Every transport desk knows the complaint. A lane that runs in three and a half hours most weeks suddenly takes most of a day, the customer asks what happened, and the honest answer is "traffic". Nobody can do anything with that answer, so nothing changes, and it happens again.
This article is about replacing "traffic" with something you can act on. The short version: the road itself rarely varies by more than about half. When a trip takes twice as long, or five times as long, it is almost always because a modest delay crossed a line in the calendar, and a line in the calendar is something you can plan around.
Two different things are hiding in "it took longer"
When a trip runs long, two separate mechanisms are usually mixed together.
- Spread. The minutes the road adds or takes away. Rain, a crash, a Friday afternoon, roadworks. This varies smoothly: a bad day is 20% or 40% worse than a normal one, not 400%.
- Steps. The hours a fixed line in the calendar adds when a truck arrives on the wrong side of it. A dock that stops accepting trucks at 17:00, an expressway ban window, a driver who has reached the end of a legal driving block. Miss one by five minutes and you lose a whole block of time.
Spread is the road's fault and you can only buffer against it. Steps are your plan's fault and you can usually remove them. Most "the trip doubled" stories are a small amount of spread multiplied by a step.
First, measure the lane properly
You cannot tell spread from steps with an average, because the average hides both. You need each run of the lane as a separate record, and you probably already have it: trucks covered by the journey data recorder mandate carry one, and its position records give gate-out and gate-in times per trip. What the recorder measures and what it only infers is covered in the article on what truck GPS tracking tells you.
Take eight to twelve weeks of one lane. For each run, write down three numbers: departure, arrival at the customer's gate, and the time unloading actually started. Then work out:
- The median road time, the middle run. That is your "normal".
- The 95th percentile road time, the run that only one in twenty exceeds. With twenty runs, that is the second slowest.
- The buffer index: the 95th percentile minus the average, divided by the average. It says how much time you must add to the average to be on time on nineteen runs out of twenty.
Here is an illustration, twenty runs of the same lane, departing 12:30, customer's dock accepting trucks until 17:00. The road times are in minutes, gate to gate:
| Runs | Road time (minutes) |
|---|---|
| Fifteen ordinary runs | 195, 200, 205, 205, 210, 210, 212, 215, 215, 220, 220, 225, 230, 235, 240 |
| Three slow runs | 255, 262, 268 |
| Two very slow runs | 285, 310 |
The average is 231 minutes. The median is 220. The 95th percentile, the second slowest, is 285. The buffer index is (285 − 231) ÷ 231, or 23%. The very worst run, 310 minutes, is 41% slower than the median.
So on the road, this lane never came close to doubling. Now look at what the customer saw.
Departing at 12:30, anything slower than 270 minutes arrives after 17:00. Two runs did. Those two trucks stood outside overnight and started unloading at 08:00 the next morning, 19 hours 30 minutes after they left. That is 5.3 times the median.
A worst case 41% slower on the road became 5.3 times slower at the customer, and the whole difference came from one fixed line. If the lane left at 11:45 instead, the cut-off would move to 315 minutes, and all twenty runs would have been unloaded the same day. Forty-five minutes of earlier departure would have removed both overnight delays.
Sort the slow runs by cause
Once the slow runs are separated, give each one a reason. In practice they fall into four families, and each family needs a different answer.
The clock: slow at predictable times
Some slowness repeats by day, hour and month. The Department of Highways' record of accidents on its own network in 2025 is a useful proxy for when the road is under most strain, because every recorded crash is a potential blockage. It is not a measure of delay, and should not be read as one, but its shape is clear:
| Pattern | What the 2025 record shows |
|---|---|
| Day of week | Saturday 3,518 and Friday 3,433, against Thursday 2,988. The busiest day is 18% above the quietest |
| Time of day | 13:00 to 17:00 carries 4,936 accidents, 22% of the year's 22,609, in 17% of the day |
| Month | April 2,720, December 2,493 and January 2,378, against 1,480 in September. April is 84% above the quietest month |
The months with the most crashes are the long-holiday months, when the whole country is on the road at once, and those need planning of their own. For the rest of the year the point is simpler: a Friday afternoon departure and a Tuesday morning departure are two different lanes, and should have two different planning times.
Weather: slow in a way you can see coming by the hour
Rain in Bangkok reduces speed in two ways: the rain falling while you drive, and the water still lying on the road from the rain before. A study of about 4,800 probe vehicles across Bangkok from 2018 to 2020 put numbers on both:
- 10 mm an hour of rain while driving cut speed in an average district by about 10.3%.
- The same rain after a further 10 mm over the previous five hours cut it by about 20.7%.
- The effect was larger in the morning and evening peaks, in low-lying districts, and in districts with many narrow side streets.
Turn a speed cut into time and it grows: a 20.7% lower speed means 26% more time. A 60-minute urban leg becomes about 76 minutes.
Two practical conclusions follow. The buffer belongs to the urban ends, not the line haul, and most of all to ends in low-lying areas reached through small streets. And the rain that matters most is the rain that fell in the hours before the truck left, which is something you can check at dispatch.
On the national highways, the Department of Highways recorded 2,837 of 2025's accidents in rain, 12.5% of the total. Rain is a real contributor on the trunk roads too, but it is not the dominant one.
Incidents: slow at random, but not everywhere
A crash is the classic random delay. There were 22,609 recorded on the Department of Highways' network in 2025, about 62 a day, on a network of about 52,400 km. That is roughly 0.43 per kilometre per year on average.
The average is misleading, because incidents concentrate. The Department lists every control section with more than 30 accidents in the year, ranked per kilometre, and the sections near the top are exactly the ones freight uses to get in and out of Bangkok and the Eastern Seaboard:
| Section | Length | Accidents in 2025 | Per km |
|---|---|---|---|
| Motorway 7, Khlong Song Ton Nun to Pim Pa | 31.0 km | 591 | 19.06 |
| Motorway 9, Ram Intra to Bang Phli | 19.7 km | 313 | 15.89 |
| Motorway 7, Nong Kham interchange to Laem Chabang port | 8.1 km | 109 | 13.47 |
| Motorway 9, Bang Pa-in to Ram Intra | 44.3 km | 540 | 12.19 |
The first of those is 44 times the network average per kilometre, and it averages more than one and a half recorded accidents every day.
This changes how you treat incidents. On a long rural stretch, a crash is rare and you can only buffer against it. On one of these sections, a crash on your route on any given day is closer to normal than unusual, so the plan needs an agreed alternative before it is needed, not a call from the driver after the queue has formed. Deciding what to do in the moment, and who is allowed to decide, is covered in the article on the in-day control loop.
The ends: slow at the gate, not on the road
Look at the three numbers you wrote down per run: departure, arrival at the gate, unloading start. If the slow runs had normal road times but a long gap between arrival and unloading, the problem is not the road at all. It is the customer's dock, or yours, and no amount of road buffer fixes it.
This is the most common finding when a lane is actually measured, and the cheapest one to act on.
The steps that turn a delay into a day
With the causes sorted, go back to the steps. These are the lines that multiply spread:
- The receiving window. The illustration above. Arrive after the last acceptance time and the delay is no longer the minutes you were late, it is the hours until the dock opens.
- The expressway ban windows. A truck that was meant to clear the city before a ban starts, and does not, either waits for the window to end or takes a slower road.
- The driver's continuous driving limit. Section 103 bis of the Land Transport Act, read in Thai, bars a licensed driver from driving for more than four consecutive hours; after a continuous rest of at least half an hour, a further period of not more than four hours may follow. A block planned at 3 hours 45 minutes that picks up 20 minutes of rain now needs a half-hour stop before arrival. Twenty minutes of spread has become fifty.
- The start of the next working day. A trip that cannot finish today does not finish at midnight. It finishes when the driver can lawfully start again, and at the far end's opening time.
The rule for all four is the same: know where each line sits on each lane, and keep your 95th percentile arrival on the right side of it. How the lines combine over a full door-to-door journey is worked through in the article on how long inland transport really takes.
Put the slack where the cause is
A flat percentage added to every lane is the most common response to a bad week, and the most expensive. It pays for buffer on the good lanes and the good days to cover a problem that lives on a few sections, a few hours and a few gates. Match the answer to the family instead:
| Cause | What to do |
|---|---|
| The clock | Set separate planning times by departure band and by day. Friday afternoon gets its own number |
| Weather | A rainy-season planning time for the urban ends, and a check at dispatch of how much rain has fallen in the last few hours |
| Incidents | On the high-incident sections, an alternative route agreed in advance, and a rule for when the driver switches |
| The ends | Measure arrival-to-unloading separately. If it is the slow part, fix the dock, not the road plan |
| Steps | Move the departure, or move the line: ask for a later acceptance time, or plan the rest stop where it costs nothing |
What to do this month
- Pick your three most complained-about lanes and pull eight to twelve weeks of recorder data for each: departure, gate arrival, unloading start.
- Calculate the median, the 95th percentile and the buffer index for the road time, and separately for arrival-to-unloading. If you have fewer than twenty runs, use the second slowest as your 95th percentile and treat it as provisional.
- Mark every fixed line on each lane: the far dock's last acceptance time, any ban window on the route, and where the driver's four-hour block ends.
- Check whether the 95th percentile run clears every line. If it does not, move the departure before you add buffer. In the illustration, 45 minutes earlier removed every overnight delay.
- Give every slow run a reason code: clock, weather, incident, gate or step. Read them monthly. A reason that keeps recurring is not bad luck, it is a plan that needs changing.
- Report the 95th percentile to your customer, not the average. The average is the number that will be wrong on the weeks they remember.
"Traffic" is not an explanation. It is five different causes, and every one of them has a different fix.