Why your production reporting data is fake: buddy punching, backfilling, and end-of-shift catch-up

The system is live and reports come in every day. But when you use those labor hours to calculate cost, you don't believe the numbers yourself. This post lays out the four ways production reporting data goes fake, the incentive structure behind it, and four ways to get the shop floor to tell the truth. It also says the thing system vendors won't write: after you implement the system, the data can still be fake.

Johnny Yang21 min read
Why your production reporting data is fake: buddy punching, backfilling, and end-of-shift catch-up

"We put in a production reporting system a long time ago. Data comes in every day. But when I use those labor hours to calculate cost, I don't believe the numbers myself."

I've heard this too many times, and the person saying it has usually already spent real money. He bought the system, the tablets, the barcode scanners. There's a reporting procedure taped to the wall. The database holds hundreds of thousands of reporting records, and it looks thorough and professional. But look closely and you find every entry on the same work order logged at exactly three hours, one person starting work on two machines fifty meters apart at the same moment, and not a single downtime or rework record for the entire month.

No factory goes a whole month without something going wrong. If the data says nothing went wrong, it means one thing: it isn't the shop floor that's error-free. It's the data that's lying to you.

I'm Johnny Yang, Brightstream Technology, Taiwan. I've led implementation teams for seven years, across 252+ projects in 32+ industries. Let me say this plainly: very few system vendors would write this article, because it contradicts their own pitch. The core selling point of what they sell is "see your real production status." Admit that the data is still fake after installation, and how do you close the sale? But I don't sell licenses. My part is the stretch where things actually get made to work. So I can say it.

Conclusion first.

Fake production reporting data is not a system problem, and it's not a character problem with your workers. It's an incentive problem. For a frontline operator, reporting honestly only hurts him: report an abnormality and he gets questioned, report his real hours and he looks slow, and the reporting screen is so awkward he has to pull off his gloves and tap five times. In that structure, no matter how expensive the system, all you get is "fake data, produced faster." There are only four ways out: tie the reporting action to something he already does, remove the penalty for reporting abnormalities, cross-check against traces left by the machines or the process itself, and most important of all, accept data that is coarser but true over data that is precise but made up. If the shop floor truly can't do it, admit that the number can't be used for decisions yet. Don't calculate cost from fake numbers. That is more dangerous than having no numbers.

Four ways production reporting data goes fake, and your shop floor has at least two

Before talking about fixes, you have to separate the problems. Each of these four kinds of fakery gets something different wrong, and each needs a different fix. If you hit all of them with one move, "tighter audits," you won't solve any of them.

Type one: everything filled in before clocking out, all the times wrong

This is the most common one, so common it's practically the default. The operator works a normal day and reports nothing. Fifteen minutes before the shift ends, he sits down at the tablet and fills in every work order he did today in one go.

The "quantity" he enters is usually right, because it's visible, countable, and the line supervisor checks it. But every time he enters is from memory, on the spot. Which order came first, which came later, how long things stopped in between, when the changeover happened — all of it is recalled. Human memory for time is unreliable to begin with, let alone for something eight hours ago.

The nastiest thing about this data is that the totals are right and total hours are roughly right, so your monthly report looks fine. But the moment you try to go one level down — which time slot output dropped in, how much time changeovers ate, how day shift compares with night shift — that whole layer is noise. You paid for real-time data. What you got is an electronic copy of a handwritten daily log.

Type two: reporting and clocking in for each other, the wrong person

The second type gets the person wrong. Worker A's work order gets scanned, swiped, or started by Worker B. The reasons usually sound reasonable: A's gloves are oily and he can't use the screen, A left his card in his locker, A got pulled to support another line, or the simplest one — B is faster, so one person swiping for three saves time.

The quantity and time in this data may be close to real, but the person is wrong. Once the person is wrong, two things are dead on arrival. One is individual efficiency analysis: the performance ranking you produce doesn't reflect what those people actually did. The other is piece-rate pay: if part of your pay is tied to reporting records, you are paying people based on wrong data.

Worse, once this becomes a habit, it turns into a private balancing system inside the crew. Whoever's output looks bad this month, everyone shifts a little to him. At that point the problem is no longer about convenience. It's organized behavior, and no system will block it.

Type three: only the good gets reported, and the problems disappear

The third type does the most damage. The operator reports everything he's supposed to — quantity right, time right, person right — except nothing bad ever makes it into the system.

Downtime not reported. Waiting for material not reported. Rework not reported. Die changes that ran over not reported. Minor machine faults he fixed himself not reported. Your database shows a clean production curve, and under that curve is a pile of holes.

So you get an absurd picture: the report says line utilization is good and there are almost no abnormalities, yet your deliveries keep slipping, inventory keeps piling up, and overtime keeps getting paid. Because every loss has been absorbed into "normal hours" — he stood waiting for material for forty minutes, and those forty minutes got counted as processing time on that work order. The result is an inflated standard time for that product, and everything you calculate from it, cost and quotations, is off.

Of the four, this one hurts business decisions the most, because it hides exactly where improvement is possible. You want to find the bottleneck, and the data tells you there isn't one.

The downstream damage usually blows up in delivery dates first. The schedule looks fine on paper, and deliveries keep missing anyway, because the schedule is built on these fake hours. I wrote about that chain of cause and effect separately: Deliveries keep slipping? Check whether these three data sources are accurate.

Type four: rounded numbers, every work order exactly three hours

The fourth is the easiest to spot: open the reporting records and the hours are all 0.5, 1, 2, 3. Messy decimals almost never appear.

Real production isn't that clean. Real hours look like one hour forty-seven minutes, two hours eight minutes — ugly numbers. When you see a whole column of neat round numbers, the shop floor isn't stable. Someone is filling in a form.

Rounding usually shows up together with type one — when you backfill before clocking out, you can't recall to the minute, so you drift toward round numbers. But it can also exist on its own. For example, the operator knows "nobody asks questions if I put three hours on a ticket," so every ticket gets three hours. This is filling in whatever keeps people off your back, and it's completely rational behavior.

Why do people fake it? Because honest reporting only hurts the shop floor

At this point, a lot of owners react with: "So my people aren't honest."

I'll be direct: No. Each of these four behaviors is rational. Rational enough that if you were in that seat, you'd probably do the same.

To fix this, you have to look honestly at the incentive structure first. When an operator reports honestly, what does he get?

First, report an abnormality and he's the one who gets questioned

This is the core of it. The floor reports "waiting for material, forty minutes." What happens next? Usually, at the next day's production meeting, someone gets called on to explain. And the person standing in the meeting room explaining is often not purchasing or production control, who caused the wait, but the crew that reported it.

As long as "reporting an abnormality" earns a follow-up question, a review, or a bad mark in your organization, reporting abnormalities is asking for trouble. You've designed a system that punishes honesty, and then you're surprised nobody is honest.

Second, report real hours and he looks slow

The second is just as real. If one person actually took two hours forty minutes to finish, and the next crew reports one hour fifty, and both numbers sit on the same report, what does the slow one do?

He has two choices: explain why he was slow — but usually nobody wants to hear it, and the report only shows numbers — or adjust the number down a bit. Guess which one he picks.

When labor-hour numbers are used to judge individuals, labor-hour numbers get distorted. It's close to a law. And no manager has to say it out loud. If someone got chewed out once in a meeting because their hours looked bad, the whole factory learned.

Third, the reporting screen is so awkward it's a punishment in itself

The third is the most overlooked, and has the biggest effect. Many reporting screens were designed by someone sitting in an office: pick department, then line, then work order, then process step, then operator, then enter quantity, then pick an abnormality code, then press confirm. Seven steps.

On the floor, the operator is wearing gloves, his hands are oily, the machine is running, and someone is waiting for him to step aside. You're asking him to do seven actions per ticket, and to find the right code out of fifty in a dropdown list.

Any reporting action that takes more than three or four seconds will naturally turn into "I'll do it all later" on the shop floor. That's not a discipline problem. It's a physics problem. You can't ask people to focus on production and also ask them to stop every ten minutes to operate an awkward screen.

This has the same root cause as what I wrote about in why ERP implementations fail: when the system was designed, people only thought about which fields the database needed, not about where the person filling in those fields is standing.

Four kinds of fake data, four fixes: start with this table

Matching each problem to its fix makes things much clearer. You can print this table and go through it with your plant manager, checking one type at a time which ones your floor has.

Type of distortionWhat's wrongDirect consequenceDirection of the fix
Everything filled in before clocking outTimes are wrong, order is wrongLabor-hour, time-slot, and changeover analysis all uselessTie reporting to actions already being done, so timestamps come from machines, not from people typing
Reporting and clocking in for each otherWrong personIndividual efficiency rankings distorted, piece-rate pay wrongSwitch to identification that needs no extra action; also check whether the process is forcing people to report for each other
Only the good gets reportedLosses vanishInflated standard times, skewed cost and quotations, bottlenecks invisibleRemove the penalty: abnormality reports trace the process, not the person, and cut abnormality codes to a number the floor can remember
Rounded numbersThe precision is fakeLooks detailed, has no analytical valueLower precision requirements to what the floor can actually deliver, cross-check the real range against machine traces

Once you understand this table, you'll notice something: none of the four fixes is "tighter audits" or "set penalties." Those two moves deal with "people not following orders," and the problem here was never people not following orders. It's a system punishing the people who do. Tighten audits and the fakery just gets more polished — people who were rounding to whole hours start making up decimals, so it's harder for you to catch.

Illustration of how hard each of the four types of production reporting distortion is to fix

Fixes one and two: make reporting easy, and remove the penalty

The first two fixes have to be discussed together, because one handles "can they do it" and the other handles "are they willing." Without both, neither works.

Tie reporting to something he already does

This is the most important sentence in the whole article: reporting shouldn't be extra work. It should be a byproduct of something he would do anyway.

What does a shop-floor worker definitely do every day? He picks up material, he gets a work order or drawing, he presses the machine's start button, he puts finished goods in a certain place, and he hands in the ticket when the job is done. These actions are part of production itself. Without them, the work doesn't happen.

So the reporting trigger should sit on top of those actions. Scan the code on the material rack when picking material, and you have the start time. Scan the pallet code when putting finished goods on the pallet, and you have the finish time. No extra actions for the worker, and the timestamps are real.

On the other hand, if your reporting is "after you finish, walk over to the tablet and do seven steps," it's extra work, and it will get postponed, batched, and rounded off.

In practice, doing this means walking through the full process with the shop floor, finding "the few moments where he stops anyway," and putting triggers only at those moments. You can't design this from an office. You have to stand on the floor and watch for half a day. That's why I keep saying the first step in digital transformation isn't picking a system. It's seeing clearly what actually happens on the floor.

Abnormality reports trace the process, not the person

The second direction is about policy, and only the owner can do it. Consultants and system vendors can't.

You have to announce publicly, and prove through behavior: reporting an abnormality won't get you questioned, but hiding one will.

How, concretely? Three things:

First, take abnormality records out of individual performance reviews. A crew that reports a lot of waiting for material isn't bad. It's honest. If your performance sheet counts abnormalities against people, you're paying for fake data.

Second, change how you ask about abnormalities in production meetings. Don't ask "why did it stop for these forty minutes." Ask "how many times did this kind of stop happen this month, and where are the causes concentrated." The first is looking for someone to blame. The second is looking for a process problem. The floor is very sensitive to this. Within three meetings they'll have decided which kind you are.

Third, cut the abnormality codes down to what the floor can remember. I've seen abnormality code lists with forty-plus items, finely divided and very logical, and the floor picks the first item every time. When the number of codes goes beyond what a person can react to instinctively, the options collapse into the default. Start with five to eight broad categories that are real and usable, and subdivide later if needed.

Fixes three and four: cross-check against shop-floor traces, and get true before you get detailed

Don't rely only on what people type; use the traces the process leaves

The third direction is technical, but it doesn't have to cost much.

Your shop floor is constantly producing evidence on its own; nobody is collecting it. Machine current or run signals, finished-goods receiving times, material requisition times, packing and shipping scan times, even door access logs and weigh-scale records — these are all traces that exist without anyone having to be honest.

The value of these traces isn't replacing reporting. It's that when the cross-check doesn't add up, you know what to ask. Someone reports three hours on a ticket, but the machine signal shows the machine was stopped for one of those hours. Where did that hour go? You're not trying to catch the person. You're trying to find out whether that hour was waiting for material, a die change, or a minor machine fault — which is exactly the improvement opportunity you couldn't see before.

One warning: don't try to connect every machine in the plant from day one. Pick the one or two most critical, most expensive, most bottleneck-like machines and start there. I covered the logic of this trade-off in how to calculate ERP implementation cost. The point is always to confirm first that you'll actually use what the money buys.

Accept data that is coarser but true

The last direction is the most counterintuitive: deliberately lower your precision requirements.

Many factories start by requiring reports accurate to the minute, broken down by every process station, with detailed abnormality causes. The floor can't do it, so they make it up. You get data precise to the minute, and every number in it is invented.

Try the other way: start by requiring only "morning / afternoon" or "this shift" level detail, but require that it be true. Coarse true data at least lets you see trends, ratios, and which line keeps falling behind. Detailed fake data lets you see nothing, and it misleads you into wrong decisions. It costs more than having no data.

Once the floor is used to it, reporting has become muscle memory, and everyone has confirmed that reporting abnormalities really doesn't get them yelled at, then subdivide layer by layer. This order can't be reversed.

What the data is for decides how precise it needs to be

One level deeper: you don't need all data to be equally precise. Precision requirements should be decided by use.

A few common uses, side by side:

  • Finding which line or machine is the bottleneck: you need relative comparison, and coarse is enough. Which station keeps backing up, which keeps waiting — shift-level data shows that.
  • Finding out why deliveries keep slipping: you need the distribution of "where the time went." What matters is whether abnormalities get recorded, not whether hours are accurate to the minute.
  • Calculating the true cost of a single product: this is where you need finer labor hours, and abnormality time has to be separable from processing time.
  • Running piece-rate pay: this needs the highest precision on "who." One wrong record means wrong pay, and it directly creates a trust problem.

See it? The problems most owners most want to solve — finding the bottleneck, finding out why deliveries slip — actually need the least precision. What needs high precision is cost and piece-rate pay, and those two should come later.

If you design reporting from the start to the standard of "must support accurate costing," you're making the hardest level the entry test. The floor can't pass it, starts making things up, and then even the simplest task, finding the bottleneck, is ruined along with it. It's a bit like the trap of forcing Excel to carry everything. The limits of managing with Excel is about the ceiling of the tool itself; this is about the requirement itself being set too high.

When you can't do it, be willing to say the number can't be used

Finally, something that doesn't work in my favor.

Some shop floors simply can't produce trustworthy reporting at this stage. Staffing may be too tight, products too varied, the type of work may not suit per-order reporting at all, or the organization may not have enough trust yet.

When that's the case, the right move is not to force a system in and pretend the data is real. The right move is to admit: this number can't be used for decisions yet.

Concretely, that means don't use these hours to calculate product cost, don't use them to adjust quotations, and don't use them to evaluate individuals. You can keep collecting, and use them to look at broad trends, but mark their reliability clearly on the report, so everyone reading it knows how much weight the number carries.

I know saying this hurts anyone trying to sell systems, because admitting the data can't be used means admitting the investment won't show results for a while. But adjusting quotations based on costs calculated from fake numbers costs real money, and you won't find out for months. Better to say honestly "this cell is empty for now" than to fill in a fake number that looks professional.

As for what kind of system can really make reporting easy and classify abnormalities automatically, that's a different level of question. Some things can be solved with process design; some genuinely need tools. I go into it more fully in what AI ERP can do. But the order is always: fix the incentive structure first, then talk about tools. Do it the other way around, and the tools just produce fake data faster.

FAQ

If reporting data is fake, will face recognition or fingerprint clock-in stop buddy punching?

It solves part of it, not the root cause. Biometrics do block the most direct form, "B swipes A's card," and that is their value. But first ask: why does the floor need buddy punching? If the answer is "swiping is a hassle, and one person swiping for everyone saves time," then after you install the devices, the floor will just evolve a new workaround — for example, everyone lining up to swipe once before clocking out, which makes backfilling worse. If the answer is "people often get pulled to support other lines on short notice, and the system has no way to report that," then what needs fixing is the process design, not the hardware. Spend a week collecting the real reasons for buddy punching before deciding. The wrong equipment becomes a new reason for the floor to resent the system.

Does it help to set penalties, like docking performance reviews for late reporting?

It works short-term, and long-term it makes the data harder to save. After penalties are set, the first thing the floor does is not report honestly, but make the records "look compliant." Those who backfilled everything before clocking out start backfilling at a few spread-out times. Those who rounded to three hours start rounding to two hours fifty-three minutes, so it doesn't look invented. The harder you audit, the harder the fake data is to spot. The bigger cost is trust: once reporting is defined as "something you get punished for not doing," it will never become "something useful to me," and the floor will never volunteer the problems the system didn't ask about. Penalties should only be the last line. The main tools should be making the action simple and removing whatever punishes honesty.

We run multi-skilled workers, one person tending several machines. How should a floor like this report?

This kind of floor is the worst fit for detailed per-order, per-person reporting. Force it and it will be distorted. If one person tends three machines and you ask him to split his hours across each machine and each order, he can only guess, and the guesses have no analytical value. There are two practical directions. First, change the reporting subject from "person" to "machine or work order" — record when this machine ran from and to, how long this order took, and for people only record "who was responsible for these machines this shift," with no hour allocation. Second, if you really need labor cost allocation, divide total shift hours and clearly mark it "this is an allocated value, not measured." The key is not to let allocated numbers mix with measured data, or later no one will be able to tell which numbers are real.

We already bought a reporting system but the data is fake. Replace the system or redo the process?

Redo the process first. Most of the time you don't need a new system. The test is simple: take your current reporting screen, put on gloves yourself, stand next to a machine, run one full report, and time it. If it takes more than three or four seconds, or you have to search through more than ten options, the problem is most likely in configuration and process design, not the system itself. Most systems let you adjust fields, required inputs, code lists, and triggers; it's just that during implementation nobody stayed with you to tune them until the floor could actually use them. On the other hand, if the system fundamentally doesn't support the kind of reporting you need, such as one person on multiple machines or scan-triggered reporting, that's a product limitation. Even if you do replace it, fix the process and incentives first, or you'll just get fake data in a different format.

Can coarse data really be used for decisions? Won't it mislead?

Yes, and it's much safer than detailed fake data, as long as you know what it can and can't answer. True shift-level data can answer trend questions like "which line keeps falling behind," "which time slot output drops in," and "are there more abnormalities this month than last," and those are exactly what most factories need to solve first. It can't answer "what's the unit cost" or "how much piece-rate pay does this person get," which need absolute precision. So the key isn't how precise the data is. It's whether you've drawn the boundary on what this data can be used for, and made sure everyone reading the report knows where that line is. Misleading never comes from data being coarse. It comes from using coarse data to answer questions it can't answer.

If you happen to be stuck here

If you finished reading and thought "our floor has all four," don't rush to blame anyone. The first thing to do is simple: pick a day, put on gloves yourself, stand next to a machine, and go through your current reporting process from start to finish. When you're done, you'll probably understand why the data looks the way it does.

Reporting is only one part of data distortion on the factory floor. The same set of floor problems has two other common forms. One is rush orders throwing the whole schedule into chaos, and how to schedule so you're not haggling every time, which I wrote about in Rush orders wrecking your factory schedule? Production scheduling in practice. The other is waiting for material — the downtime that shows up in reports often starts at the supplier, and that topic is Chasing suppliers for delivery dates? Turning material follow-up into a system.

If you want to get the direction clear before spending money, we have free tools and a case library you can look through, and our Podcast covers plenty of practical shop-floor issues. You can run through your own check against them first, and bring someone in when you actually need help. Honest coarse data can be refined over time. Fake data, however long you keep it, only grows into a better-looking lie.

Found this useful? Pass it on
30 minutes · free

Bring the one thing you are stuck on — a free 30-minute diagnostic

You do not need to know what system you want. Tell Johnny Yang about the one thing that keeps blocking you, and in 30 minutes you will know whether to skip it, build it with coaching, or pay for a custom build. No pitch — if we cannot help, you will hear that straight.

Johnny Yang replies himself · limited spots each month · or just book a free 30-minute diagnostic

Johnny Yang (熊董)
Johnny Yang (熊董)
Founder, Brightstream Technology

Serial founder leading a small team that builds systems into an owner's everyday work.

As a founder
7 yrs
Projects delivered
252+
Industries served
32+
About Johnny