Two people in the same transport office describe the same operation in opposite words.
The planner says no two days are the same, so the loads have to be built fresh every afternoon. The owner says the customers barely change, so why is somebody redrawing the map every day.
Both are talking about the same order file, and neither of them has counted it. That count is the whole decision, and it takes an afternoon.
The choice is a property of your orders, not a preference
There are two questions, and only two.
Do the same delivery points come back? Not the same customers on your sales report. The same physical addresses receiving goods, week after week.
Can you know the quantity before the day? Not exactly. Well enough to pick a vehicle.
Cross them and you get four situations, and each one has an answer that is not really up for debate.
| Quantity predictable | Quantity unpredictable | |
|---|---|---|
| Same points every week | Fixed routes, held for months | Fixed routes, variable vehicle |
| Points change | Plan every day | Plan every day, and it is hard |
Most operations put themselves in the wrong box, and they do it the same way every time: they see their volumes bouncing around, conclude that nothing can be fixed, and start planning daily.
That is the mistake. It is the addresses that decide, not the volumes. A route is a sequence of places. If the places repeat, the route can be fixed, and the fact that Tuesday's load is 40% bigger than last Tuesday's changes which truck you send, not where it goes. Retail delivery, bread, beer and factory collection rounds have all run on fixed schedules for decades while their daily quantities moved constantly.
Schedules built this way, from history rather than from today's orders, hold for three to six months at a time. They need editing when a customer opens or closes, not redrawing.
The genuinely different case is the one where the points move. A parcels operation, a spare parts service, a contract carrier working for a dozen clients: today's list of addresses is not last week's list, so there is nothing to fix. That operation has to plan daily, and it should stop wishing otherwise.
The test, on eight weeks of your own delivery notes
You do not need software for this. You need the last eight weeks of delivery notes and an afternoon.
Count the repeat rate. Take this week's delivery addresses. For each one, count how many of the previous eight weeks it also appeared in. Then read the whole list:
- Addresses appearing in six weeks or more are your spine. These can be fixed.
- Addresses appearing once or twice are your tail. These get planned.
If the spine is most of your drops, you are in a fixed-route operation that has not been organised as one. If the tail is most of them, you are planning daily and you are right to.
Then check the day, not just the week. Sort the spine by which weekday each address received on. Two operations with the same repeat rate behave completely differently depending on whether Tuesday means anything. If a customer has been taking deliveries on a random weekday for two years, nobody has ever asked them to pick one, and asking is free.
Then size the vehicle against the peak. For each candidate route, list the weekly volume for those eight weeks. Say one eastern round comes out at 18.2, 21.6, 16.9, 24.8, 19.4, 22.1, 17.5 and 26.3 cubic metres.
The average is 20.85. The peak is 26.3. A six-wheel box body of roughly 29.5 cubic metres holds the peak, so a fixed six-wheeler run works every week of the eight.
Now price the honesty. In an average week that vehicle runs at 20.85 out of 29.5, which is 70.7% full. Measured against its own peak, 1 β 20.85 Γ· 26.3, about 21% of the vehicle is being bought for weeks you mostly do not have.
That number is the real cost of the fixed route, and it is the number to compare against everything in the next two sections. If it comes out near 20%, fix the route. If the peak is more than twice the average, you are buying a truck for a fortnight a year, and that route belongs in the tail.
What a fixed route actually saves, and it is not kilometres
This is where most of the argument goes wrong, because both sides assume the prize is distance.
A vehicle collection operation feeding an assembly plant in Laem Chabang Industrial Estate was studied on five months of its own records. It collected from 35 supplier plants, 12 in Chonburi and 23 in Rayong, averaging 941.12 cubic metres a day into a plant that receives between 08:00 and 19:00.
As it was running, it used 17 vehicles doing 49 rounds a day, covering 3,118.5 kilometres, with each round carrying an average of 17.99 cubic metres, or 57.2% of what the vehicle could hold. That cost 21,859,083 baht a year.
Rebuilt properly, it used 10 vehicles doing 37 rounds a day, covering 2,310.5 kilometres, at 25.43 cubic metres a round and 80.9% full. That cost 11,962,011 baht a year, a saving of 9,897,072.
Now look at what moved and what did not.
| Before | After | Change | |
|---|---|---|---|
| Vehicles | 17 | 10 | β41.2% |
| Rounds a day | 49 | 37 | β24.5% |
| Distance a day | 3,118.5 km | 2,310.5 km | β25.9% |
| Distance per round | 63.6 km | 62.4 km | almost none |
| Cost a year | 21,859,083 | 11,962,011 | β45.3% |
Cost tracked the vehicle count, not the distance. Vehicles fell 41.2% and cost fell 45.3%. Distance fell 25.9%, and distance per round barely moved at all, because the roads between Rayong and Laem Chabang did not change. Nothing was found by driving cleverer. The saving was twelve fewer rounds and seven fewer trucks.
That is not a quirk of one study. Look at what a round costs before it moves. In the collection operation described in the next section, the standing cost of putting a six-wheeler on the road for one round is 2,505.25 baht, and it burns 6.45 baht a kilometre. At those numbers a round has to cover 388 kilometres before its running cost equals what it cost just to exist. A four-wheel round stands at 809.78 baht, a long six-wheeler at 3,072.90, a ten-wheeler at 3,422.04.
So the question to ask a routing change is not how many kilometres it removes. It is how many rounds and how many vehicle-days it removes. If the answer is none, there is no saving in it, however much neater the map looks.
This is also the honest answer to what a fixed route buys. It removes rounds, because a repeating route can be filled to a known level and sized once, instead of being assembled each day out of whatever happens to be sitting in the system by four o'clock.
Repetition then buys a second thing that never appears in a routing model. Unloading time at a customer has a fixed part, ten to twenty-five minutes depending on access and equipment, plus a variable part that depends on how much is on the truck. Experience shrinks the fixed part and nothing else does. A driver on his fortieth visit knows which gate, which security desk, which forklift and which person signs. A driver on his first is standing at the wrong entrance. The Kasetsart study allowed a flat 30 minutes per collection point, and on a route with eight stops that is four hours of the day spent standing still.
What a fixed route costs: the rule nobody revisits
Here is the failure that is worth the price of the whole article, because it looks like nothing.
A logistics company runs collection rounds from automotive component plants to assembly plants across five zones of Thailand. Its busiest zone has 69 supplier plants feeding four assembly plants. It receives volumes quarterly, divides them by working days, and builds a plan every day. So it is already planning daily, with real data, using people who know the job.
It planned inside two rules that had been there so long nobody argued with them:
- Pickups are sorted into four time windows, 07:00 to 12:59, 13:00 to 18:59, 19:00 to 23:59 and 00:00 to 06:59, and a route never crosses one.
- A route serves one assembly plant. Two pickups going to different plants cannot share a truck.
Both rules are reasonable. Both are the kind of thing a well-run operation writes down once and never looks at again. And the result was that vehicle capacity utilisation sat below the company's own 80% target in every single month across a full year, averaging 58.07%. Trucks were leaving 42% empty, every month, while a competent team planned them daily.
The study changed the rules, not the planning. Merge the four windows into two, and let one route serve two plants that are near each other. Same demand, same fleet, same algorithm.
| What was allowed to share a truck | Cost | Utilisation |
|---|---|---|
| Existing practice | 594,502.15 | 58.07% |
| Same plant, same window (four windows) | β45.75% | 79.83% |
| Same plant, two merged windows | β49.64% | 72.40% |
| Nearby plants, two merged windows | β58.52% | 92.60% |
Baht, averaged across four quarters. Cost fell from 594,502.15 to 246,626.57.
Read the first row of results carefully, because it is the quietest and the most useful. Simply planning the existing rules well, with no rule changed at all, was worth 45.75%. Then loosening the buckets took it to 58.52%.
The expensive part of a fixed route is not the route. It is the rule that decides which jobs are allowed onto it. Every bucket, every window, every "we keep those separate", every minimum notice, every one-customer-per-truck convention is a wall, and each wall leaves a piece of a truck empty. Nobody costs the walls, because each one was sensible on the day it was written.
Two honest caveats. This is a modelled comparison, not a saving somebody banked. And the authors say plainly that getting it in practice needs the parts plants to stage and label goods by destination, which is work at somebody else's site. A number this large is a direction to investigate, not a cheque.
What planning every day costs
Daily planning is not free, and its price is in three places.
The plan itself. Historical data is enough to build a fixed schedule once. Live scheduling needs the day's real demand, complete and on time, every day, and somebody working it before the pick can start. That is the reason not every operation schedules live: the data and the hours cost money whether or not the day turns out to be different from yesterday.
Everything repetition was giving you. A route that changes every day cannot build driver familiarity, cannot let a customer prepare, cannot be handed to a carrier as a standing job worth planning around, and cannot have a return load arranged in advance.
A queue somewhere else. This is the one that surprises people. In the Laem Chabang case, the rebuilt routes were correct and cheaper, and when the arrival times were plotted they bunched into 10:00 to 11:00, 12:00 to 13:00 and 17:00 to 18:00, with hours in between where nothing arrived at all. The plant had a fixed number of forklifts and receiving staff. Trucks queued at the peaks, near the production area, while the equipment stood idle in the gaps.
The fix was not a better algorithm. They kept the routes and moved the arrivals: push each vehicle's last round of the day into an empty slot, then push the rounds before it back to suit, staying inside the 19:00 close. Same trucks, same kilometres, spread arrivals, four receiving stations matched to the forklifts that existed.
A route plan that ignores the receiving resource does not remove a queue, it relocates one. Whatever the plan says, the gate can only take what the gate can take, and the hours a truck spends waiting there are rent on a driver's legal day.
The answer most Thai operations should reach
Almost nobody is at either extreme. The workable answer is to fix the frame and vary the load.
Fix the territory, the day and the driver. Vary the vehicle and the quantity.
In practice that means a customer is on the eastern round, on Tuesdays and Fridays, with the same driver. What changes week to week is which truck goes and what is on it. The customer experiences a fixed schedule. You keep the freedom that matters, which is vehicle choice, and give up the freedom that was never worth anything, which is the right to visit them on a different day for no reason.
Then the tail gets planned into it. A new address or an urgent order is tested against the fixed rounds first: is there room on Tuesday's truck, and does it still fit the day. If yes, it joins. If no, it goes on a separate vehicle and gets priced as one. That is a five-minute check, not a replan.
The habit worth keeping is to review the frame on a schedule, quarterly, rather than to keep the frame and re-plan the days. Fixed routes fail slowly. They keep running long after two customers have closed and a third has quadrupled, because nothing in a fixed route ever announces that it has gone stale.
Three Thai constraints that fix the frame before you do
The driving day. Under Section 103 bis of the Land Transport Act B.E. 2522, subject to the labour protection law, a licensed driver may not drive continuously for more than four hours from the moment of starting, and after a continuous break of at least half an hour may drive up to four more consecutive hours in the same twenty-four.
For a fixed route, this is the binding test, and it must be applied to the bad week. A round built to fit on a clear day is not a fixed route; it is a fixed disappointment that fails whenever it rains. Build it so it survives the slowest week in your eight, and it will still be there next quarter.
The Bangkok expressway windows. Six-wheelers are barred from the listed expressways between 06:00 and 09:00 and again between 16:00 and 20:00. Trucks of ten wheels and over, and trailers, are barred between 06:00 and 09:00 and between 15:00 and 21:00.
So in and around Bangkok the usable middle of the day is short and it is fixed by regulation, not by you. That has a useful consequence: the departure time is often the only thing you get to fix, and fixing it well is worth more than the route order. It also means a daily plan that does not carry these windows will keep producing rounds that cannot legally be run, and the driver will silently fix them on the road in ways nobody measures.
The receiving window at the other end. The Laem Chabang plant took deliveries between 08:00 and 19:00. Your customer's window, or your own dock's, is very often the real constraint, and it is shorter than the road allows. Collect the actual open hours for every address on the spine before designing anything. It is dull work and it changes the answer more than the routing method does.
You already have the data to build this
The awkward part of a fixed route is honest travel and stop times, and the usual source is a planner's memory, which is generous about traffic and stingy about unloading.
You do not have to rely on it. Vehicles registered for transport in Thailand carry a journey data recorder that logs position, speed and time, identifies the driver by card and transmits to the Department of Land Transport at least every five minutes, with the data retained for six months. Six months is more than the eight weeks the test above needs.
Ask your carrier, or your own fleet office, for the recorded arrival and departure times at each address for one quarter. Then build the route from what happened rather than from what everyone believes happens. The same records are what a carrier needs in order to quote the route accurately instead of pricing the unknown at worst case.
What to change on Monday
- Pull eight weeks of delivery notes and count, for each address, how many weeks it appeared in. Six or more is your spine and can be fixed. One or two is your tail and must be planned. The split decides the argument
- For each spine address, check which weekday it receives on. Any address with no consistent day is a free win: ask them to choose one
- For each candidate route, list the eight weekly volumes, take the average and the peak, and work out 1 minus average divided by peak. That percentage is what the fixed route costs you in vehicle you are not using
- Write down every rule that stops two jobs sharing a truck: time buckets, one customer per vehicle, one destination per route, separate product groups. Cost each one. That list, not the map, is where the money is
- Judge any proposed routing change by the rounds and vehicle-days it removes, not the kilometres. If no vehicle-day disappears, no money does either
- Plot your own gate arrivals by hour for one week. If they bunch, respread the plan against the forklifts and staff you actually have before buying anything
- Collect the real open hours of every address on the spine, and time a fixed route against your slowest week, not an average one, so it still fits four hours plus a break plus four hours when the traffic is bad
- Put the frame in the diary for a quarterly review. Fixed routes do not fail loudly, they just quietly stop matching the business
The two positions in that office were never really about routing. One person wanted the cost of standing still and the other wanted the cost of never settling, and both of those are real. What settles it is that the vehicle-day is what you are buying either way. Fixed or planned, the win is the same win: fewer rounds carrying more, which is also why drop size rather than distance decides whether a hub pays, and why two deliveries a week can beat five.