Inspection Data Collection Cost and Payback | Four Gates in a Thai Factory

“Put in automated inspection data collection and the transcription work disappears.” That statement is true. But transcription disappearing and an inspection record being complete are two different things. What a measuring instrument sends to the system is a number and nothing else. What that number is, where on the part it was taken, when it was taken and who took it does not travel with it. This article uses a hypothetical model factory – a Japanese-owned metal machining plant in Chonburi, Thailand – to break down what actually pays back an initial investment of 1,366,000 THB, in a form you can follow on a calculator. The conclusion first: 77.4% of the return comes from somewhere other than transcription labor.

Connecting the instruments does not finish the inspection record

Connect a digital caliper by cable or wirelessly, and the moment the operator presses the send button the reading flies across. The screen shows “12.043”. That part works reliably. The problem is what comes next. The system has received the number 12.043 and nothing else. It holds no information at all that this is the outside diameter at measurement point 3 on the drawing, for part number A-1120, lot 2608-03.

On a handwritten inspection record, that information was guaranteed by the inspector’s head and by the format of the paper. The inspection sheet had the part number and the measurement points pre-printed on it, and the inspector wrote the reading into the matching cell. The act of choosing which cell to write in was the identification work. Automated collection removes the paper form. Once it is removed, the identification work does not disappear – it survives as an on-screen operation, “select the measurement point”, on a shop-floor terminal.

This is where the most easily overlooked trap in this investment sits. A transcription error can be found later by rechecking. The recorded number differs from what the instrument displayed, so comparing the two reveals it. A selection error is different, because the number itself is correct. If the value for measurement point 3 lands in the column for measurement point 5, it is still a real, correct measured value. Rechecking will not find it. Re-measuring will not find it. It gets found when the part will not assemble at the customer, or when an auditor lays the drawing and the record side by side.

This article is not about which instruments can be connected. The list of connectable instruments grows every year. What this article covers is what remains after you connect them – the mechanism that gives the numbers meaning, what it costs, and what comes back. Note also that the scope here is the automatic recording of measured values such as dimensions, weight and torque. Camera-based visual inspection equipment that judges pass or fail is a different animal in both design philosophy and cost structure. Selection criteria for that equipment are covered separately in how to choose visual inspection equipment.

The four gates inspection data has to pass, and how the initial cost splits

Between the number leaving the instrument and a submittable inspection record, there are four gates. Below are the initial costs estimated for the model factory, laid out gate by gate. Every figure is a hypothetical value.

The shares add up to 99.9% because each row is rounded to one decimal place.

Before the quotations arrive, most capital requests budget for Gate 1 only. “Fit wireless units to 12 calipers and install a receiver.” But Gate 1 is only 24.6% of the initial cost. The largest single item is Gate 2 at 30.7%, and Gate 2 plus Gate 3 comes to 680,000 THB, or 49.8% of the initial cost. In other words, roughly half of the initial spend goes not to moving numbers but to giving numbers meaning. If you do not share that split up front, the moment the real quotation lands you get the “this is not what we were told” conversation.

The gates are in series. If Gate 1 cannot capture the reading, nothing downstream runs – but passing Gate 1 alone does not produce a record either. Skip Gate 2 and you accumulate a list of timestamps and numbers. Skip Gate 3 and you accumulate records that never notice an out-of-specification result. Skip Gate 4 and you end up in the state where the correct data sits inside the system while the document you hand the customer is still built by hand in Excel. All three happen in practice.

One warning in advance. The 77.4% of the return shown later in this article comes from automatic judgment at the moment of capture plus an immediate alert – that is, from Gate 3. It might look as though you could therefore skip Gate 2 and buy only Gates 1 and 3 for less money. That does not work. Judgment means deciding which item and which measurement point’s upper and lower specification limits this number should be compared against, and building that linkage is Gate 2. Gate 3 without Gate 2 is a judgment engine with nothing to compare against. Understand that Effect B only arises once you have paid the 420,000 THB for Gate 2. The same applies to Gate 4. Skip it and you lose Effects C and D, and on top of that you get to explain to a customer auditor that the data is in the system but the submission is made in Excel.

Gate 1 capture — the physical entry point of a measuring instrument data collection system

The model factory has 28 measuring instruments: 12 digital calipers, 8 micrometers, 4 height gauges, 2 torque meters and 2 electronic balances. The first thing to check is whether each instrument has a data output function. The same model number is often sold in both a version with an output port and a version without, and you cannot know how many of the 28 units on your floor can output data until you physically turn each one over and look. A unit with no output will not accept a wireless module no matter what you buy. It has to be replaced.

If output exists, the next question is wired or wireless. Height gauges and electronic balances that sit fixed beside a surface plate are fine on a cable. Instruments that get carried over to the machine, like calipers and micrometers, should be wireless. Wireless transmitter units, of which the Mitutoyo U-WAVE is the best known example, use 2.4 GHz band radio such as Bluetooth. That band is licence-exempt in Thailand as well, so unlike the 920 to 925 MHz RFID band there is no need to rebuild the hardware country by country. This is one of the few areas where a configuration validated in Japan can be brought to Thailand unchanged.

The catch is that 2.4 GHz is the same band as the factory Wi-Fi, so it interferes with your access point channel plan. Instrument transmissions are intermittent and short, so real damage is uncommon – but once a reading is dropped, the inspector experiences “I pressed it and nothing arrived”, and after that happens a few times the inspector goes back to paper. Wireless design itself is covered in factory wireless LAN and industrial networks, but at minimum you should decide before installation to fix the channels of the access points in the measuring area and keep them clear of the channel used by the instrument receivers.

The number of receivers drives the cost directly. The published specification for U-WAVE is a maximum of 100 instruments registered per receiver and a maximum communication distance of 20 m. Looking only at the count of 28 instruments, one receiver would be plenty. What actually binds is the distance. If three machining lines stretch from one end of the building to the other, you are effectively laying out 20 m circles, and you need three or four receivers. A large part of the 336,000 THB for Gate 1 is not the transmitter units on the instruments but the receivers plus their mounting, cabling and power work. When you ask for a quotation, start from the coordinates of the places where measurement happens, not from the instrument count.

Gate 2 identification — where the information the instrument cannot send comes from

Gate 2 is the largest cost in this investment, and it is the part that vendor product articles almost never touch.

Every single measured value needs four attributes. First, what – the item and lot. Second, where – the measurement point ID. Third, when – the measurement time. Fourth, who – the inspector. Of these, the only one the instrument can realistically supply on its own is the third, the time. And since the time is normally stamped by the receiving side anyway, it matches reality better to assume the instrument supplies only the number itself.

This structure shows up plainly in the de facto industry standard. The Q-DAS DFQ format, widely used for quality data exchange, is built from numbered items called K-fields, and the layers are clearly separated. The K0001 series holds the measured value, the K1000 series holds the part, meaning the item, and the K2000 series holds the characteristic, meaning which dimension at which location. The instrument can supply only the K0001 series. The K1000 and K2000 series have to be prepared from somewhere else. The format’s own design assumes that the number and its meaning arrive from different sources. Gate 2 is exactly the work of deciding where the K1000 and K2000 series come from, how they are filled, and by whose action.

Do not make people choose – fix the measurement sequence

There are two ways to fill in the attributes. Let a person choose, or let the system instruct.

The choose-it-yourself approach puts an item list and a measurement point list on the terminal and the inspector taps to select. It is simple to build and cheap. But the more options there are, the more mis-selections occur. With 80 items and 16 measurement points per item, the combinations on screen multiply. And as noted above, a mis-selection cannot be caught downstream because the number is correct.

The other approach has the system instruct which point to measure next. The inspector selects the item and lot once at the start, then the screen calls out “measurement point 1, outside diameter”, “next, measurement point 2, inside diameter” in order, and the inspector simply measures the indicated feature and presses the instrument button. Once the measurement sequence is fixed, the only decision left to the inspector is whether to follow the current instruction or skip it. Removing the free-choice screen brings selection time down from 3 seconds per point to 0.5 seconds.

The calculations later in this article deliberately use the conservative 3 seconds. There are two reasons. First, a fixed sequence presumes that the item master and the measurement point master are complete, and exceptions always appear immediately after go-live. Second, if a capital request states a benefit figure that assumes 0.5 seconds and the number is not achieved, the credibility of the whole investment goes with it. If the payback works at 3 seconds, then whatever you gain by approaching 0.5 seconds can be reported as upside.

Keep it joinable with a machining condition record system

Looking one step further, the attributes you want tied to a measured value are not limited to inspection data. If you can also pull the machining conditions used to cut that part – machine number, cumulative parts machined since the last tool change, setpoints – then when an out-of-specification result appears you can trace back which conditions were in play when the drift started. The important point here is that you do not need to install a machining condition recording system at the same time. All that is needed is that the inspection data carries the lot, the machine number and the time correctly. With those three in place, a machining condition record can be added later as a separate system and still be joined. Conversely, if those fields sit empty while several years of data pile up, no amount of sophisticated analytics platform bolted on later will make the join possible.

The design of the linking keys themselves is covered in quality data management systems. Understand Gate 2 as the investment that fixes those keys at the instant of measurement.

Gate 3 judgment — the gate that blocks readings from an out-of-calibration instrument

Much of the 260,000 THB for Gate 3 goes into building the upper and lower specification limit master. For every item and every measurement point you hold the upper tolerance, the lower tolerance and the control limit. It looks like simple data entry from the drawings, but in practice you hit drawing revision history against physical parts, and it is not unusual for a factory to discover at this exact moment that the drawing in use on the floor is not the same as the controlled master drawing.

Once the specification limit master is in, judgment changes from something a person reviews afterwards into something that happens automatically the instant the number arrives. What needs designing here is who gets told, within how many seconds, and through which route. Does it sound on the line leader’s terminal, or go to the quality assurance group? If a control limit exceedance – in specification but trending – and an actual out-of-specification result travel by the same route, the thing rings so often that nobody looks at it any more. The 77.4% of the return described later depends on nothing except whether this routing design works. That is precisely why Gate 3, the cheapest of the four gates at 19.0% of the initial cost, generates most of the benefit. Put the other way round, if you buy the judgment function without deciding the alert recipients and who is accountable for responding, the 77.4% never materializes.

And there is one risk that exists only once you automate. Readings from an instrument whose calibration has expired enter the quality record without anyone laying eyes on them.

In the handwriting era, the inspector looked at the calibration label every time they picked up an instrument. Not as a conscious check – it is simply stuck where your eye falls when you grip the tool. If it had expired, they went and fetched another unit. That “eye” was never designed as a quality assurance control. It worked as a by-product of the way the work moved. Automated collection removes that eye. Press the button and the number flies, so there is less reason to look at the instrument at all. The number that flew becomes a legitimate inspection record inside the system, gets aggregated monthly, and appears on the inspection certificate submitted to the customer. Only six months later, when a customer audit compares the instrument register against the records, does it emerge that the records for that period were taken with an out-of-calibration instrument. And what is then in question is not one instrument. It is the validity of every record in that period.

The countermeasure is mechanically simple. Put the instrument register at the entrance to collection. The register only needs three fields – control number, calibration expiry date, and calibration certificate number. When a reading arrives from an instrument, look up the sending unit’s control number in the register, and if the calibration has expired, refuse the data. Not only refuse it, but display “this instrument is out of calibration” on the terminal and prompt the inspector to fetch a replacement. Set the warning to start 30 days before expiry and you can also plan when to pull instruments off the line for calibration.

Clause 7.1.5.2 of ISO 9001:2015 covers measurement traceability, requiring that measuring equipment be calibrated or verified, be identified, and be safeguarded against impairment of its status. A calibration expiry gate can be described as the implementation that moves this requirement from “human attention” to “the system’s front door”. In an audit, being able to explain that the register and the records are not two separate things – that a record only comes into existence by passing through the register – makes your audit position much stronger. The three-layer relationship of measuring, judging and correcting is handled with the same structure in temperature and humidity monitoring systems, where calibration is likewise the prerequisite for the first layer.

Gate 4 retention and submission — how far automated inspection certificate creation really goes

The 220,000 THB for Gate 4 is what it costs to get the accumulated data back out. Underestimate this and you land in the state where the correct data is in the system while the document that goes to the customer is still assembled by hand in Excel.

The easy things to automate are internal-format inspection records and your own standard inspection certificate. Specify item and lot, and out comes a form already populated with the measured values, the judgments, the measurement date and time, the inspector and the control number of the instrument used. The hard part is customer formats. Field names, ordering and unit notation differ per customer, and on top of that there is processing of the raw data, such as “the mean, maximum and minimum of a sample of 5”. The model factory’s estimate assumes building out the formats for its three main customers, and each additional fourth or fifth customer triggers extra cost. The honest thing to write in the capital request is “three main customer formats in initial scope, additional formats quoted separately per format”.

If your factory has already moved ahead with digitizing the forms themselves, you can reuse those output format assets directly. The issues on the digitizing-paper-forms side are collected in electronic forms systems and the paperless factory. Automated inspection data collection and electronic forms are separate investments, but sharing the exit point brings the Gate 4 cost down.

On the retention side you need a design that guarantees what an audit will ask about – that the record has not been altered. Make the measured value itself non-updatable, add a re-measurement as a new record, and leave an invalidation flag, a reason and the person responsible on the old record. With that approach the history is never lost.

Where inspection data collection pays back — shorter detection delay is 77.4%

Now to the core of it. Here is the breakdown of how much comes back per year at the model factory. Every figure is hypothetical and none of it is the actual result of a real company.

The measurement point count and hourly labor rate we assume

The model has three machining lines running 250 days a year. Measurement points are as follows.

That is 1,600 points a day, and 1,600 x 250 = 400,000 points a year.

The hourly labor rate is taken as 100 THB per hour. The minimum wage in Chonburi is 400 THB per day (Wage Committee Announcement No. 14, effective 1 July 2025, 17 tiers running from 337 to 400 THB nationwide), but inspectors are assumed here to be paid not at minimum wage but at the equivalent of 18,000 THB per month. Dividing that by contracted hours of 173 per month (equivalent to a 40-hour week) gives roughly 104 THB per hour. Tidying up the remainder, the calculations from here on use a rounded 100 THB per hour. Actual rates differ by factory, so substitute your own figure as you read.

Effect A – reducing transcription and cross-checking labor

Here is the labor per point, today against after automated collection.

Note that those 8 seconds survive automated collection. The physical act of writing goes away, but this figure includes the time to read the instrument display and judge for an instant whether the value is plausible, and inspectors do not stop making that judgment just because collection is automated. What disappears is 6 seconds of transcription plus 4 seconds of cross-checking, 10 seconds in total, offset by 3 seconds added for the selection operation. Net 7 seconds per point.

Today’s annual labor is 400,000 points x 18 sec = 7,200,000 sec = 2,000 hours, which is 200,000 THB per year. After automated collection it is 400,000 x 11 sec = 4,400,000 sec = 1,222.2 hours, multiplied by 100 THB gives 122,222 THB, rounded to the nearest hundred baht 122,200 THB per year. The difference is 77,800 THB per year. You can confirm it by another route: 400,000 x 7 sec = 2,800,000 sec = 777.8 hours, multiplied by 100 THB gives 77,778 THB, rounded to the nearest hundred baht 77,800 THB. That is Effect A.

Work that cost 200,000 THB a year now costs 122,200 THB. Not a bad number, but set against the 1,366,000 THB initial cost and 168,000 THB annual running cost that follow, it recovers nothing on its own.

Effect B – shrinking the detection delay

At the model factory there are 24 process abnormalities a year, meaning 24 out-of-specification occurrences. From the moment sampling inspection produces an out-of-specification result to the moment quality assurance becomes aware of it currently takes 6.5 hours. The reason is structural. The inspector records on paper, transcribes into Excel in a batch, and quality assurance reviews that file once a day. With automated collection judging against the specification limits at the point of capture and alerting on an out-of-specification result on the spot, that becomes 0.2 hours. The reduction is 6.3 hours.

The equipment in question – that one machine producing the abnormality – runs at 120 pieces per hour. 6.3 hours x 120 pieces = 756 pieces. That is the extra quantity produced per occurrence purely because detection was late. Read it as the output of the one machine that should have been stopped and kept running, not as the output of the whole factory.

Of the 24 occurrences a year, and taking a three-year historical average as the assumption, 2 a year escape to the customer and 22 a year are contained internally. These two behave differently, so they are handled separately.

For the 2 escapes a year, the cost of dealing with an escape is put at 161,000 THB per occurrence. That breaks down as sorting personnel dispatched to the customer’s premises, 3 people x 3 days at 72,000 THB, recovery transport at 25,000 THB, preparation of the corrective action report at 40 hours x 100 THB for 4,000 THB, and 60,000 THB of price reduction associated with a concession, which is the practice of delivering out-of-specification parts with the customer’s approval. Total 161,000 THB. The 72,000 THB for the sorting dispatch works out at 8,000 THB per person-day, an order of magnitude away from the internal rate of 100 THB per hour, because it is the rate for outsourced sorting including travel, accommodation and emergency response. If detection happens within the same shift the parts are stopped before shipment, so the assumption is that 2 occurrences a year fall to 0.5 a year. That is a reduction of 1.5 occurrences, and 161,000 x 1.5 = 241,500 THB. Call this B1.

For the 22 occurrences that did not escape, 756 pieces x 22 occurrences = 16,632 pieces are identified only after they have already been made. Assuming 15% of those were genuinely out of specification gives 2,495 pieces. The 15% assumption reflects abnormalities of the gradual-drift type, such as tool wear, where machining conditions move slowly out of position. Parts made just after the drift starts are still inside tolerance, and only in the later part of the period do they break the specification. For an abnormality where every part suddenly goes out of specification at a single point in time, the ratio would be considerably higher. Multiplying by a manufacturing cost of 85 THB gives 212,075 THB. This is the reduction in over-production and scrap, and it is B2. The remaining 14,137 pieces turn out to be good, but since they are under suspicion they all have to be sorted. At 25 seconds per piece that is 353,425 seconds, or 98.2 hours, multiplied by 100 THB gives 9,820 THB. That is B3.

B = 241,500 + 212,075 + 9,820 = 463,395 THB per year.

Effects C and D – document preparation and document hunting

Effect C is the preparation of inspection certificates and customer submission documents. Assuming that takes 40 hours a month today, that is 480 hours a year, or 48,000 THB per year. Effect D is hunting for documents during customer audits and corrective action reporting, 12 times a year x 8 hours = 96 hours, or 9,600 THB per year. Both are easy for the shop floor to recognize and easy to write into a capital request.

Total return and its composition

The shares add up as 13.0 + 77.4 + 8.0 + 1.6 = 100.0%. So 77.4% of the return comes from cutting the volume built while nobody had noticed. Transcription labor is only 13.0%.

That split determines how you write the capital request. What you are recovering is not transcription time but the time it takes to notice an out-of-specification result. A request that writes only labor savings in the benefit column cannot even cover the annual running cost, as the next section’s arithmetic shows.

Cost and payback — what is inside the 3.2 year figure

Here is the cost side. Initial 1,366,000 THB, annual running cost 168,000 THB. The running cost is 96,000 THB of maintenance and licences plus 72,000 THB of master data upkeep. Budgeting for master data upkeep matters, because new items, specification updates following drawing revisions and additional measurement points occur every year. Estimate that line at zero and from year two onwards the master drifts away from physical reality, and the judgments stop being believed.

Annual net benefit is 598,795 − 168,000 = 430,795 THB. Simple payback is 1,366,000 ÷ 430,795 = approximately 3.2 years, or about 38 months.

How you judge 3.2 years depends on your factory’s investment criteria. In a factory with a three-year payback hurdle for capital equipment, this number does not clear. Sometimes it is right that it does not clear. Read the sensitivity section before deciding.

It is worth noting that conditions around manufacturing in Thailand are far from easy. The Office of Industrial Economics under the Ministry of Industry reported a Manufacturing Production Index for June 2026 of 94.99, down 3.10% year on year, with capacity utilization at 57.61%. For the second quarter the index was 95.96, down 1.79% year on year, with average utilization of 57.47%. Getting a payback of more than three years approved in a period where utilization is below 60% rests on the benefit breakdown surviving a recalculation.

Sensitivity — the factories this investment does not suit

Change the assumptions and the payback moves a long way. The important discipline here is not to apply a single blanket discount factor to every benefit. When an assumption changes, only the items it actually touches move. Here are three cases.

Case 1 is a factory that has never had an escape to a customer. The 241,500 THB of B1 vanishes entirely and the total benefit becomes 598,795 − 241,500 = 357,295 THB. Net benefit is 189,295 THB and payback is 1,366,000 ÷ 189,295 = approximately 7.2 years. B2 and B3 are about quantity built internally, so they survive regardless of escape history. That is why the figure does not fall to zero.

Case 2 is a factory where detection is already fast. Assume an operating practice where quality assurance reviews sampling results on the spot, so detection takes only 1.5 hours. The gap against 0.2 hours after automation is 1.3 hours, so the reduction shrinks from 6.3 hours to 1.3 hours. Only the items derived from detection delay move. B1 falls to 40%, or 96,600 THB, because the probability of stopping an escape drops. B2 and B3 scale with the reduced time, so multiplying by 1.3 ÷ 6.3 gives 43,762 THB and 2,026 THB, rounded to the nearest hundred baht and nearest ten baht respectively as 43,800 THB and 2,030 THB. Total B becomes 142,430 THB. A, C and D have nothing to do with detection time, so they do not move. Total benefit is 77,800 + 142,430 + 48,000 + 9,600 = 277,830 THB, net benefit 109,830 THB, and payback approximately 12.4 years.

Case 3 is not a change of assumption but a problem with how the capital request is written. List only A, C and D in the benefit column and the total is 77,800 + 48,000 + 9,600 = 135,400 THB. That is below the annual running cost of 168,000 THB, so the net benefit is −32,600 THB. Not only does it fail to recover a single baht of the initial cost, it accumulates a loss of 32,600 THB every year. The inequality 135,400 < 168,000 is the single most practical line in this article. A request that lists only labor savings and document preparation falls the moment the approver reaches for a calculator. And it deserves to fall.

What Case 1 shows should be stated bluntly. The less escape history a factory has, the further away the payback is. Automated inspection data collection is not the kind of investment a factory with no problem installs just in case. The numbers stand up at a factory that has caused a customer trouble because an out-of-specification result was caught too late, or that has actually had to run 100% sorting. Conversely, if a factory where detection is already fast wants to install it, the decision should rest on some other reason – inability to recruit people, rising audit requirements – and not on payback years.

Three more things to settle for a factory in Thailand

When you bring a configuration validated at head office in Japan over to Thailand, there are three additional things to settle.

First, the calibration traceability chain. Customer audits ask how far back a measured value can be traced. Thailand’s national metrology institute is NIMT, established on 1 June 1998 under the National Metrology System Development Act B.E. 2540. NIMT’s own calibration laboratories operate under ISO/IEC 17025, and many of its calibration services are listed under the CIPM-MRA, the mutual recognition arrangement of the International Committee for Weights and Measures. Certificates you receive from a commercial calibration laboratory, on the other hand, are internationally accepted on a different basis – the ILAC-MRA, which is the mutual recognition arrangement between accreditation bodies. Do not mix up these two tiers. The chain you need to be able to explain has four nodes joined by three arrows: your factory’s instrument, then an ISO/IEC 17025 accredited calibration laboratory, then NIMT, then the SI units. Accredited laboratories inside Thailand are listed publicly by TISI on the basis of TIS 17025, the Thai Industrial Standard equivalent. Alongside that, decide at the equipment relocation stage where you will send instruments brought from Japan once their JCSS certificates – issued under the calibration laboratory registration scheme of Japan’s Measurement Act – expire. Factories really do end up in the situation of “nobody knows where to send it” after relocation and keep using instruments past their expiry date.

Second, the radio band. As noted above, the 2.4 GHz band is licence-exempt so no country-specific redesign is needed, but it shares the band with the factory Wi-Fi. The published specification of up to 100 instruments registered per receiver and a maximum communication distance of 20 m determines how many receivers you install, and therefore the Gate 1 cost. Plot the measuring locations on the building layout drawing, draw 20 m circles, and decide the receiver count from that.

Third, the language of measurement point names. If the Gate 2 measurement point master exists only in Japanese, local inspectors cannot select from it. This is the single biggest cause of selection errors. Hold item names and measurement point names in three columns – Japanese, English and Thai – and display the local language plus the balloon number from the drawing on the terminal. Screens for Japanese managers can show the Japanese column, and the same data satisfies both. Adding a column to the master after go-live means translating every item in the catalogue. Building three columns into the initial master design is by far the cheapest solution.

Where to start with inspection record automation

For the order of implementation, start with the parts that cost nothing.

First, count your current measurement points. For each inspection type, physically count the number of points measured per day. The 1,600 points per day at the top of this article is a hypothetical value, and Effect A cannot be calculated until you substitute your own number. Next, actually measure the time from an out-of-specification result appearing to quality assurance becoming aware of it, using the most recent handful of cases. The 6.5 hours in this article is hypothetical too. With just those two measured values, the skeleton of Effect A and Effect B can be rebuilt using your own numbers.

On that basis, check the output capability of all 28 instruments one unit at a time. How many units have no output changes the Gate 1 cost substantially. At the same time, check the current state of the instrument register – whether the control numbers and calibration expiry dates are up to date. Turn on the Gate 3 expiry check while the register does not match reality and it will reject perfectly good instruments from day one.

Starting with a narrow scope is the rational choice. Take one of the three lines and a handful of items for a major customer. Run the whole chain from Gate 1 to Gate 4 once, all the way through to an inspection certificate actually coming out. Most of the problems that surface will be concentrated in Gate 2, that is, in the master data and the selection operation. Expanding to all lines comes after that.

One word about timing. The revised edition of ISO 9001 went to FDIS ballot in mid-April 2026, with publication expected in September 2026. How the requirements on monitoring and measuring resources will change in concrete terms cannot be confirmed from primary sources at this time, so this article will not set out clause-by-clause changes. However, standards revisions are customarily followed by a transition period after publication, so if you intend to review how your records are built, there is a reasonable case for putting it on this period’s agenda. Getting the master data and the instrument register in order first, rather than waiting for the revision to be finalized before starting design work, will not be wasted either way.

Frequently asked questions

Where should we start with automated inspection data collection?

Not with instrument selection and not with product comparison. Start with two measurements. One is the number of measurement points per day for each inspection type. The other is the actual measured time from an out-of-specification result appearing to quality assurance becoming aware of it. Without those two, the benefit figure is not your own factory’s figure. The second one in particular is the variable that decides 77.4% of this investment’s return. At a factory already running at 1.5 hours’ detection time, the arithmetic puts payback beyond 12 years, and the investment decision itself changes.

Can automatic data recording work with any measuring instrument?

Only with units that have a data output function. The same model number is frequently split into different part numbers depending on whether an output port is fitted, so each physical unit has to be checked. A unit with no output cannot take a wireless module, so the choice is to replace it or to leave manual entry in place for that instrument only. When counting units before installation, count individual units, not model numbers.

Should caliper data transfer be wired or wireless?

It depends on where the instrument is used. If it stays fixed beside a surface plate, wired is enough, cheaper, and drops nothing. If it is carried over to the machine, use wireless. Wireless uses the 2.4 GHz band and one receiver has a maximum communication distance of 20 m, so if the measuring locations are scattered around the building the receiver count goes up. Estimating that count depends on the positions where measurement takes place, not on the number of instruments.

Can weighing scale data integration use the same mechanism as other instruments?

The capture mechanism is the same, but there is one thing to watch. Electronic balances and torque meters are frequently used in 100% inspection processes, where the point count is an order of magnitude larger. Even in this article’s model, 1,120 of the 1,600 points per day come from 100% inspection processes. In a process with that many points, making a person select the measurement point is not realistic. Fix the measurement sequence to the process order of the equipment and design the selection operation out entirely.

Can digitized quality records replace paper inspection records as they are?

Not as they are. A paper inspection sheet guaranteed which value belonged to which location through its pre-printed part number and measurement point columns. Remove the paper and that guarantee has to be replaced by either an on-screen selection operation or a system-driven measurement sequence instruction. Remove the paper without designing that replacement and what accumulates is a list of numbers. Gate 2 accounts for 30.7% of the initial cost precisely because of this replacement.

Where should we send instruments for calibration at a factory in Thailand?

To a calibration laboratory accredited to ISO/IEC 17025. The list of accredited testing and calibration laboratories inside Thailand is published by TISI on the basis of TIS 17025, and you select from it a laboratory whose measurement quantities and ranges match yours. Certificates from an accredited laboratory are internationally recognized under the ILAC-MRA, so they are accepted in customer audits. International equivalence on the national standards side is underwritten by NIMT through the CIPM-MRA. Decide where to send instruments brought from Japan once their JCSS certificates expire at the equipment relocation planning stage.

Summary

Here are the points to hold on to when deciding on an automated inspection data collection investment.

• What the instrument sends is a number only. What, where, when and who measured it does not come with it
• Of the 1,366,000 THB initial cost, Gate 1 capture is 24.6%. Identification plus judgment, at 49.8%, is the cost of giving the numbers meaning
• A selection error carries a correct number, so neither rechecking nor re-measuring finds it. That is why you fix the measurement sequence instead of making people choose
• Automation removes the eye that used to read the calibration label. Put the instrument register at the entrance to collection and refuse data from expired instruments
• Of the 598,795 THB annual benefit, 77.4% comes from shorter detection delay. Transcription labor is 13.0%
• Simple payback is about 3.2 years. But with no escape history it is about 7.2 years, and with detection already at 1.5 hours it is about 12.4 years
• A capital request listing only labor and document preparation totals 135,400 THB, below the 168,000 THB annual running cost, and never pays back

Every figure in this article is a hypothetical value based on a Japanese-owned metal machining plant in Chonburi, Thailand. Substitute your own measurement point count and detection time and you can produce your own payback period from the same formulas.

If you are at the stage of wanting to count your measurement points, or to establish the actual time it takes to detect an out-of-specification result, we are happy to talk. TOMAS TECH supports Japanese manufacturers in Thailand starting from a stocktake of the shop-floor instrument register and the inspection process. Getting your current state into numbers is a perfectly good place to stop, so please get in touch through our contact form.

References

• Office of Industrial Economics, Ministry of Industry, Thailand – Manufacturing Production Index for June 2026 at 94.99, down 3.10% year on year, capacity utilization 57.61%, second quarter at 95.96, down 1.79%, average utilization 57.47%
• NIMT – About the National Institute of Metrology Thailand
• TISI – List of accredited testing and calibration laboratories under TIS 17025
• AQDEF K-field list – the layer structure of the Q-DAS DFQ format
• Mitutoyo U-WAVE wireless measurement data communication system
• DQS – ISO 9001 revision at a glance, FDIS balloted mid-April 2026 with publication expected September 2026
• Minimum wage in Thailand – Wage Committee Announcement No. 14, effective 1 July 2025, 337 to 400 THB per day across 17 tiers