A peak is not the same thing as a surprise. Most of the days that break a transport plan were visible weeks or months earlier, in data the operation already had. They break the plan anyway, because nobody converted them into a number of trucks in advance.
There are two kinds, and they need opposite treatment. One is written into the calendar and cannot be argued with. The other was built by your own company, usually by a sales target, and can be dismantled. Telling them apart is most of the work.
The calendar peak is measurable to within half a point
Thailand's Office of Industrial Economics publishes its monthly industrial indices twice: once raw, and once seasonally adjusted. Divide the raw series by the adjusted one and you get that office's own estimate of what the calendar alone does to a month. Averaged over the five complete years from B.E. 2564 to B.E. 2568, the national production index gives this.
| Month | Calendar effect | Range across the five years |
|---|---|---|
| January | 103.7 | 103.4 to 103.7 |
| February | 101.4 | 100.9 to 101.8 |
| March | 110.1 | 109.7 to 110.3 |
| April | 92.5 | 91.9 to 93.4 |
| May | 101.9 | 101.2 to 102.6 |
| June | 99.3 | 99.1 to 99.5 |
| July | 97.2 | 96.8 to 97.8 |
| August | 98.7 | 98.3 to 99.2 |
| September | 98.0 | 97.4 to 98.7 |
| October | 98.9 | 98.5 to 99.3 |
| November | 100.0 | 97.8 to 100.8 |
| December | 97.4 | 96.9 to 98.2 |
March runs about 10 per cent above trend and April about 7.5 per cent below. From one to the other is a fall of 16 per cent in the size of the month.
The ranges are the important column. March has landed between 109.7 and 110.3 in every one of five years. April between 91.9 and 93.4. This is not a forecast that might be wrong by a third. It is a repeating feature of the Thai calendar, stable to within about half a point, and it was knowable last year.
The shipment index says the same thing in raw form. Normalise each year to its own twelve-month average and March is the highest month of the year in all five complete years, and April the lowest in four of them.
| Year | March | April |
|---|---|---|
| B.E. 2564 | 114.88 | 94.67 |
| B.E. 2565 | 114.41 | 96.43 |
| B.E. 2566 | 112.57 | 92.37 |
| B.E. 2567 | 107.88 | 94.07 |
| B.E. 2568 | 108.32 | 93.32 |
| B.E. 2569 | 110.99 | 93.87 |
The month gets smaller. The day does not
Here is where a monthly figure will mislead anyone booking trucks with it.
Take B.E. 2568. Factory shipping fell from 108.32 in March to 93.32 in April, down 13.8 per cent. Read as a month, April looks quiet, and a plan built on monthly volume will book fewer vehicles for it.
Now count the working days. March 2025 carried no public holiday at all, and had 21 of them. April carried four: Chakri Day fell on a Sunday and moved to Monday 7 April, and the three Songkran days on 13, 14 and 15 April brought a substitute day on Wednesday 16 April. April had 18 working days.
Divide the index by the days. March is 5.16 a day, April 5.18. The two months are within half a per cent of each other per working day. April's month shrank by 13.8 per cent because it lost three working days out of twenty-one, not because the work went away.
That has a direct operational consequence. A truck is hired by the day, not by the month. An operation that scales its April vehicle plan down by 14 per cent because the monthly forecast fell by 14 per cent will be short every single day it runs, and will spend the shortfall buying capacity at short notice in the one week of the year when everyone else is doing the same thing.
The reverse trap sits in March. A month that is 10 per cent above trend with a full complement of working days is a month where the ordinary day is genuinely busier, not just longer. That one really does need more vehicles.
Doing this with your own numbers
The national figures show the method and the scale. They are an average across 22 sectors, so they are not your seasonality. Yours is in your own despatch file and takes an afternoon to extract.
- Take three or more years of despatches, monthly, in vehicle-days rather than tonnes or baht. The vehicle-day is what gets bought.
- Divide each month by that year's own twelve-month average. Dividing by the year's own average is what removes growth, so the shape left behind is seasonal rather than trend.
- Average each month's figure across the years, and write down the range as well as the average.
- Divide each month by its working days to get the daily figure, which is the one that becomes a truck booking.
The range decides what you do with it. A month whose figure sits in a tight band across three years is a seasonal allowance and belongs in next year's committed plan. A month that swings widely is not seasonal at all; it is random variation, and the answer to random variation is buffer stock and short-term hire, not a bigger standing fleet. Treating the second like the first is how operations end up owning capacity they use twice a year.
The peak inside the month
Now do the same exercise on the days of the month rather than the months of the year. Plot every despatch of the last six months against the day it left the yard. Almost every distribution operation gets the same shape: a flat, quiet run through the first three weeks and a wall at the end.
The best measurement of it in the literature comes from a Brazilian branch of a large multinational in non-durable consumer goods, studied with system dynamics and interviews at 26 of its customers. In 2007, 57 per cent of its monthly sales volume was sold and invoiced in the last five days of the month, on average, across every channel and every region. It had run that way for more than twenty years.
Two findings from that study matter more than the number.
The first is what the company believed. Its executives held that the spike was inherent to the industry and came from outside the company, so no other policy was available. The model said the opposite. The spike was generated inside the company, by the gap between the profit target and actual profit, which pushed sellers to offer discounts at the end of each period. Customers learned the discount was coming and waited for it, which made the gap worse the following month.
The second is what it cost. Shipping, stocking, warehousing, handling and overtime were all directly affected, but with a delay between the spike and the cost showing up. That delay is why nobody connects them. Over the mid to long term the invoiced value did not rise at all, while volume and total cost did, so total profit fell. The company was buying its own sales twice.
Why the Thai month-end is welded to the truck
In some countries a company under revenue pressure can move the paper without moving the goods. In Thailand it cannot, and the reason is worth knowing precisely, because it also tells you which dates are free to move.
Three sections of the Revenue Code lock the invoice to the delivery.
- Section 78(1). For a sale of goods, the whole VAT liability arises when the goods are delivered. It arises earlier only if ownership transfers, the price is received, or a tax invoice is issued before delivery, and then only in proportion to that act.
- Section 86. A registered operator must make a tax invoice for every sale, and must make it immediately when that liability arises, then hand it to the buyer.
- Section 83. The return is filed by tax month, whether or not anything was sold in the month, and the return and the payment for a tax month are due within the fifteenth day of the following month.
Read together: the delivery date is the revenue date. A sales target measured on invoiced value is therefore a target measured on trucks physically leaving the yard before midnight on the last day of the month. The Thai month-end rush is not a bad habit. It is arithmetic, and no amount of asking the sales team nicely will change it while the target is written that way.
The same three sections show the way out. Everything that is not the delivery date is free to move. The date the customer places the order, the date you pick and stage the load, the date the customer expects to receive it inside their own agreed terms. Flattening a month-end means pulling orders forward, not pushing invoices back, and that is a commercial change rather than a tax one.
What the peak costs, in trucks
Take a month with 22 working days and 440 vehicle-days of despatch work in it. Assume a committed vehicle-day costs 8,000 baht and a day bought on short notice costs 10,000, a 25 per cent premium. Substitute your own two numbers; the wider the gap between them, the more everything below costs.
Spread evenly, that month needs 20 vehicles a day. Commit 20 and you spend 22 × 20 × 8,000, or 3,520,000 baht, with nothing bought outside and nothing idle.
Now put 40 per cent of the work into the last four days, which is a milder version of the Brazilian case. The last four days need 44 vehicles a day. The other eighteen need 14.7.
| Committed trucks | Committed vehicle-days paid | Idle days | Days bought outside | Total cost |
|---|---|---|---|---|
| 14 | 308 | 0 | 132 | 3,784,000 |
| 15 | 330 | 6 | 116 | 3,800,000 |
| 18 | 396 | 60 | 104 | 4,208,000 |
| 20 | 440 | 96 | 96 | 4,480,000 |
| 25 | 550 | 186 | 76 | 5,160,000 |
The cheapest column in a peaked month is 14 committed trucks at 3,784,000 baht. The same 440 vehicle-days of work, spread evenly, cost 3,520,000. The shape of the month, not the amount of work, costs 264,000 baht a month, or about 7 per cent of the transport bill. Across a year that is 3,168,000 baht spent on nothing but the order in which the work arrived.
Flattening it only part of the way is worth most of that. Move from 40 per cent in the last four days to 25 per cent, and the best commitment becomes 18 trucks at 3,608,000 baht. That single change is worth 176,000 baht a month.
Why the average is the worst possible commitment
Look again at the row for 20 committed trucks, the average of the month. It leaves 96 vehicle-days idle and buys 96 vehicle-days outside.
Those two numbers being identical is not a coincidence in this example. It is always true. Commit exactly at the average and the days you paid for and did not use must, by arithmetic, equal the days you needed and did not have. You pay twice for the same capacity and end up with the worst cost on the table.
This is the point where a peaked month behaves in a way that surprises people. The peakier the month, the fewer trucks you should commit to, not more. In the flat month the right commitment is 20. In the peaked month with the identical total workload it is 14. A peak day cannot be covered by standing capacity at a price that makes sense, so the standing fleet retreats to what the ordinary day needs and the peak is bought. What the arithmetic will not let you do is buy the peak cheaply. That is a separate negotiation, and it is why how much capacity to commit to is decided by the shape of the week, not the size of the year.
The four levers, in the order of how hard they are
Move the target measurement. The Brazilian study's central finding is that the target period creates the spike. A sales target measured on the month's invoiced value produces a month-end delivery wall in Thailand because of Section 78. A target measured on orders taken, or measured on a rolling basis, does not. This is the only lever that removes the cause; the rest manage the symptom. It also has a cost the study measured: profit falls in the short term before it rises, because one period's apparent sales are given up in the change. Nobody signs that off inside the logistics function, which is why the study concludes the conversation has to be held in financial terms with whoever owns the annual result.
Publish shipping days by customer. Give each customer two or three named days a week on which their orders ship. This spreads the month by design and costs nothing, and it is the same mechanism that makes an order cut-off worth more than it looks.
Price the last four days. A delivery on the 29th costs more to serve than the same delivery on the 12th. Most carriers charge a flat rate through the month, so that difference currently sits entirely with you. A small discount for taking delivery in the first three weeks is easier to sell internally than a surcharge at the end, and it moves the same orders.
Stop paying the end-of-month discount. This is the direct one and the hardest. The discount is what trains the customer to wait, and each round makes the next month's gap larger.
The peak your customer made, arriving as yours
One last check before you conclude that a peak is yours to fix. A small, steady change in what shoppers actually buy gets amplified as it passes back up the chain, because every level between them and you reorders independently and in its own batch sizes. By the time it reaches a factory gate, a modest movement at the shelf can look like a violent one.
Practically, this means part of your month-end is not yours. It is your customers' month-ends stacked on top of each other and landing on the same four days. You cannot dismantle that from inside your own company, but you can see it coming, because it repeats. Plot despatches by day of month for each large customer separately. Customers whose shape differs from the pack are worth serving on their own rhythm; customers who all pile onto the same three days are a reason to buy short-notice capacity deliberately rather than by emergency.
Where this leaves next month
The output of all of this is small and specific. For each of the next twelve months you should be able to write down two numbers: the vehicle-days that month needs, and the vehicle-days its busiest day needs. Everything above exists to produce that pair honestly.
If the two are close, the operation is easy to resource and the commitment sits near the average. If the second is two or three times the first, the peak is the expensive thing in the plan, and the first place to look is not the carrier's rate card. It is whichever internal rule decided that all the work should arrive in the same four days.