10 Smart Factory Case Studies and the Numbers Behind Them

Most manufacturers in Thailand and the wider ASEAN region have heard the term smart factory often enough. What stays unclear is far more practical — what exactly should we do in our own plant, how far should we take it, and what comes back in return. There is no shortage of conceptual explanations. What actually helps a decision is knowing what a plant of a similar size, in a similar industry, actually did and what changed as a result. This article lines up ten smart factory case studies across four angles — large enterprises, small and mid-sized plants, food processing, and Southeast Asian sites — with company names, countries, the technologies deployed, and the numbers they reported.

Why case studies get you further than theory

Smart factory projects rarely stall because information is missing. Usually the opposite is true. AI, digital twins, predictive maintenance, 5G, private networks — the list of options keeps growing until it becomes impossible to judge how far any of them sit from your own shop floor.

The way out is to start from the shape of the plant rather than the name of the technology. How many people work there, what do they make, how old is the equipment, how much was spent, and what changed. Read a handful of case studies written with all of that on the page and the picture of your own plant suddenly becomes concrete. Conversely, a success story with no figures for budget or scale is unusable for your own decision, no matter how impressive the headline percentage looks.

Four things to look for in any case study

Scale is better measured by the number of machines you need to monitor than by headcount. A plant with 30 machines and a plant with 300 need different networks, generate different data volumes, and justify investment on different logic.

Industry should be read through the character of the process. Machining and assembly put equipment utilisation at the centre. Process industries such as food or chemicals live on yield and temperature control. Logistics and warehousing turn on operator movement and picking accuracy. The same phrase — visibility — points at completely different measurements in each case.

The state of existing equipment drives cost directly. A line of new machines that can expose signals externally is one situation. A workhorse fleet of 20-year-old equipment with no communication capability is another, and the countermeasures differ. If that second description fits your plant, our article on IoT for older machines in Thai factories covers the retrofit approach in detail.

How results are measured is the point most readers skip. A claim of 30% higher productivity means very different things depending on whether it refers to labour hours, output per hour, or a figure that still includes defective units. If a case study is going into your own capital request, you need to capture that definition too.

The ten case studies at a glance

Here is the full list covered in this article. The fastest way in is to find the row closest to your own plant and read from there.

The order above follows the structure of this article rather than company size. Let us start with the large enterprises.

Large enterprise cases — what integration and automation actually changed

It is worth knowing what companies with deep investment capacity are doing, even though the scale is not directly copyable. Seeing where the road eventually leads makes it much easier to decide where your own first step belongs.

Toyota Motor Corporation — one place for all the data

Toyota Motor Corporation built a shared platform it calls Factory IoT, which centralises digitised data and enables real-time information sharing across the organisation. What deserves attention is not which sensor went on which machine, but the sequence — the company decided where the data would live before it started collecting it.

In most plants the opposite happens. Machine monitoring runs on the equipment vendor’s system, quality sits in a second system, energy in a third, and each is locally optimised. When somebody later tries to analyse across all three, the timestamp granularity does not match, the identifiers are structured differently, and reconciliation swallows an enormous amount of effort. Deciding where data lives and how it is structured before you scale is a lesson that applies at any size.

Mitsubishi Electric in Thailand — 5G and autonomous robots for automated inspection

At its Thailand site, Mitsubishi Electric deployed 5G, autonomous mobile robots, and AR and VR, achieving high-speed data processing and automated production line inspection that raised productivity. As a working example of 5G used for industrial purposes inside Thailand, it is a useful reference point for anyone evaluating the technology locally.

What really does the work here is not raw wireless bandwidth but the operational freedom to rearrange a line without rewiring it. In plants that switch product mix frequently, a cabling-dependent layout quietly becomes a constraint on improvement itself. Once wireless is the assumption, the installation work and downtime that used to accompany every layout change largely disappear.

Zuellig Pharma — 30% higher picking productivity and 100% pick accuracy

Zuellig Pharma, a major pharmaceutical logistics provider in Singapore, deployed 5G, edge computing, AR headsets, and drones across its warehouse operations. Picking productivity rose by 30% and pick accuracy reached 100%. Drone-based stock counting achieved 95% count accuracy.

What this case demonstrates is not the replacement of human work by machines, but a change in how information is presented to the person doing the work. The AR headset puts the next location and item directly in the operator’s field of view, simultaneously removing the time spent moving between a paper list and the physical goods and the mix-ups that occur during that movement. In a category where picking the wrong item is simply not acceptable, reaching 100% pick accuracy is what shows why this particular configuration matters.

The drone figure needs to be read carefully. A 95% count accuracy is not the level at which drones replace a manual stocktake outright. The practical reading is that they raise frequency. If something that could only be counted twice a year can now be counted weekly, a modest drop in per-count accuracy is more than offset by how much sooner discrepancies surface.

Small and mid-sized plants — starting with sensors that cost a few thousand yen

This is where the article becomes realistic for most readers. If the large enterprise cases left you thinking none of this is possible at your plant, the next two companies are the ones to look at.

Asahi Tekko — optical and magnetic sensors retrofitted to old machines

Asahi Tekko, an automotive parts manufacturer, retrofitted optical and magnetic sensors costing between a few thousand and a few tens of thousands of yen onto old machines, and built its own system to visualise the data in real time on smartphones and tablets. The result was labour cost savings in the hundreds of millions of yen per year and an increase in output of tens of percent.

Note the gap between those two orders of magnitude. Sensors costing a few thousand to a few tens of thousands of yen each produced annual effects measured in hundreds of millions of yen. What is actually happening here is not sophisticated automation. The plant simply became able to hold, as numbers rather than as recollection, exactly when and for how long each machine stopped.

As long as the shop floor records this on a daily paper report, a stoppage is only ever written down at the granularity of roughly half an hour. Capture the same events at second-level resolution and a different reality appears — dozens of three-minute micro-stoppages per day, each too small to write down. That accumulation is invisible in a handwritten report, and most of Asahi Tekko’s result comes from converting those previously unseen losses into figures.

Kuno Kinzoku Kogyo — managing die shot counts in the cloud

Kuno Kinzoku Kogyo, a sheet metal processor, fitted sensors to its dies to count shots automatically and manage the totals in the cloud, eliminating the risk of line stoppages caused by die breakage.

This is a good example of a tightly bounded scope. Rather than attempting to make the whole plant visible, the company reliably automated the capture of a single value — how many times a die has been struck. Because die wear tracks shot count, holding that one number accurately is enough to schedule maintenance before a failure. Once a die actually breaks, you pay for the repair, the line stands idle, and the delivery date is suddenly at risk.

In investment terms, this is a case to be justified by loss avoided rather than cost removed. When you present it internally, it helps to first establish how many times a die failure has stopped the line historically, and what each of those events cost in downtime and missed output.

A realistic entry point at smaller scale

What both of these companies share is that neither replaced any equipment. They left the machines alone and attached sensors on the outside to pick up signals. Approached that way, the initial investment is nowhere near the order of magnitude of an equipment renewal.

That said, fitting sensors does not automatically produce results. Unless somebody is designated to look at the data every morning and somebody is empowered to act on it, all you gain is another dashboard. Our article on why smart factories are not only for large corporations works through the phased approach for mid-sized plants, including the order in which investment decisions should be made. Budget ranges are covered separately in the cost of factory IoT.

Food processing — the real utilisation rate only appeared after digitisation

The next case is from Thailand, and in terms of what the numbers actually mean it may be the most instructive one in this article.

A major Thai food processor — starting with ten sausage production lines

The company is a major food processor founded in 1967 with operations across Thailand, Cambodia, Laos, and Myanmar. It deployed Advantech edge devices and IoT gateways together with WISE smart factory and OEE solutions, starting with ten sausage production lines.

The key decision was not covering every line at once but narrowing the scope to ten. Phase 1 was completed in three months. Instead of a multi-year, company-wide programme, the work was cut into a unit that could deliver results within a quarter.

The paper data was only 60-75% accurate

This is the heaviest fact in the case. Before the project, the accuracy of the data being collected on paper was only 60-75%. In other words, somewhere between a quarter and 40% of the numbers being used in management meetings did not match reality.

After digitisation, the actual OEE turned out to be 60-65% — substantially lower than assumed. The starting point for improvement was simply not where anyone believed it to be.

The same pattern shows up constantly in Japanese-owned plants in Thailand. Daily reports keep aggregating to a comfortably high utilisation rate, and then reading the equipment’s own signals directly turns up a number well below that. The difference is made up of short stoppages nobody writes down, the real duration of changeovers, and the time when equipment is running but not producing good parts. We explain how to break OEE apart in our guide to improving OEE, which is worth reading if you want to get the metric itself straight first.

Average OEE in Thai food processing runs at 40-60%

Taken alone, 60-65% might look weak. In context it is not. Average OEE across the Thai food processing industry is reported at 40-60%, well below the 85% figure generally treated as world class. Within the industry, this company sits above average rather than below it.

The lesson is that the question is not whether your OEE is low. The prior question is whether you know your true OEE at all. As long as the discussion runs on paper-based data, even the comparison against an industry average is meaningless. The first value of visibility is not improvement itself but obtaining numbers solid enough to argue from.

Southeast Asia — closing the gap between sites

The final group concerns companies operating multiple plants. If you have a Thai site and a Vietnamese site, or a Japanese head office and a local subsidiary, producing different results from the same equipment and the same drawings, this section is written for you.

A transferred plant in Vietnam — a 20-30% efficiency gap against the parent plant

Where production has been transferred from China to Vietnam, an efficiency gap of 20-30% against the parent plant in China has been reported during the early phase. Identical equipment does not deliver identical output, because operator proficiency and the fidelity with which work procedures are transmitted both differ.

One documented response combined AI vision with digital SOP management, meaning standard operating procedures held and maintained digitally. Cameras capture the work as it is actually performed and detect deviations from the standard, so it becomes possible to identify which motion in which process differs without stationing someone on site permanently.

A Thai power equipment manufacturer — rolling out best practice with AI vision

A power equipment manufacturer in Thailand used the PowerArena HOP platform to replicate optimal workflows through AI vision and to monitor operations remotely. The results were a shorter learning curve for new hires and full transparency of operations across every site. Efficiency gains of more than 10% have been reported even on mature production lines.

The phrase replication of optimal workflows captures the essential point. Every site has someone who performs the work faster and more accurately than anyone else. The traditional approach converts that person’s movements into written words for a procedure manual, and information is inevitably lost in the translation. Capture the motion from video instead and you can hand other sites the parts that never made it onto the page.

Daviteq — wireless sensors for temperature and vibration

Daviteq in Vietnam deployed wireless sensors and smart IoT gateways to resolve challenges in temperature monitoring, vibration measurement, and the storage and analysis of the resulting data.

Temperature plus vibration is a rational starting pair for maintenance. Vibration changes before bearing or motor faults become visible, and temperature signals overload or cooling problems early. Continuously capturing just those two variables is enough to detect warning signs ahead of a good share of unplanned stoppages. Because the sensors are wireless, they can be retrofitted to running equipment without any cabling work.

Makino Asia — cutting downtime with predictive maintenance

Makino Asia in Singapore deployed voice-controlled robots, robotic arms, AR remote support, driverless forklifts, and IIoT, preventing equipment failures and reducing downtime through predictive maintenance.

There is significance in a machine tool builder applying predictive maintenance to its own operations. The AR remote support setup lets engineers in Singapore view a shop floor in another country through live video and give instructions, compressing both the travel cost and the time lag of cross-border technical support. For a company with manufacturing in Thailand and its engineers in Japan, the concept transfers directly.

Three patterns shared across the ten cases

The ten cases differ in scale, country, and industry, yet the successful ones are built the same way.

One — a narrow scope and a small start

The Thai food processor started with ten sausage lines. Kuno Kinzoku Kogyo captured nothing but die shot counts. Asahi Tekko captured nothing but machine stoppage time. None of them tried to cover the whole plant at once.

The benefit of a narrow scope is not only a smaller budget. When results arrive within three months, the internal mood shifts. A two-year plan, by contrast, collects staff rotations, currency swings, and demand changes along the way until nobody can remember what the original objective was. We have turned the small-start approach into a concrete sequence of steps in our guide to small-start PoC projects for factory IoT.

Two — measurement comes before automation

Even in the large enterprise cases where robots and AI take the spotlight, data capture always comes first. The Zuellig Pharma drones exist to count inventory, and predictive maintenance at Makino Asia depends on measuring equipment condition.

Measurement is also what exposes reality. The Thai food processor discovering that its actual OEE was 60-65% is the archetype. Skip this step and go straight to automation, and you end up investing without any yardstick to measure the improvement against. The practical steps for machine monitoring are set out in our article on implementing factory IoT monitoring.

Three — gaps between sites are closed by systems, not by people

What the Vietnamese transferred plant and the Thai power equipment manufacturer have in common is that they closed the gap between sites with a system rather than by flying people in. The Vietnamese plant applied AI vision and digital SOP management to a 20-30% efficiency gap against its parent plant in China, while the Thai manufacturer used AI vision on the PowerArena HOP platform to replicate best-practice workflows and monitor remotely, reporting gains of more than 10% even on mature lines. The tactics and the numbers differ, but the character of how the gap was closed is the same.

Sending instructors from Japan works, but it costs money and time, and performance tends to drift back once they leave. When the procedure itself is held as data, the result no longer depends on how long anyone stays on site, and the same approach can be rolled out again each time a new plant is added.

How to pick the case study closest to your own plant

The table below is a checklist for deciding which of the ten cases to use as your reference.

If two or more rows describe your plant, the rule is to start with the row carrying the largest financial loss. Running several initiatives at once makes it impossible to tell which one produced the result.

For examples drawn specifically from Japanese-owned plants inside Thailand, we maintain a separate collection of IoT case studies from Thai manufacturing.

Context worth knowing before you start in Thailand

Manufacturing accounts for roughly 25% of Thailand’s GDP, and under the Thailand 4.0 policy automation and robot adoption are forecast to increase by 50% by 2026. Thailand’s digital transformation market was valued at around 10 billion US dollars as of 2025 and is expected to grow at an average of 8.75% per year through 2031.

Those figures carry two practical implications. First, competitors around you are moving in the same direction, so the later you start the harder it becomes to secure the people you need. Second, an expanding market attracts more vendors, and it becomes easier to run into proposals with no real track record behind them. The habit of checking the numbers in a case study is, among other things, a defence against that second risk.

One local consideration deserves specific attention. Operator turnover in Thailand is faster than in Japan, and systems need to be designed on that assumption. A process that lives only in the head of one experienced worker is more dangerous here than it would be at home. Seen in that light, it makes complete sense that the Thai power equipment manufacturer lists a shorter learning curve for new hires among its results.

Frequently asked questions

How should we estimate the return on a smart factory investment?

Do not start from market benchmarks. Start by putting a value on the losses already occurring in your plant. In the Asahi Tekko case, sensors costing a few thousand to a few tens of thousands of yen led to annual labour savings in the hundreds of millions of yen, because the loss being addressed was large to begin with. How much shipping delay does an hour of downtime cause? What is a one-point reduction in defect rate worth per year? Once you have that denominator, judging whether an investment is proportionate becomes straightforward.

Can a small plant really achieve results like these?

Yes. Kuno Kinzoku Kogyo, covered above, began with a system that automatically captured a single value — the shot count of a die. What matters is not the size of the budget but whether the scope is narrow enough. If anything, smaller organisations have an advantage, because decisions reach the shop floor faster and results can be reviewed in three-month cycles.

Most of our equipment is old. Do we have to replace it first?

No. As the Asahi Tekko case shows, machines with no communication capability can still be monitored by detecting their motion externally with optical or magnetic sensors. The range of information available is narrower than when signals can be read directly from the controller, but running and stopped — the highest-value information of all — can be captured. Assuming a full replacement is required is exactly what causes projects to stall, so the practical move is to try a retrofit first.

Are there considerations specific to plants in Thailand?

Three. First, assume from the outset that numbers derived from daily reports and paper tallies may not match reality. In the Thai food processor case, paper-based data accuracy was 60-75%. Second, build procedure digitisation into the plan on the assumption of operator turnover. Third, confirm that the system can be maintained locally. However well designed the initial configuration is, if no engineer in the country can respond to a failure, its value drops to zero the moment it stops.

Where should we start to minimise the risk of failure?

Choose one or two of your main production lines and begin by capturing running and stopped time automatically. Every case among the ten with a clearly demonstrated result went through this stage. The measured values you obtain here become the reference for every investment decision that follows.

Summary

Put ten smart factory case studies side by side and a pattern appears. Despite the differences in scale, country, and industry, the plants that are getting results have all followed remarkably similar paths. Narrow the scope and start small, measure before automating, and close gaps between sites with systems rather than with people. Those are the three.

Among the large enterprises, Zuellig Pharma delivered 30% higher picking productivity and 100% pick accuracy, and Mitsubishi Electric’s Thailand site automated line inspection using 5G and autonomous mobile robots. Among smaller manufacturers, Asahi Tekko achieved annual labour savings in the hundreds of millions of yen with sensors costing a few thousand to a few tens of thousands of yen each, and Kuno Kinzoku Kogyo eliminated line stoppage risk simply by managing die shot counts.

The Thai food processor case makes the sharpest point of all — paper data was only 60-75% accurate, and actual OEE turned out to be 60-65%. Before improvement comes knowing your own real numbers. If even one of these cases resembles your situation, start by measuring, on your own main line, the same thing that plant measured first.

TOMAS TECH is based in Thailand and supports Japanese-owned manufacturers with factory IT deployments, including the PEGASUS production and energy management systems. You do not need to be at the stage of selecting a product. If you simply want help working out which of these cases most closely matches your situation, that is a perfectly good place to begin. We are happy to visit your plant and think through where measurement should realistically start, so please get in touch through our contact form.

References

• ABI Research – Smart factory examples in Southeast Asia
• Advantech – Driving digital transformation in Thailand’s agro-food sector with smart manufacturing solutions
• QUANTS – IoT case studies and low-cost small starts for small and mid-sized manufacturers
• PowerArena – Smart factory trends and transformation steps in Southeast Asia
• Iconic Thai – The state of Thailand’s manufacturing industry and Thailand 4.0
• IT trend – Smart factory case studies and IoT platforms in manufacturing