Factory Call and Notification System in Thailand: NBTC Rules, Cost and ROI

Walk into enough Japanese-owned plants around Bangkok, Chonburi and Rayong and you start hearing the same request. “Our parent plant in Japan uses a call system. We want the same setup here.” A machine stops and the maintenance technician’s wrist buzzes. Material runs low and a notification lands on the logistics team’s phone. In Japan that picture is close to standard equipment.

Almost every one of those projects stalls, and not because of budget or shop-floor resistance. It stalls because you cannot assume that the wireless call equipment running in Japan will work in Thailand as-is. The 429MHz band commonly used by Japan’s specified low-power radio category is not one of the bands Thailand has opened for licence-exempt use. On top of that, radio equipment in Thailand requires NBTC certification, and that process requires a local representative in Thailand. The starting premise — “order the same part numbers the parent company uses” — does not survive contact with the local rules.

This article is not written for readers still deciding *whether* to install a factory call and notification system. It is a working guide for getting to the point where you can answer four concrete questions: which of your calls, routed to which device, at what cost, with what response record. It covers the four call types, a comparison of the three notification device families, Thailand’s radio regime, a five-layer cost breakdown, and a two-scenario investment model built on a 30-machine plant — with every number shown so you can rebuild it with your own data.

No figure appears here without a source behind it. The goal is that by the end you could draft the skeleton of your capital request.

Why factory call systems are on the agenda now — and what they will not fix

Andon has moved from the board to the wrist

Andon is the mechanism that signals abnormalities in a production process using indicator lights or electronic display boards, enabling early detection and rapid response. The colours used are mainly green, yellow, white and red. Most people picture a rotating beacon hanging near the ceiling or a display board at the end of a line.

That destination is changing. Traditional andon was a visual signal on a board; today the notification increasingly goes to mobile devices, smartwatches and email, and some implementations connect directly to MES and ERP. In other words, andon is expanding from “a tool for making things visible” to “a mechanism for moving people”.

The reason for the shift is simple: a display board only reaches people who happen to be looking at it. The larger the floor area, and the more machines a single person covers, the lower the odds that anyone is standing in front of that board. On a thinly staffed night shift. While the maintenance technician is in another building. While the forklift driver is somewhere across the site. The board is lit correctly and nobody has noticed. That is the structural ceiling of board-based andon.

Wrist-worn receivers and smartphones attack that ceiling by delivering the signal to the person rather than to the room. In digital andon implementations, the intended operating model includes automatic escalation when a call goes unanswered, and a record of how it was handled. At that point you are no longer extending a call bell — you are running a business process.

There is academic work in the same direction. A paper in MDPI’s *Applied Sciences*, volume 15, issue 5, article number 2806 (published 5 March 2025, DOI: 10.3390/app15052806), titled “Smartwatch-Based Monitoring and Alert System for Factory Operators Using Public Cloud Services”, presents a smartwatch-based framework using public cloud services to deliver two-way notification between employees and supervisors and among colleagues. It addresses real-time situational awareness, task management and working-time recording, with the aim of simplifying reporting procedures. The point worth carrying forward is that the research frames the problem as a two-way exchange, not a one-way alert. This article does not use any latency or cost figures from that paper; the reference stops at “a two-way notification framework has been proposed in the research literature”.

One paragraph on market context, and no more. The overall wearables market was estimated at USD 83.5 billion in 2025, with an expected compound annual growth rate of 8% over 2026–2035. For industrial wearables specifically, one estimate puts the 2026–2035 CAGR at 12.0%, describing the category as moving from pilots into full deployment. These come from different research firms using different methodologies, so they cannot be compared directly — and more importantly, market size is not a basis for your own investment decision. Everything after this point runs on your plant’s own numbers.

What “chokotei” actually is, and where calls help

Any discussion of call system benefits eventually reaches minor stoppages — *chokotei* in Japanese shop-floor vocabulary. A minor stoppage is a temporary equipment stop that is resolved in a short time, generally lasting from a few seconds to a few minutes. It is one of the seven major losses in manufacturing and is classified as a performance loss.

This is a decisive fork in the road. Stoppages that clear in seconds to minutes usually generate no call at all — the operator clears the jam and moves on. Call systems earn their keep one level up: stoppages the operator cannot resolve alone, where someone from maintenance, quality or logistics has to be summoned.

So before you look at any product, establish three things about your own plant:

• What share of your stoppages can operators resolve without help?
• How many stoppages per day genuinely require calling someone?
• How many minutes pass between the call and arrival, on average — and how many calls end up ignored entirely?

Start comparing products without these three numbers and your capital request will come back. The financial reason for that appears in the model section below.

What a call system solves, and what it does not

Set expectations before anything else. A call system addresses “searching”, “waiting” and “not noticing”. It does not address “fixing”.

A call system does not reduce how often calls occur. If your goal is fewer calls, invest in monitoring equipment condition and catching failures before they happen — that is the territory of predictive maintenance systems and equipment monitoring. A call system shortens the handling time of calls that have already happened. The two are complementary. Blur that boundary in your internal pitch — “install a call system and stoppages will drop” — and disappointment after go-live is guaranteed.

The four call types, and how each one changes the design

Treating all your calls as a single category and then shopping for a product is a reliable way to fail. Sort your calls into four types first. Each type differs in the device it needs, the response time it tolerates, and the fields worth recording.

Where the design branches by type

Type 1, equipment abnormality, should not depend on a human pressing a button. Once you rely on people, you get forgotten presses and “it’s faster to just walk over and ask”, and neither leaves a record. Capture the signal-tower contact with a transmitter instead, and the call is raised automatically the moment the machine shows yellow or red. This type pairs best with the response log discussed later, and it maps directly to the financial benefit of reduced stoppage time.

Type 2, material replenishment and transport, turns on one question: can the call be raised early? Call when material has already run out and the machine stops. Call when 30 minutes of material remains and it does not stop. That reframes the call as a *forecast* rather than an *abnormality*, which lets you lower its urgency tier. If forklift dispatch is in scope, remember that the driver is in motion — a fixed display board will never reach them. A vibrating wrist receiver or a smartphone is the realistic choice.

Type 3, quality judgement, needs information travelling with the call: *what should I come and look at?* A wrist receiver that only transmits numeric messages runs out of capacity here. Because the judgement result itself has to be recorded, a practical division of labour is to capture the input on the electronic forms system side and let the call system do one job only: “please come”.

Type 4, emergency and safety, should be designed as a separate channel from the other three. This is the one place where “the notification might not arrive” is not an acceptable risk. Treat it as a supplement rather than a replacement for existing emergency-stop and fire alarm systems, leaving the equipment required by regulation and internal safety standards fully in place. Confirm this with your own safety management function before you design anything.

Of the four types, type 1 produces by far the clearest return. In the model calculation later, the large majority of the monetary benefit comes from type 1. Types 2 and 3 make life on the floor noticeably easier, but the money attached to them is small. Understanding that asymmetry early keeps your investment sequence sensible.

Choosing the notification device: comparing three families

Notification targets fall into three families: indicator lights and display boards, wrist-worn receivers, and smartphones. In practice, most deployments combine them.

How this differs from handing everyone a two-way radio

Someone will always ask: “why not just issue intercoms instead?” It is a fair question — replacing shop-floor intercoms is a real use case for a call system. There are three differences.

First, with a radio you never know whether the person you called was listening. Voice passes and disappears; no record remains. Second, speech loses to noise. Near a press or an injection moulding machine, vibration beats audio for reliability. Third, broadcasting to everyone fragments attention. Listen to irrelevant calls all day and people learn to tune them out.

What radios do better is conversation. “I can’t come right now, give me ten minutes” is a reply that only voice handles well, which makes it telling that the academic framework cited above treats notification as two-way. A realistic design splits the work: the call system carries the call and the response record; the radio or the phone carries the negotiation.

Specification examples from real equipment (third-party products)

To make this concrete, here are the published specifications of two products sold for the Japanese domestic market. Neither is a TOMAS TECH product — both are examples from other manufacturers. Pricing is not publicly available to us, so this article does not discuss it.

Imao Corporation “Message Watch Type-N” (example of a wrist-worn receiver)

• Transmitter-to-receiver range is 270 m line of sight; transmitter-to-repeater is 1,000 m line of sight
• Three push buttons and three external input contacts
• Stores up to 16 call history entries
• Displays a numeric message on a wristwatch-style display and alerts by vibration
• Supports a multi-hop configuration in which a transmitter also acts as a repeater, extending range in roughly 1,000 m increments
• Groups can be configured per work area to avoid cross-talk

The specification worth pausing on is “stores up to 16 call history entries”. In a plant generating dozens of calls a day, on-device history alone will never be an improvement dataset. Device history and a server-side response log are two different things.

Herutu’s wireless call system (example of a push-button and indicator combination)

• Uses specified low-power radio in the 429MHz band
• Communication range is approximately 120 m outdoors and approximately 200 m indoors
• A repeater (TRV426M-A) extends communication range by a factor of 1.5 to 2
• Built from push-button transmitters, wall-mounted indicators, LED indicator lights, management software, and LAN-connected wireless receivers

Here the line to note is “429MHz band”. As the next section explains, that band is not among Thailand’s licence-exempt allocations. Even a configuration with a long and successful track record in Japan cannot be assumed to transfer to Thailand — and you can already read that off the specification sheet.

Why Japanese call equipment cannot simply be shipped to Thailand: radio rules and NBTC certification

This is the core of the article, and it is the part almost entirely absent from Japanese product pages. It is also exactly the sort of thing you want settled early when you run a plant in Thailand or elsewhere in ASEAN.

The Japanese side: specified low-power radio stations and technical conformity

The reason factory call equipment is so easy to deploy in Japan is a regulatory category called the specified low-power radio station. It is a type of licence-exempt station under Japan’s Radio Act: equipment meeting the conditions on application, frequency, antenna power and communication method can be used without an individual station licence. Equipment that has obtained construction design certification (commonly called *giteki*) can be built in and used without a station licence.

The representative bands are the 400MHz band (used for telemetry and telecontrol among other purposes, including the 429MHz band) and the 920MHz band. The Herutu product above uses the 429MHz band precisely because it sits inside that framework.

And every serious explanation of the Japanese regime carries the same caveat: taking equipment abroad requires case-by-case verification, because radio regulations differ from country to country. Skimming past that line and concluding “it works in Japan, so it will be fine” is the first trap.

The Thai side: the licence-exempt bands defined by NBTC

In Thailand, NBTC has added the 920–925MHz band as a licence-exempt band, and it is used for IoT and LoRa applications. The bands identified as usable without a licence are 920–925MHz and 2.4–2.5GHz.

Line those two up against the representative Japanese bands.

Three facts follow.

1. Japanese-market call equipment using the 429MHz band does not fall within Thailand’s licence-exempt bands.
2. Even 920MHz equipment cannot be assumed to transfer, because band edges, output limits and duty-cycle conditions differ by country.
3. In either case, NBTC certification is required in Thailand, and that process requires a local representative in Thailand.

To avoid any misreading: what can be stated here is that you cannot assume the equipment works as-is, and that a certification process applies. This is not enough information to declare any specific device or deployment unlawful. Verify the actual position for your own case with the equipment manufacturer and the certification body.

The technical conditions, in numbers

The standards relating to Thailand’s 920–925MHz band came into force on 24 November 2017.

The main technical conditions are as follows.

• Output ceiling: 4W e.i.r.p. for both RFID and non-RFID
• Reclassification for RFID: devices below 50mW EIRP moved from Class A authorisation to SDoC, a category exempt from permission for manufacture, possession, use, import and export
• Spectrum access for non-RFID devices — choose one of two methods:

– Duty cycle: 1% per hour below 50mW; 10% per hour from 50mW to 4W

– Frequency hopping: 0.4 seconds dwell time per channel within 8 seconds, with a maximum of 20 hopping channels

• Test references: ETSI EN 300 220-1 or FCC 15.247

The duty-cycle condition has direct operational consequences, so convert it into a unit you can reason about. An hour is 3,600 seconds:

1% per hour = 3,600 seconds × 1% = 36 seconds
10% per hour = 3,600 seconds × 10% = 360 seconds (6 minutes)

That is your entire transmit budget. For a call system — short packets sent at the moment a button is pressed — this is comfortably sufficient. But a design that continuously polls machines to keep pulling status data may not fit inside that envelope. If you are considering putting call functionality and condition monitoring on the same radio link, design the transmission frequency first.

The frequency-hopping option can be checked the same way:

0.4 seconds per channel ÷ an 8-second cycle = 5% of the cycle per channel
20 channels maximum × 0.4 seconds = 8 seconds = the entire cycle

So using the full 20 channels fills the 8-second cycle exactly.

Certification classes and the local representative requirement

IoT device certification is divided into Class A and Class B. Class A — high-output or high-sensitivity devices — requires testing at an accredited laboratory inside Thailand, and that laboratory must hold ISO/IEC 17025 accreditation.

And here is the clause that bites hardest in practice: the manufacturer must appoint a representative in Thailand. The representative must be a Thai national or a company registered in Thailand.

Translated into purchasing terms:

• “Import the device through a Japanese trading company and use it in our own plant” is not a straightforward path. Who acts as the representative in Thailand has to be settled first.
• Lead time swings enormously depending on whether the Japanese manufacturer already holds Thai certification. If they do, this is a purchasing conversation. If they do not, it is a testing and paperwork conversation.
• “We’ll just bring in one unit for a trial” is also not something to proceed with before checking how the rules treat it.

Other ASEAN countries, Vietnam included, have their own regimes and require country-by-country verification. This article can only present specific technical conditions for Thailand, but if you are planning to roll out one common configuration across several countries, build the schedule on the assumption that certification status must be confirmed separately in each one.

The practical workaround: Wi-Fi and cellular

If your reaction so far is that dedicated radio sounds like a lot of trouble, that reaction is correct. There is a way around it.

A configuration built on Wi-Fi (2.4GHz band) or the cellular connection of a smartphone avoids this dedicated-radio certification problem, because smartphones and Wi-Fi equipment sold in Thailand have already been certified for the Thai market. Put the notification on smartphones and pick up machine signals over wiring or the existing network, and you no longer need to import dedicated radio hardware at all.

But this is displacement, not elimination. The problem moves to your factory Wi-Fi design. Access point placement in a hall dense with metal machinery, reflection and interference, network separation between production equipment and notification traffic, availability during a power event — none of these are trivial. Settle the design of your factory wireless LAN and industrial network before you move on to the notification layer.

The communication options summarise as follows.

Designing past “it rings”: response logs, escalation and visible non-response

The most common failure in call system projects is spending months on device selection and then going live with no operating design at all. Making it ring is the easy part. The hard part is recording who responded, and when.

A call bell can only call. The improvement loop closes only when three things are recorded:

1. When the call was raised
2. When it was answered, and by whom
3. How many minutes recovery took

Only with all three do you get the two interval metrics that matter: average minutes from call to response, and average minutes from response to recovery. And the corrective actions for those two intervals are completely different.

Most “we installed a call system and stoppage time didn’t fall” conversations turn out to be discussions that never separated these two intervals, which is why nobody can identify the cause. Measure them separately and the right place to invest changes.

Escalation: automatically raising unanswered calls

Without a mechanism that automatically forwards unanswered calls to someone more senior, “it’s ringing and nobody is going” hardens into normal operation. In digital andon implementations, the intended model includes automatic escalation on non-response, together with a record of how the call was handled.

Escalation design comes down to four decisions.

• After how many minutes without a response does it escalate? Vary this by call type: tens of seconds for emergency and safety, a few minutes for material replenishment.
• Who does it escalate to? Sideways to another person at the same level, or upward to a supervisor? Going sideways first tends to produce faster movement on the floor.
• How many levels? Decide the terminus — for example, the plant manager at the third unanswered level — or escalation eventually runs out of targets and stops silently.
• Is the escalation itself recorded? The count of escalations is a direct indicator of whether your staffing plan is realistic.

One caution: escalation must not become a disciplinary instrument. The moment it is used to identify “slow responders” and correct them, the floor invents a workaround — press the acknowledge button immediately, walk over later. The records get cleaner and reality gets worse. Announce internally, before you start, that escalation data will be used for staffing and coverage design rather than individual evaluation.

Making non-response visible, and where the data goes

Once response logs accumulate, useful analysis becomes possible:

• Call volume by time of day (are calls clustering around breaks or on the night shift?)
• Call volume by machine (is a small number of machines producing most of the calls?)
• Response time by call type (is quality-judgement waiting time the only outlier?)
• Escalation rate (which time bands are genuinely short-staffed?)

To act on any of this, the call record has to be joinable with the rest of your factory data. To compare against machine utilisation data, feed it into the platform described in factory IoT implementation and equipment monitoring; to compare against production plans and process results, feed it to the process management system side. Put the other way round: keep the call system self-contained and its data dies in place.

Building that integration is Layer 4 of the cost model below. Cut it and the implementation cost falls — along with the improvement loop.

The five-layer cost breakdown: what to look for in a quotation

Quotations for call systems are structured differently by every vendor. To compare them, you have to normalise them into a common frame yourself. This article splits cost into five layers.

Fix the definition of “investment” before anything else

The investment figure in your capital request is the sum of Layers 1 through 4. Layer 5 belongs in a separate line as annual operating cost.

The reason is double counting. In the payback calculation, annual net benefit is annual benefit minus Layer 5. If Layer 5 is also folded into the investment figure, the same cost appears both in the numerator (investment) and as a deduction in the denominator (annual net benefit), and payback looks worse than it is.

Investment = Layer 1 + Layer 2 + Layer 3 + Layer 4
Annual net benefit = Annual benefit − Layer 5 (annual operating cost)
Payback period = Investment ÷ Annual net benefit
5-year ROI = (Annual net benefit × 5 − Investment) ÷ Investment

Put those four formulas in a footnote to the capital request and you will field far fewer questions from finance and from head office in Japan.

The costs most often missed in each layer

In Layer 1, the commonly missed item is extracting a contact from the existing signal tower. Some machines have no spare contact available and need one added. Check in advance whether the modification affects the machine manufacturer’s warranty conditions.

In Layer 2, the commonly missed item is spares. Wrist-worn receivers are worn on the body, so drops and water ingress cause failures. Buy exactly the number in service and every failure leaves one person unreachable.

In Layer 3, the commonly missed item is power. Repeaters need power. Mounted near the ceiling with no outlet nearby, they generate electrical work — and that work can cost more than the hardware. In a Wi-Fi configuration, the additional access points bring their own power and cabling checks.

In Layer 4, the commonly missed item is the configuration effort itself. Deciding “which colour on which machine reaches which group, in how many minutes, escalating how many levels” — for every machine — takes longer than people expect. Add an interface design exercise on top if you are connecting to an existing production management system or IoT platform.

In Layer 5, the commonly missed item is batteries. In a plant with dozens of push-button transmitters, battery replacement becomes a standing task. Unless you build into operations who replaces them and how a flat battery gets detected, the number of buttons that no longer ring quietly grows.

Model calculation: plant-wide rollout (A) versus bottleneck-first (B)

Now to concrete numbers. Everything below is a model calculation built on assumed values — not quoted prices and not measured results. Read it as a template to populate with your own figures.

Shared assumptions

Annual call volume, for reference: 120 calls/day × 312 days = 37,440 calls/year, of which 48 × 312 = 14,976 involve a machine stop and 72 × 312 = 22,464 do not.

A word on the 50% realisation rate. Of the time saved, only half is assumed to convert into money through added output or reduced overtime, because free machine capacity is worthless without orders to fill it. You will see model calculations that use 100%, but that embeds a strong assumption — that every saved minute becomes cash — and it will not survive questioning in a capital review.

The basis for THB 65/hour is as follows. Thailand’s daily minimum wage ranges from THB 337 to THB 400, with a national average of approximately THB 374 per day. The THB 400 daily rate applies in Bangkok, Phuket, Chachoengsao, Chonburi, Rayong, and Ko Samui District in Surat Thani Province. This model places the plant in an area where the THB 400 rate applies.

THB 400 ÷ 8 hours = THB 50/hour
THB 50/hour × 1.3 (employer contributions assumed at 30%) = THB 65/hour

The uplift covers employer-side social security and related contributions. On the regulatory side, from January 2026 the ceiling on the contribution base for social security (SSO) was raised from THB 15,000 to THB 17,500 per month — an increase of THB 2,500, or approximately 16.7%. The Employee Welfare Fund (EWF) is scheduled to start in October 2026. Employer-side labour cost is trending upward, which is worth holding in mind as background to the investment decision.

That said — and the numbers below make this plain — the labour-side benefit is not the main act in this investment.

Scenario A: plant-wide rollout (30 machines)

Investment (Layers 1–4)

Layer 5 (annual operating cost) is THB 90,000/year, and is not included in the investment figure. As a ratio to investment: 90,000 ÷ 1,680,000 = approximately 5%.

Look at the composition of the investment: Layer 1 is about 29%, Layer 2 about 21%, Layer 3 about 14%, and Layer 4 about 36%. The largest single block is not hardware — it is Layer 4, software and integration. Treat a call system as “buying some devices” and that 36% will come as a shock the moment the quotation lands.

Annual benefit

Benefit 1 (reduced machine downtime)
48 calls/day × 5 minutes × 312 days = 74,880 minutes = 1,248 hours
1,248 hours × 50% realisation rate = 624 hours
624 hours × THB 800 = THB 499,200/year

Benefit 2 (reduced time spent searching for people)
72 calls/day × 5 minutes × 312 days = 112,320 minutes = 1,872 hours
1,872 hours × 50% realisation rate = 936 hours
936 hours × THB 65 = THB 60,840/year

Total annual benefit = 499,200 + 60,840 = THB 560,040/year
Annual net benefit = 560,040 − 90,000 = THB 470,040/year

Payback period = 1,680,000 ÷ 470,040 = approximately 3.6 years
5-year ROI = (470,040 × 5 − 1,680,000) ÷ 1,680,000 = +39.9%

Now look at the split. Benefit 1 is THB 499,200 and Benefit 2 is THB 60,840, which makes Benefit 1 roughly 8 times Benefit 2 (499,200 ÷ 60,840 = approximately 8.2). Benefit 2 accounts for 60,840 ÷ 560,040 = about 11% of the total annual benefit.

This is the point worth stating loudly. Labour savings alone will not pay this investment back. At Thai daily minimum wage levels, cutting the time people spend hunting for each other produces a small number. What decides payback is reduced machine downtime — and even that only becomes money when the order book is full. Submit a capital request built entirely on “we will eliminate wasted walking” and the review will proceed without ever examining the downtime argument that carries nearly 90% of the value.

For reference, benefit per call is 560,040 ÷ 37,440 calls = approximately THB 15 per call. Carrying that unit figure in your head lets you size the impact quickly if actual call volume comes in below the assumption.

Scenario B: bottleneck process first (8 machines)

Instead of the whole plant, this scenario limits the first deployment to the 8 machines in the bottleneck process. Call volume arising in that process is assumed at 20 calls/day involving a machine stop and 24 calls/day not involving one.

Investment (Layers 1–4)

Layer 5 (annual operating cost) is THB 30,000/year. As a ratio to investment: 30,000 ÷ 479,000 = approximately 6%.

Composition here is about 28% Layer 1, about 24% Layer 2, about 17% Layer 3 and about 31% Layer 4. Layer 4 has fallen from the 36% it represented in Scenario A because this scenario assumes no integration with existing systems — starting with the standard package functionality only.

Annual benefit

Benefit 1
20 calls/day × 5 minutes × 312 days = 31,200 minutes = 520 hours
520 hours × 50% realisation rate = 260 hours
260 hours × THB 800 = THB 208,000/year

Benefit 2
24 calls/day × 5 minutes × 312 days = 37,440 minutes = 624 hours
624 hours × 50% realisation rate = 312 hours
312 hours × THB 65 = THB 20,280/year

Total annual benefit = 208,000 + 20,280 = THB 228,280/year
Annual net benefit = 228,280 − 30,000 = THB 198,280/year

Payback period = 479,000 ÷ 198,280 = approximately 2.4 years
5-year ROI = (198,280 × 5 − 479,000) ÷ 479,000 = +107.0%

The same asymmetry appears in Scenario B. Benefit 1 is roughly 10 times Benefit 2 (208,000 ÷ 20,280 = approximately 10.3), and Benefit 2 accounts for 20,280 ÷ 228,280 = about 9% of the annual benefit.

A versus B: investment rises 3.5×, net benefit only 2.4×

Side by side:

The ratios expose the structure:

Investment ratio: 1,680,000 ÷ 479,000 = approximately 3.5×
Annual net benefit ratio: 470,040 ÷ 198,280 = approximately 2.4×

Investment rises by roughly 3.5 times while annual net benefit rises by only about 2.4 times. That is why starting at the bottleneck process pays back faster. On 5-year ROI the gap is stark: +39.9% for A against +107.0% for B.

Two reasons drive this. First, calls do not scale proportionally with machine count. In this model, stop-related calls per machine run at 48 ÷ 30 = 1.6 per day in Scenario A but 20 ÷ 8 = 2.5 per day in Scenario B, reflecting the higher call density at the bottleneck. Second, Layer 4 integration cost does not shrink in proportion to unit count. Layer 4 is 600,000 in A against 150,000 in B — but only because B abandons integration with existing systems. Do the same integration in B and Layer 4 barely falls at all.

The evaluation horizon here is five years. Given device service life, replacement occurs at around the five-year mark, so a ten-year calculation is not used. A model that assumes wrist-worn receivers still in service after ten years loses credibility on that basis alone.

Sensitivity: what if downtime loss were half as large?

The single most influential assumption in this model is the THB 800 loss per hour of machine downtime. Replace it with THB 400.

Benefit 1: 624 hours × THB 400 = THB 249,600/year (half of the 499,200 at THB 800)
Total annual benefit: 249,600 + 60,840 = THB 310,440/year
Annual net benefit: 310,440 − 90,000 = THB 220,440/year
Payback period: 1,680,000 ÷ 220,440 = approximately 7.6 years
5-year ROI: (220,440 × 5 − 1,680,000) ÷ 1,680,000 = −34.4%

Payback stretches from about 3.6 years to about 7.6 years — four years longer. Annual net benefit falls from THB 470,040 to THB 220,440, roughly 47% of the original. The 5-year ROI turns negative, and under these conditions the capital request will not be approved.

Halving one assumption reverses the conclusion. The practical implication is unambiguous:

Until you can calculate your own loss per hour of machine downtime, you cannot make this investment decision.

Producing that single number matters more than any product comparison. The standard approach is to take the contribution margin of the target line (selling price minus variable cost) and divide it down to an hourly figure. If your order book is full and you are covering it with overtime and weekend shifts, an hour of downtime costs real money. If you have spare capacity and can recover any stoppage within plan, the loss should be estimated conservatively. Which also means the same plant can reach different conclusions in peak season and in slow season.

One more assumption to interrogate: the 5-minute reduction in response delay, derived from an assumed 8-minute average before and 3-minute average after. If you have never measured the “before” figure, that 5 minutes has no basis at all. Which is precisely why the 90-day plan below spends its first 30 days measuring the current state.

Six failure patterns

Six ways call system projects in Thai plants either stall or go live and then fall out of use.

Failure 1: planning around the same part numbers as the parent company in Japan

This is the most common entry point. Head office specifies “install what we use here”, a quotation is obtained on that basis, and the project stops at the import stage when radio certification surfaces. 429MHz equipment does not fall within Thailand’s licence-exempt bands, and even 920MHz equipment faces country-specific conditions. The fix is to ask the manufacturer about Thai certification status in the first week of the project — and to ask it precisely. Not “can we use this in Thailand?” but “do you hold NBTC certification, and if not, what happens regarding a local representative and testing?”

Failure 2: starting with a plant-wide rollout

As the A-versus-B comparison shows, investment rises by roughly 3.5 times while annual net benefit rises by only about 2.4 times. Plant-wide deployment pays back more slowly, and it is far harder to embed new operating rules across every machine at once. The fix is to start with a handful of machines at the bottleneck, measure the effect from the response log, and expand from there. Measured results from a first deployment are exactly what lets you replace assumptions with actuals in the next capital request.

Failure 3: stopping at “it rings”, with no response log

A call bell can only call. Without a response log, nobody can answer “how many minutes did we save?” after go-live — which means the next capital request fails and the system becomes a one-off purchase. Stored history on wrist-worn receivers is capped depending on the model; the third-party product cited earlier stores up to 16 entries. Confirm at quotation stage that device history and server-side response logging are treated as two separate things.

Failure 4: no escalation, so ringing becomes background noise

Without automatic forwarding of unanswered calls to someone more senior, “it’s ringing and nobody is going” becomes permanent. The insidious part is that this state is invisible: the calls themselves are logged, but if nothing tallies how many went unanswered, management never sees the problem. The fix is to put “unanswered call count” into the daily report from the first month of operation.

Failure 5: saturating the notification target

Design it as “we’ve paid for it, so send every abnormality to everyone” and wrists buzz dozens of times a day. Within days people stop checking. When emergency and safety calls get buried in that noise, it becomes a safety issue. The fix is to separate notification targets by call type. Broadcast to everyone only for emergency and safety; route everything else to the responsible group. Some equipment supports group configuration per work area to avoid cross-talk and misdirected notifications, so check this during design.

Failure 6: never budgeting Layer 5

The capital request carries only the initial cost, with no budget for battery replacement, failed-device swaps or rule revision effort. In a plant with dozens of push-button transmitters, dead batteries accumulate. After a few experiences of “I pressed it and nothing happened”, the floor stops using it. The fix is to state Layer 5 as annual operating cost in the capital request from the beginning. In the model calculation it is THB 90,000/year in Scenario A and THB 30,000/year in Scenario B — about 5% and about 6% of investment respectively. Losing credibility to conceal a number that small is a poor trade.

How to run the project: three phases over 90 days

Turning all of the above into something executable: 90 days, split into three phases of 30 days each.

Phase 1 (days 1–30): measure

The task in this phase is not selecting equipment. It is turning your current state into numbers.

• Inventory the calls: for one to two weeks, on paper or in a spreadsheet, record every call as it happens — time, machine, purpose, who responded, how many minutes until arrival. That gives you volume and response time for each of the four types.
• Calculate the downtime loss rate: divide the contribution margin of the target line down to an hourly figure. This is joint work with finance. As set out above, without this number no decision is possible.
• Verify radio certification: ask candidate manufacturers about their certification status for Thailand, whether they have a local representative, and lead time if certification is not yet held. Class A devices — high output or high sensitivity — require testing at an accredited laboratory in Thailand, so confirm the classification as well.
• If considering a Wi-Fi configuration: check current access point placement and radio conditions in the target area.

By day 30 you should be able to answer: how many calls per day, how many minutes on average, and how much money one hour of downtime costs.

Phase 2 (days 31–60): pilot

Deploy at the bottleneck process only. The scale of Scenario B — around 8 machines — is a realistic range.

• Configure notification targets and escalation: of the four types, start with just type 1 (equipment abnormality) and type 2 (material replenishment and transport). Adding quality judgement and emergency/safety is safer once operation has stabilised.
• Brief the floor: put “who presses what, and whose device rings” on a single diagram and post it. Material in Thai is essential. State clearly at this point that escalation records will not be used for individual evaluation.
• Read the response log daily: check four items every day — volume, average response time, unanswered count, escalation count. When something looks wrong, going to ask the floor the same day is what makes the system stick.
• Adjust as you go: escalation wait times, group divisions and button placement will all need revision in the first month. Build a schedule that assumes revision.

Phase 3 (days 61–90): decide

Rebuild the model using measured values from the pilot.

• Replace assumptions with actuals: swap the 5-minute response reduction and the 50% realisation rate for the figures your pilot produced. This is the single largest source of value in the 90-day plan.
• Finalise the five-layer cost: reflect the installation and configuration effort actually incurred, and rebuild the estimate for full rollout. The number swings heavily on whether Layer 4 integration is in scope, so settle that first.
• Decide the rollout scope: everything, or only the processes with high downtime loss? With measured data you can produce a payback figure per process.
• Write the capital request: line up investment (Layers 1–4), annual operating cost (Layer 5), annual benefit, annual net benefit, payback period and 5-year ROI, and attach the sensitivity case with the downtime loss rate halved. Presenting the sensitivity yourself reduces the questions you get in review.

FAQ: six questions that come up on the floor

Q1. Call bells or smartwatches — which should we choose?

It depends on the use case. A wrist-worn receiver wins where the responder covers a wide area and where vibration is needed because of noise. Indicator lights and display boards win where everyone present needs to know, and in outdoor locations with heavy dust or water exposure. In practice, plants usually combine both. As a starting point, route type 1 (equipment abnormality) to wrist-worn devices and type 4 (emergency and safety) to every channel including indicator lights, then work outward from there.

Q2. If we only push andon notifications to phones, is that cheaper?

Layers 1 and 2 get cheaper. Layer 4 does not. Extracting contacts from existing signal towers (Layer 1) and configuring call rules, escalation and response logging (Layer 4) require exactly the same work whether the notification lands on a phone or a wrist receiver. In Scenario A of the model, Layer 4 accounts for about 36% of the investment. Pitches built on “phones are cheap because you don’t buy hardware” usually leave this out.

Q3. Can we bring equipment we use in Japan straight to our plant in Thailand?

You cannot assume it will work as-is. The 429MHz band, representative of Japanese specified low-power radio, matches neither of Thailand’s licence-exempt allocations at 920–925MHz and 2.4–2.5GHz. Even 920MHz equipment faces country-specific differences in band edges, output and duty cycle. NBTC certification is required in Thailand, and the manufacturer must appoint a representative in Thailand — a Thai national or a company registered in Thailand. If you want to sidestep this, look at a configuration using Wi-Fi (2.4GHz) or smartphone cellular connectivity. In that case, factory Wi-Fi design becomes the new problem. Always confirm the position for your specific case with the manufacturer and the certification body.

Q4. Can we use this for forklift dispatch?

Yes. Transport requests fall under type 2, material replenishment and transport. Drivers are moving around the site, so a fixed display board will not reach them; a vibrating wrist receiver or a smartphone is realistic. Two design points. First, the call has to carry “where to” — with a receiver that shows only numeric messages, you need a posted mapping between zone numbers and call content. Second, do not design anything that requires operation while driving. For safety, write the rules on the basis that acknowledgement happens after the vehicle has stopped.

Q5. Can this replace our two-way radios?

Not completely. Radios support conversation; a call system fundamentally records “called” and “answered”. What the call system does better is three things: reliable delivery in noise, targeting only the people concerned, and leaving a record. The realistic design is a split: the call system carries the call and the response record; radios or phones carry the negotiation. Note also that a two-way notification framework covering both employee-to-supervisor and peer-to-peer exchange has been proposed in the research literature, so a design that does not lock notification into one direction is worth considering.

Q6. How many years until it pays back?

It depends on your own loss per hour of machine downtime. In this article’s model, at THB 800 per hour of downtime, payback is approximately 3.6 years for a plant-wide rollout and approximately 2.4 years for a bottleneck-first deployment. Change the downtime loss to THB 400 under otherwise identical conditions and the plant-wide payback stretches to approximately 7.6 years, with 5-year ROI turning negative at −34.4%. Rather than asking for the payback period first, produce your downtime loss rate first. Once you have it, payback is just a matter of putting your numbers into the formulas above. Use a five-year evaluation horizon, in line with the device replacement cycle.

Summary: three decision points

Three things to take away.

1. Verify the radio rules first. A plan built on using the same part numbers as the parent company in Japan may simply not be viable in Thailand. The 429MHz band is not among Thailand’s licence-exempt allocations, 920MHz equipment faces country-specific conditions, and either way NBTC certification and a representative in Thailand are required. Do this check in the first week of planning, not just before purchase. If you want to avoid dedicated radio entirely, a configuration on Wi-Fi (2.4GHz) or smartphone cellular is a genuine option.

2. Do not stop at “it rings”. A call bell can only call. Only when three things are recorded — when the call was raised, when and by whom it was answered, and how many minutes recovery took — does the improvement loop start turning. Without escalation on non-response, “it’s ringing and nobody is going” becomes permanent. And the response log becomes the measured data behind your next investment decision.

3. Produce the downtime loss rate first, then start at the bottleneck. Labour savings alone will not pay this back. In the model, the labour-side benefit (Benefit 2) accounts for only about 11% of annual benefit in Scenario A; the rest comes from reduced machine downtime. Change the loss per hour of downtime from THB 800 to THB 400 and payback stretches from about 3.6 years to about 7.6 years with a negative 5-year ROI. Since plant-wide deployment raises investment by roughly 3.5 times while raising annual net benefit by only about 2.4 times, going bottleneck-first to generate measured data moves both payback and internal agreement along faster.

Call systems are one of the lower-cost entry points in factory IT and OT investment. They are also one of the few places where the first thing you receive is data about what is actually happening on the floor, in the form of a response log. Treating them as the doorway to later investments in predictive maintenance and equipment monitoring is a sequence that makes sense.

TOMAS TECH is based in Bangkok and supports IT and OT implementation for Japanese-owned manufacturers in Thailand. On call and notification systems specifically, we can work through the practical steps with you in light of local conditions — checking radio certification, inventorying calls against the four types, organising cost across the five layers, and designing a pilot deployment. We are equally happy to talk at the earlier stage, before any product is on the table, when the question is simply which of your calls to tackle first. Not knowing your current call volume, or not knowing how to derive a downtime loss rate, is a perfectly normal place to start. Get in touch through our contact form.

References

Sources referenced in this article. Some are published in Japanese.

Andon and minor stoppages

• Definition of minor stoppages and the seven major losses: https://www.keyence.co.jp/ss/general/manufacture-tips/short-stop.jsp
• How andon works and its indicator colours: https://evort.jp/article/andon

Changing notification targets and digital andon

• https://innovaromorir.com/en/andon-system-manufacturing-guide-visual-efficiency/
• https://oxmaint.com/industries/manufacturing-plant/andon-system-manufacturing-real-time-alert-software

Academic literature

• “Smartwatch-Based Monitoring and Alert System for Factory Operators Using Public Cloud Services”, *Applied Sciences*, 15(5), 2806, published 5 March 2025, DOI: 10.3390/app15052806: https://www.mdpi.com/2076-3417/15/5/2806

Equipment specification examples (third-party products)

• Imao Corporation “Message Watch Type-N”: https://www.imao.co.jp/products/messagewatch-n.html
• Herutu wireless call system: https://www.herutu.co.jp/product/product.php?categid1=2

Japanese radio regulations

• Specified low-power radio stations: https://ja.wikipedia.org/wiki/特定小電力無線局
• Overview of specified low-power radio: https://www.kinryo-electric.co.jp/telecon/support/low-power-radio/
• JEITA document: https://home.jeita.or.jp/page_file/20160831112324_IEnda8v21P.pdf

Thai radio regulations (NBTC)

• NBTC 920–925MHz usage (RFID and non-RFID): https://360compliance.co/global-market-access/thailand-nbtc-updates-920-925mhz-usage-rfid-non-rfid/
• IoT device certification in Thailand and Vietnam: https://www.iotapproval.com/articles/navigating-iot-regulatory-certification-in-thailand-and-vietnam

Wearables market

• https://www.gminsights.com/industry-analysis/wearables-market
• https://www.indexbox.io/blog/industrial-wearable-market-driven-by-labor-shortages-and-aging-workforce-to-2035/

Thai labour cost and social security

• Minimum wage in Thailand: https://www.thailawonline.com/minimum-wage-in-thailand/
• Ministry of Labour announcement: https://www.mol.go.th/en/news/starting-july-1-ministry-of-labour-revises-minimum-wage-to-align-with-economy-and-improve-workers-quality-of-life