses the granularity where two former provinces merged. Restate the history on the new hierarchy and keep the old codes as an attribute rather than a level. The reconciliation method that keeps the levels consistent afterwards is D8's.
Distribution has grown a fast modern trade in Hanoi and Ho Chi Minh City while the provinces still run through wholesalers and small shops served by motorbike, with distributors holding both the stock and the credit relationship. Sell in to the distributor is not demand, and that gap is where a Vietnamese forecast usually goes wrong; the treatment of it is D2's, and regional route to market design belongs to U6.
Where this stops
The origin work above rests on data you may not have. Supplier declarations covering the tier below your direct supplier are hard to obtain and harder to verify, and a Vietnamese assembler buying a subassembly from a Chinese vendor frequently cannot see what is inside it. Any regional value figure built on unverified supplier statements is only as good as those statements, and a customs authority will treat it that way.
The Tet capacity curve has a data problem too. Fitting it needs several years of daily attendance or output at the same site, and a plant that opened two years ago has two observations of something that varies with the labour market. Use the number, hold it loosely, and widen the pre-build until you have four years. Where a rule itself is undefined, as with the transhipment provision, scenario branches are the honest tool and a point estimate is a guess wearing a decimal place.
Take the last two years of inbound receipts from Chinese suppliers, separate the crossing from the rest of the lead time using the customs clearance timestamp, and compute the standard deviation of that leg alone; it is usually the single largest term in the buffer calculation and almost never the one being tuned.
More on this in regional on the DemandLabs platform.
