In-Plant Logistics Improvement Proceeds in Three Layers

When a plant tells us that internal material movement is not keeping up, the conversation almost always opens with the name of a machine. Conveyor, AGV, or automated storage. But the first thing to decide in in-plant logistics improvement is not the equipment. What sets the total volume of material transport is not distance — it is the number of runs and the load unit carried per run. This article lays out the three-layer sequence that fixes those two variables before any machine is chosen.

Why in-plant logistics improvement fails when it starts with equipment selection

Intuitively, the amount of movement inside a plant looks like a function of distance. It is 80 meters from the warehouse to the assembly line, so it is far, so something should carry it. The reasoning feels natural, but it does not match where the people and the hours are actually going. The total labor hours poured into material transport are set by this relationship.

Total transport labor hours = number of runs × time required per run

And within the time required per run, only the round-trip travel time scales with distance. When you actually measure it, more than half of a single run is consumed by end-point work — loading, unloading, searching, waiting, and recording. Halve the distance, and if the number of runs doubles, total labor hours are unchanged. Conversely, even with distance untouched, halving the number of runs brings total labor hours close to half. And what sets the number of runs is the load unit carried per run — the number of runs is total volume divided by the load unit per run. In other words, the variables you can actually design against total transport labor hours are the number of runs and the load unit per run. Distance is not one of them. That is what the relationship above is telling you.

What happens when you start from the equipment

Projects that begin with equipment selection follow almost the same arc every time. The current number of runs and the current load unit are handed to the vendor as fixed input conditions, and the vendor quotes the fleet size and throughput that satisfy them. Hand over the conditions of a site that runs dozens of small runs a day, and what comes back is the fleet size needed to run those same runs without a human on board. The investment scales with the number of runs.

Worse, once the machine is installed, the run design becomes hard to change. After routes are drawn, charging positions fixed, and magnetic tape or reflectors laid on the floor, any attempt to consolidate runs triggers construction work on the equipment side. Push a Layer 1 decision behind Layer 3, and a change that would have cost essentially nothing turns into a change with a construction invoice attached. That is the concrete content behind the claim that skipping the sequence inflates the investment.

Which of distance, run count, and load unit actually moves the needle

Line up the three variables and the relationship between improvement headroom and cost becomes obvious.

The table shows an inversion: the two cheapest variables are the ones with the largest effect on total labor hours. That the discussion on the floor still concentrates on distance and equipment is because runs and load units are nobody’s job. Distance belongs to the layout owner and equipment belongs to the facilities owner — those boundaries are clear. The number of runs, by contrast, falls into the gap between production planning and shop-floor discretion, and in most plants there is no record anywhere of it ever having been designed explicitly. What was never designed is not recognized as something that can be improved.

The failure mode called “equipment that speeds up the inefficiency”

What you end up buying when you skip the sequence is equipment that runs an inefficient flow faster. Travel speed goes up. But nothing about what moves, when, and how much has changed, so the empty return runs, the transfers between containers, and the waiting all survive. Utilization looks better, but total labor hours do not fall as far as the business case promised.

What has to be settled first is how many times you move material and how much moves each time. Until those two are fixed, the equipment specification cannot be written, because the throughput you write into that specification is a number derived from the number of runs and the load unit. Write the answer without fixing what it is derived from, and the vendor has no option but to propose capacity with safety margin on top. Over-investment is usually not the consequence of an inflated vendor proposal. It is the consequence of a buyer who never fixed the conditions.

There are four kinds of transport waste — mechanization removes only one

The waste contained in material transport falls into four broad kinds. The classification earns its keep in practice because each kind is removed at a different layer.

Set out side by side, the four look like this.

The single most important point in this classification is that mechanization, as a rule, removes only the fourth one — returning empty. While the transport equipment is moving, the operator can leave the spot, so an empty return leg no longer costs human time. Mechanization is a way of shifting the cost of returning empty from person-hours to equipment running time. It does nothing directly to the other three.

Searching, waiting, and rehandling do not disappear with a machine

Searching waste is a problem of storage location and labeling design. Install a transport machine while storage locations are still undefined, and there is nowhere definite to set the load down. Mechanization requires fixed positions first, so in sequence terms Layer 2 comes before it. Waiting waste behaves the same way. Waiting for the upstream process has nothing to do with the performance of the transport method — it belongs to the design of run frequency and the call method. Waiting for a forklift looks like a fleet-size problem, but in practice it is usually the result of too many load units that nothing but a forklift can move.

Rehandling waste is caused by containers and load units that are not aligned. When process A uses plastic containers, process B uses metal pallets, and shipping uses cartons, a transfer is guaranteed at every boundary. Installing a transport machine does not remove the transfer itself. If anything, transport equipment is designed around a single consistent load unit, so on a site where load units are not aligned, new manual transfer work gets created in front of and behind the machine. This is the textbook case of headcount not falling after go-live.

How to use the four kinds in a baseline study

The classification only becomes useful once it is turned into a way of measuring. The method we recommend is to allocate a transport operator’s day across the four kinds. There is no need for stopwatch precision. Ask the operator to record, in 30-minute blocks over one week, which of searching, waiting, rehandling, returning empty, or actually moving loads applies. That alone gets you the right order of magnitude. Keep this record for a week and the share of time in which material is genuinely moving comes out lower than anyone expected, and the shortfall is usually concentrated in searching and waiting. In an investment review meeting, putting one page of that week’s record on the table works faster than explaining the three-layer sequence.

Improvement runs in three layers — runs and load units, storage locations and travel paths, mechanization

In-plant logistics improvement splits into three layers that differ by an order of magnitude in cost. The sequence is fixed from the bottom up, and skipping a layer makes the layer above it impossible to design.

Here is what sits in each of the three layers.

The sequence is fixed because the design inputs of each layer are the outputs of the layer below. The throughput required at Layer 3 is calculated from the number of runs and the load unit fixed at Layer 1. The position and count of stop points required at Layer 3 follow from the storage locations fixed at Layer 2. If the lower layer is undecided, the upper one cannot be decided. Put the other way round, once the lower layer is fixed, writing the upper layer’s specification becomes mechanical.

The three layers differ not only in cost but in how long the effect takes to appear and how easily they can be reversed. Layer 1 is settled with paper and rules, so it can be started within weeks, shows up in the numbers in one to two months, and if you get it wrong you simply revert the rule. Layer 2 involves physically moving things on the floor, so it takes one to three months to get started, and it can be reversed by resetting the storage locations. Layer 3, including specification freeze and procurement, takes six months to a year, and the investment cannot be reversed. Try things first in the reversible layers, and fix the conditions before entering the irreversible one. This sequence pays off especially at overseas sites. Deciding local working practice from head-office assumptions alone always produces a gap, but at Layer 1 the gap can be corrected the moment it appears.

The symptoms shared by projects that skipped the sequence

Look back at automated logistics projects after the fact and the ones that skipped the sequence share a set of symptoms. The labor-hour savings promised before go-live do not match what actually happened. Trace the reason and you find that the hours assumed to be removable included searching, waiting, and rehandling. Mechanization does not remove those, so they never show up in the results. The second symptom is new human work created after go-live — transfers to align goods with the load unit the machine can handle, relabeling into a form the machine can read, and waiting for recovery plus manual transport during stoppages. None of these appear in the pre-project estimate. Clear Layers 1 and 2 first and none of them occur, or they occur at a much smaller scale.

Layer 1 — settle scheduled runs (mizusumashi) and the load unit first

Layer 1 costs almost nothing and yet has the largest effect on total labor hours of any layer. What it settles can be organized into five decisions.

Decision 1 — move to scheduled runs

The first decision is to change material transport from “go when called” to “circulate at fixed times.” This is the method known as mizusumashi, or the in-plant milk run — the milk-run idea of a fixed collection circuit applied inside the plant. The route and the times of the in-plant circuit are fixed, and every process prepares what it will hand over and what it will receive in time for that run.

The problem with the go-when-called method is that waiting is generated on both sides. The caller waits for the transport operator to arrive, and the transport operator waits to be called again. Neither wait is recorded anywhere, so neither is recognized as something to improve. Switch to scheduled runs and both waits collapse into a single number — the interval between runs. Once it is a number, it can be designed. A significant secondary benefit is that return-leg loads become possible to include. Fix the circuit and each stop point can handle what is set down and what is picked up in the same visit, so the fourth kind of waste, returning empty, falls structurally.

Decision 2 — set the run frequency

The interval between runs follows from the consumption rate of the process and the pick face capacity on the process side. If the pick face holds two hours of stock, runs every two hours are enough; if it holds only 30 minutes, runs every 30 minutes are required. Turned around, if you want fewer runs, enlarge the pick face capacity — which is a Layer 2 conversation. The trade-off to watch is that a shorter interval makes life easier for the process while increasing total transport labor hours. Take requests from the floor at face value and the number of runs grows without limit. Use the required interval back-calculated from pick face capacity as the reference, and ask that any request be translated into an argument about pick face capacity.

Decision 3 — set the load unit per run

Fix the load unit carried in one run. Leave this vague and the load varies run to run, which makes the time per run unpredictable. Unpredictable time never makes it into a plan, and the operation drifts back to depending on individuals. The basis is the smaller of two figures — the weight a person can safely handle, and the quantity the process can receive and consume in one go. A larger load unit means fewer runs, but too large and a forklift becomes necessary, which sharply tightens the constraints at Layers 2 and 3. This relationship is the core of the forklift reduction discussion later in the article.

Decision 4 — standardize containers

Reducing the number of container types is the single most effective way to remove the rehandling waste. Transfer points that arise from the physical requirements of a process will remain, and those are narrowed down individually at Layer 2. But as long as each process uses a different container, a transfer at every boundary is unavoidable. Unify them and you not only remove transfers — the loading pattern becomes fixed, so the time per run becomes predictable, and the preconditions for future mechanization fall into place at the same time. It does not have to be a single company-wide standard. Start by converging on one type across the high-volume main processes and keep the exceptions explicit. Manage the list of container types on a single page, so exceptions cannot quietly multiply.

Decision 5 — choose the call method

Even after switching to scheduled runs, you still need a mechanism that communicates what to load. There are three main options.

Which one you choose depends on item count and the size of variation. Rather than jumping straight to electronic instruction, building the operating discipline with kanban first and then digitizing it tends to stick. A mechanism that does not work on paper will not work electronically either.

Why Layer 1 pays off at overseas plants in particular

Layer 1 carries a benefit that is specific to overseas plants — it lowers the cost of training. Once you accept that local staff turnover is a given, any operation that depends on individuals degrades at every handover. A site that runs on “he knows where everything is” sees searching waste spike the moment he leaves.

If the scheduled-run route and times, the container types, and the load unit are all fixed, what a new person has to be taught fits on a single sheet. The smaller the room for judgment, the less differences in language and experience show up in the result. What a Japanese domestic plant absorbs through the judgment of experienced operators simply does not function at an overseas site unless it has been converted into explicit rules. That is the reason, specific to overseas sites, why Layer 1 cannot be deferred.

Layer 2 — storage locations and travel paths, where forklift reduction is not about fleet size

Once Layer 1 has settled runs and load units, the next thing to settle is the physical storage locations and travel paths. The cost stays within the range of labels, floor markings, racking, and rebuilt pick faces.

Fixing storage positions and designing the pick face

Fixing storage positions removes the searching waste. What has to be decided is less the place itself than the capacity — what can be stored there and how much of it. Without a defined capacity, the location overflows in peak season, goods get parked somewhere else temporarily, and the fixed position loses its meaning. A fixed position only works when it comes paired with a maximum permitted quantity.

Pick face design is directly linked to the run frequency from Layer 1. Give the pick face two hours of capacity and runs every two hours are enough. Compare the cost of enlarging the pick face against the labor-hour cost of adding runs, and take the cheaper one. The comparison itself is trivial, but if the pick face conversation and the run conversation sit with different people, no forum for the comparison ever gets created. Always have the Layer 1 owner in the room for the Layer 2 design.

Count the transfer points

The first thing to do in travel path design is to count the number of times material is rehandled between receipt of raw material and shipment of finished goods. Count the points at which the container or the load unit changes, not the number of processes. That figure is, directly, the volume of manual work. Transfer points split into those that disappear when containers are unified and those that cannot be removed because they come from a physical requirement of the process, such as a dedicated fixture for a heat treatment furnace. Target only the removable ones. Try to remove all of them and the process owners push back, and you stall before even reaching the points that could have been removed.

Forklift reduction is about load units, not fleet size

When forklift reduction comes up in a discussion about in-plant logistics efficiency, the conversation almost always starts with fleet size. But fleet size is a result, not a cause. Forklifts are necessary because there are loads that nothing else can move. As long as pallet-scale lots move between processes, the fleet will not shrink.

The correct procedure, therefore, is not to set a fleet reduction target but to list the loads that only a forklift can move, and to take them off the list starting with the ones whose load unit can be made smaller. Material that used to arrive at the process on a pallet gets split into containers a cart can handle and put onto the scheduled run. The fleet shrinks as the accumulated result of that. Reverse the order and cut the fleet first, and the remaining large lots simply form queues — all you have added is waiting waste.

How forklifts look from the safety side

The effort to reduce load units has a justification separate from productivity. The figures below come from the Japan Industrial Vehicles Association’s compilation of the Ministry of Health, Labour and Welfare “Industrial Accident Statistics” for 2023, and they are Japanese domestic statistics. They are explicitly not Thai or ASEAN figures.

In 2023 there were 1,989 forklift-related accidents causing injury or death, of which 22 were fatal. By accident type, caught-in or entangled accounted for 35.4% and struck-by for 27.3% of injury-and-fatality accidents.

Isolate the fatal accidents and the mix by type changes.

Tip-over is 5.7% of all injury-and-fatality accidents but rises to 17.2% of fatal ones. Falls from height rise from 11.9% to 18.0%. The types that occur most often and the types that kill when they occur are not the same. Frequency countermeasures and severity countermeasures have to be planned separately.

Training and signage can bring frequency down, but tip-over and fall types can reasonably be assumed to scale with the total amount of driving that happens. The only lever that reduces that exposure itself is reducing the load units that only a forklift can move. This is the layer at which safety measures and logistics improvement converge on the same action.

Handling intersections

The last item in travel path design is intersections. Identify every point where a pedestrian route crosses a forklift route, and where a scheduled-run circuit crosses in-process movement, and reduce the count. For those that cannot be removed, define the priority. At an intersection where it is not settled who stops, both parties slow down and yield every single time — a few seconds each, but multiply by the number of runs and it becomes time you cannot ignore. The default rule is pedestrian priority, which means the vehicle is the one that stops. Because repeatedly stopping and restarting while carrying a load is itself a cause of load collapse and contact incidents, the answer on routes that carry loads frequently is not to make people give way. It is to mark the priority on the floor with a stop line and signage, and above all to lay the paths out so that load routes and pedestrian routes do not cross in the first place. As long as the decision is left to the two parties yielding to each other on the spot, neither the waiting time nor the contact risk goes down.

Layer 3 — choosing among material handling equipment, from conveyor and AGV or AMR to automated storage and sorting

Only once Layers 1 and 2 are done can mechanization be designed. Selection among material handling equipment uses three decision axes.

The axes are route stability, run frequency, and load unit weight. Distance and annual transport volume are inputs to the capacity calculation performed after those three are fixed — they cannot be used to select the method. “It is a long distance, therefore a conveyor” does not hold, because at the same distance the answer changes depending on whether the route can be held fixed.

The column that matters in this table is the weakness column. Method selection is not decided by comparing what each one can do, but by comparing whether each one’s weakness can survive your own variation. Choose the fixed route of a conveyor in a plant whose product mix changes several times a year, and every change generates construction cost. Conversely, in a process whose route has not changed in ten years, that same fixity is not a weakness but an advantage.

How to judge route stability

Judge route stability from historical evidence, not from intuition. The figure to use is how many times the layout or process order of that segment has changed in the last three years. Zero means high stability. If it has changed once a year or more, avoid methods that assume a fixed route. Look at future plans as well — expansions, relocations, or product changeovers on the plan all lower stability. The point to watch is that whether such plans exist should not be judged by the facilities owner alone. Without checking the product mix trajectory that production planning is assuming, the facilities-side view on its own overestimates stability.

Run frequency and load unit weight

Run frequency uses the figure fixed at Layer 1 as it stands. The higher the frequency, the higher the utilization of fixed equipment and the better the payback conditions. Install fixed equipment on a low-frequency segment and it sits idle most of the time. Load unit weight also comes from the Layer 1 load unit. The heavier it is, the fewer methods remain available and the higher the investment. This is where a decision to go back to Layer 1 sometimes emerges. If the conclusion is that a smaller load unit would allow a cheaper method, redo Layer 1. The three layers are not strictly one-way, but redesign always restarts from the lower layer.

Keep storage and transport separate

In Layer 3 work it is better not to mix storage with transport. Automated storage automates the storage point, while AGVs and AMRs automate transport between processes. They solve different problems, so comparing them on the same terms produces a discussion that never lands. Investment decisions on the storage side are handled separately in our article on the cost and payback conditions of automated storage.

On the transport side, the fleet calculation directly determines the investment. The required fleet size is derived from run frequency, time per run, and charging or standby time — and the premises of that calculation are the Layer 1 numbers. The approach to fleet calculation and the influence of layout conditions are covered in our article on AGV placement and fleet sizing, and the conditions under which AGVs, AMRs, or manual transport come out ahead are summarized in our comparison of transport methods.

What the buyer has to decide before ordering a conveyor design

Once a conveyor system is installed, changing its route means construction work. Precisely for that reason, there are more items the buyer must settle before ordering than with any other method. The items below are the ones that come back as problems on projects handed wholesale to the vendor.

Of these, the top three are the ones only the buyer can decide. The load unit to be carried comes from the Layer 1 load unit, and peak throughput comes from the run frequency. Operation during a stoppage is a management decision about how production continues when that segment goes down, and it is not something a vendor can decide.

Setting the peak value and the crossing points

The common failure with throughput design values is designing to the average and jamming at peak. The average is easy to produce — divide monthly output by operating hours — but actual transport peaks right after shift start, right after breaks, and right after changeovers. The method we recommend is to pull daily peaks from the last twelve months of production records and set the design value at the top 5% of days. Using the absolute maximum as the design value oversizes the equipment. Unless the split is agreed in advance — take the top 5% of days, and absorb anything above that operationally — the capacity discussion never converges.

Last come height and crossing points. A conveyor occupies floor and space, so it severs walking routes, and severed routes lengthen walking distance. Place crossing points where people cross frequently, and determine that frequency by observing how people actually walk today. This is where the Layer 2 travel path design pays off. Fix the route while the travel paths are still disorganized and the crossing points will no longer line up once you come back to fix the paths.

What conditions make sorting automation start to pay

Sorting automation is one of the more narrowly conditioned areas within material handling. Pin the conditions under which it starts to pay on structure rather than on volume.

The four conditions under which it starts to pay

First, the number of sort destinations has to be stable. In an operation where destinations are added and removed frequently, the equipment configuration is being changed continuously. Second, volume has to have peaks over time. If volume is leveled, staffing it at a level rate is enough. It is the peaks that create the waste of staffing to the peak, and that is what gives the equipment something worth absorbing.

Third, variation in item shape and weight has to be small. A sorter’s transport surface and pusher mechanism are designed around an assumed shape, so wide variation produces more jams and drops. In that case the first move is container standardization — which means going back to Layer 1. Fourth, the cost of a mis-sort has to be high. In a process where a part delivered to the wrong process gets assembled in, the loss per occurrence is large, and that is exactly the amount by which a lower error rate from equipment can be booked as a benefit.

Build the volume threshold from your own numbers

There is no cross-industry general answer to the question “from how many lines does sorting automation become viable.” A line here means a single sort instruction for one item on one order detail line. There is no general answer because the number of sort destinations, the processing time per line, the labor rate, and the cost of a mis-sort all feed into it. The way to build the threshold is in this order. First, total the labor hours currently going into sorting, split between peak days and other days. Next, confirm with the four-kind classification which portion equipment can actually replace. The sorting work itself can be replaced, but if the transfers and label checks before and after it remain, subtract those from the hours saved. Finally, multiply the remaining hours saved by the labor rate to get the annual benefit. Only once that number exists is the ceiling on the investment defined. It is the same procedure as the one covered in the next section.

The middle ground between manual and automated

Sorting is also an area with an unusually wide range of options between fully automatic and fully manual. Simply combining digital display instructions, weight verification, or barcode verification with manual work brings the error rate down substantially, at an order of investment that is not remotely comparable to an automated sorter. So the order of investigation is to first bring the error rate down with display and verification mechanisms, and only if labor hours are still short to consider automation. Where the cause of errors sits upstream in the labeling, the errors survive the equipment. Here too, Layers 1 and 2 come first.

How to think about the cost and payback of in-plant logistics efficiency

The place where investment decisions go wrong most often is how the benefit is booked. Fix that first.

Fix a single baseline

Improving material transport produces several benefits at once — reduced labor hours, reduced outsourced transport cost, fewer weekend shifts, fewer losses from mis-sorts, and lower accident risk. Add all of these together to compute a payback period and the period comes out shorter than reality. The reason is simple. The same saving is being counted twice.

If, for instance, weekend shifts fell as a result of reduced transport labor hours, then the labor-hour saving and the weekend-shift saving are two descriptions of the same phenomenon. Add both and the benefit doubles, while the actual reduction in cash out is only one of them.

The practical answer is to fix a single baseline against which the benefit is taken, and to describe everything else qualitatively as a secondary effect. In the worked example below, the baseline is fixed to the reduction in transport labor hours only. Outsourced transport cost, weekend shifts, and accident risk are not converted into money.

Assumptions for the worked example

The following is an illustration of the method. Every amount is a placeholder figure, meant to be replaced with your own actuals.

From these assumptions we derive the annual transport labor hours and cost. 4 people × 8 hours = 32 person-hours per day. 32 person-hours × 25 days = 800 person-hours per month. 800 person-hours × 12 months = 9,600 person-hours per year. 9,600 person-hours × 50 baht = 480,000 baht per year. That is the transport labor cost before improvement.

Benefit by layer and the ceiling on investment

The reduction rates for each layer are placeholder figures too. In practice they come from the baseline study records.

Let us check the arithmetic. At 12% for Layer 1 alone, 9,600 × 0.12 = 1,152 person-hours, worth 1,152 × 50 = 57,600 baht. At a cumulative 20% through Layer 2, 9,600 × 0.20 = 1,920 person-hours, so the Layer 2 increment is 1,920 – 1,152 = 768 person-hours, worth 768 × 50 = 38,400 baht. At a cumulative 45% through Layer 3, 9,600 × 0.45 = 4,320 person-hours, so the Layer 3 increment is 4,320 – 1,920 = 2,400 person-hours, worth 2,400 × 50 = 120,000 baht.

From here we derive the ceiling on the Layer 3 investment. The numerator is the Layer 3 investment; the denominator is the Layer 3 incremental benefit of 120,000 baht per year and nothing else. The 96,000 baht per year of Layer 1 and Layer 2 benefit must not be added to the denominator. Adding it means counting benefits you would have obtained without buying the machine as benefits of the machine.

Checking the arithmetic. 120,000 × 2 = 240,000. 120,000 × 3 = 360,000. 120,000 × 5 = 600,000. All three agree.

Align the definition of investment with the formula

Make explicit what “investment” means in the comparison against the ceiling. It includes the equipment itself, installation work, power and communications cabling, control system interfacing, first-year maintenance, and training during ramp-up. Put only the quoted equipment price in the numerator and the gap against actual spend feeds straight into the payback period.

Whether recurring annual maintenance goes into the numerator or comes off the denominator does not matter, as long as you do not mix the two. Our recommendation here is to include the first year in the numerator and to deduct maintenance from year two onward from the benefit in the denominator. If annual maintenance from year two is, say, 20,000 baht, the year-one denominator is 120,000 baht (first-year maintenance sits in the numerator), and the year-two and year-three denominators are 120,000 – 20,000 = 100,000 baht each. The three-year investment ceiling is therefore 120,000 + 100,000 + 100,000 = 320,000 baht, which is 40,000 baht below the 360,000 baht you get by ignoring maintenance. Treating maintenance entirely in the numerator lands in the same place — 360,000 – 20,000 × 3 = 300,000 baht, plus the 20,000 baht of first-year maintenance added back, gives the same level. Either treatment reaches the same conclusion, so the only thing that matters is not mixing them.

Sensitivity to the labor rate

The labor rate feeds directly into the payback period, so it is worth looking at the sensitivity. At 60 baht per hour, the Layer 3 incremental benefit becomes 2,400 × 60 = 144,000 baht per year, and the three-year investment ceiling becomes 144,000 × 3 = 432,000 baht. At 40 baht, it is 2,400 × 40 = 96,000 baht per year, and 288,000 baht over three years.

Thailand’s provincial minimum wages sit in a range of 337 to 400 baht per day as of 2026 and are not uniform nationwide. The same equipment, in other words, supports a different payback period depending on the province it is installed in. How to build wage trends into the model is covered in our article on automation investment under rising Thai labor costs, which works through a payback model that incorporates a rate of wage increase.

Avoid asserting price levels

We deliberately do not quote market prices for conveyors or AGVs here. The order of magnitude changes with specification, fleet size, installation conditions, and the scope of control system interfacing. What the buyer should do is not research prices but fix the investment ceiling first, using the procedure above. When a quotation comes in above the ceiling, there are exactly two options — change the method, or go back to Layers 1 and 2 and build up more reduction.

The order that pays off first in Thailand and other overseas plants, and what 57.47% capacity utilization means

Thai manufacturing has a macro condition that shapes the order of investment.

Capacity is spare and labor is short

Average capacity utilization in Thai manufacturing stands at 57.47% as of the second quarter of 2026. The industrial production index for June 2026 was down 3.1% year on year. What these two numbers describe is a state in which there is slack in capacity. Investing in more capacity while capacity is already slack is hard to pay back, because the added capacity does not run. Labor tightness, meanwhile, continues, and securing workers is a persistently cited issue across ASEAN warehousing and logistics.

The character of the investment, therefore, should be labor-hour reduction rather than capacity expansion. In-plant logistics is precisely a labor-hour reduction target, and because it does not touch production capacity itself, it produces results even while utilization is low. Layers 1 and 2 can be executed without stopping equipment, which makes the fit with this environment better still.

Layer 3 is hard to justify on labor savings alone

At the same time there is a fact worth looking at soberly. Thailand’s hourly labor rate is low compared with Japan. As the worked example above shows, at a rate of 50 baht per hour the Layer 3 incremental benefit is on the order of 120,000 baht a year, and on a three-year payback condition the investment ceiling stops at 360,000 baht. So if you want a Layer 3 investment to stand up at a Thai site, you have to pick a segment where factors beyond labor savings — safety, quality, round-the-clock operation — apply simultaneously. Even then, only the one factor chosen as the baseline is booked in money. Convert safety and quality into money and add them in, and you are back to the double counting described in the previous section.

The flip side is that at a Thai site the return on Layers 1 and 2 is relatively very high. There is no reason to skip a layer where containers, labeling, and organized storage locations put a 20% labor-hour reduction within sight, in order to start arguing over a 360,000 baht envelope. Note also that the reduction in training cost from standardization is not included in this calculation. It does not appear in an existing cost account and the basis for monetizing it is weak — but in reality you can treat it as sitting on top. The upstream judgment of which processes to select for automation is covered in our article on choosing the target process for automation.

BOI incentives are aimed at upgrading existing equipment

The Thailand Board of Investment (BOI) operates a “Smart and Sustainable Industry” measure that supports the upgrading of existing facilities. In the first quarter of 2026 it drew 61 applications worth 7,071 million baht. Over the same period, applications in logistics and high value-added services numbered 68, worth 14,548 million baht.

Because this measure targets improvement of existing facilities rather than new capacity expansion, it matches the character of investment appropriate to a low-utilization environment. If you are considering a Layer 3 investment, it is worth checking the application requirements and eligible scope early. Requirements are revised, however, so check the latest published details. Similarly named measures are easy to confuse, so we recommend quoting the exact measure name when you make an inquiry.

Checklist for running a logistics improvement, or what to settle on paper before ordering

Here is everything above organized as the items to settle on paper before ordering. Confirm the list is filled in before you walk into a vendor meeting.

Not one item on this list is filled in by asking a vendor. Every one of them is settled internally on the buyer’s side. Filled in, quotation comparison simplifies to a comparison of capacity and price. Take quotations with the list unfilled and the axis of comparison differs by vendor, and the proposal that explains itself best wins.

The order in which to run it

In practice, start with the one-week baseline study, run Layer 1 over one to two months, measure the effect, and only then move to Layer 2. Once Layer 2 has produced its effect, break the remaining labor hours down by the four kinds again. If returning empty is what mainly remains at that point, the conditions for entering Layer 3 are in place. If the other three still remain, go back to Layers 1 and 2. The criterion is simple but effective. If none of the waste that mechanization removes is left, mechanization will not help.

Frequently asked questions

Here are five questions that come up most often on the floor.

Where should we start with in-plant logistics improvement?

Start with a baseline study. Have transport staff record one week of their time in five categories — searching, waiting, rehandling, returning empty, and actually moving loads. Precise measurement is unnecessary; self-reporting in 30-minute blocks gets the order of magnitude. Decide on actions without this record and you will forecast waste that mechanization cannot remove as a benefit of mechanization. Once the record exists, settle the scheduled run route and times, the load unit per run, and the container types. Up to that point the cost is close to nothing and the effect shows in the numbers within one to two months. Looking at transport equipment comes after that.

For in-plant logistics efficiency, is a conveyor or an AGV the better fit?

Neither distance nor volume decides it. The three things used to judge are route stability, run frequency, and load unit weight. If the route has not changed once in the last three years, no change is planned, and run frequency is high, a fixed-route conveyor system has the advantage. On segments where the route changes once a year or more, or where expansions or product changeovers are planned, methods that require construction to change the route are best avoided. If loads that only a forklift can move are still present, look at whether the load unit can be split before choosing between the two at all. A smaller load unit widens the options and lowers the order of investment.

Is forklift reduction realistically achievable?

Set fleet size as the target and it is hard. Set the load unit as the target and it moves. Forklifts are needed because loads exist that nothing else can move, and fleet size is the result of that. The procedure is to list those loads and accumulate substitutions — material arriving at a process on a pallet gets split into containers a cart can handle and put onto the scheduled run. There is safety evidence behind it as well. In Japanese industrial accident statistics, 2023 saw 1,989 forklift-related accidents causing injury or death and 22 fatalities, and among fatal accidents tip-over accounted for 17.2% and falls from height for 18.0% — far above their 5.7% and 11.9% shares of all injury-and-fatality accidents. Accidents of that type scale with the total amount of driving, and the only way to reduce that exposure is to reduce load units.

At what volume should we start considering sorting automation?

Absolute volume does not decide it. There are four conditions under which it starts to pay — the number of sort destinations is stable, volume has peaks over time, variation in item shape and weight is small, and the cost of a mis-sort is high. Even at high volume, if it is leveled, staffing handles it. The order of judgment is to first bring the error rate down with middle-ground measures such as digital display instructions and weight verification, and to consider automation only if labor hours are still short. Where shape variation is large, go back to container standardization before automating.

How much does in-plant logistics improvement cost?

The order of magnitude differs by layer. Designing runs and load units at Layer 1 costs close to nothing beyond containers and labeling. Layer 2 stays within labels, floor markings, racking, and rebuilt pick faces, which is small next to production equipment investment. The only layer where the order of magnitude changes is mechanization at Layer 3. So the way to answer a cost question is not to research prices but to fix the investment ceiling first. Multiply the annual hours saved by the labor rate to get the annual benefit, then multiply by the target payback period to get the ceiling. When you do, fix the benefit baseline to one thing, and do not book labor savings and outsourcing savings at the same time.

Summary

This article makes two claims. First, what sets the total volume of material transport in in-plant logistics improvement is not distance but two things — the number of runs and the load unit per run. Second, of the four kinds of transport waste, mechanization removes only returning empty; searching, waiting, and rehandling can only be removed at Layers 1 and 2.

The sequence therefore has three layers — settle runs, load units, and containers at Layer 1; organize storage locations and travel paths at Layer 2; design mechanization at Layer 3. Skip the sequence and start from equipment selection, and you buy equipment that runs an inefficient flow faster, at an excessive fleet size and an excessive investment. The decision axes for method selection at Layer 3 are the three variables of route stability, run frequency, and load unit weight — never distance or volume on their own.

Forklift reduction has the same structure. What has to be reduced is not the fleet but the load units that only a forklift can move. While those load units remain the fleet will not shrink, and as Japanese industrial accident statistics show, severe accident types such as tip-over and falls from height scale with the total amount of driving. On cost and payback, fix a single baseline for the benefit and align the definition of investment with the payback formula. Do not book labor savings and outsourcing savings at the same time.

Thai manufacturing is in a state of slack equipment and short labor, with capacity utilization at 57.47% in the second quarter of 2026 and industrial production in June 2026 down 3.1% year on year. In the sense that labor-hour reduction investment comes before capacity expansion investment, in-plant logistics ranks early as a target. And in an environment where the hourly labor rate is low, the relative value of Layer 1 — which works at essentially zero cost — is higher still.

In in-plant logistics improvement, more is decided before the equipment purchase stage than at it. We are happy to talk at the stage where there is not yet a single scheduled-run timetable, or where the container list is still to be written. If you want to work through how to run the baseline study records, how to calculate the investment ceiling, or whether the conditions for moving to Layer 3 are in place, get in touch through our contact page. Getting in touch well before you request equipment quotations is entirely fine.

References

• Hitachi, Ltd. “In-plant logistics — the issues companies face and the points for improvement” https://www.hitachi.co.jp/products/infrastructure/portal/industry/factory_automation/column_2/index.html
• Genba Kaizen Lab “The reality of forklift accidents” (Ministry of Health, Labour and Welfare Industrial Accident Statistics, compiled by the Japan Industrial Vehicles Association) https://tebiki.jp/genba/useful/logi-forklift-accidents/
• Japan Industrial Vehicles Association “Occurrence of industrial accidents caused by forklifts” (2025) http://www.jiva.or.jp/pdf/25_SafetyDay_1-1.pdf
• Trading Economics / Office of Industrial Economics (OIE) “Thailand Capacity Utilization” https://jp.tradingeconomics.com/thailand/capacity-utilization
• Xinhua “Thailand’s industrial output slips 3.1 pct in June” (2026-07-27) https://english.news.cn/20260727/f8b7ceb202d04b1f8facbcbf44ed60ce/c.html
• Thailand Board of Investment (BOI) investment statistics and the Smart and Sustainable Industry measure https://www.boi.go.th/un/boi_event_detail?module=news&topic_id=138788&language=en
• Thai Law Online “Minimum Wage in Thailand 2026” https://www.thailawonline.com/minimum-wage-in-thailand/
• Mujin Logistics Robot Consulting Room “What is in-plant logistics” https://media.mujin.co.jp/logistics/basic/fa-logi/
• Funai Consulting “In-plant logistics improvement — five steps to solve the issues” https://smart-factory.funaisoken.co.jp/blogs/column/factorydx-5942
• Value Chain Asia “Asia warehouse labour shortage 2026” https://valuechainasia.com/articles/human-resources/asia-warehouse-labour-shortage-2026