"Last month I guaranteed the customer the date would hold this time. It came in five days late again. Now when I call, the first thing they ask is, 'Is it real this time?'"
The person saying this was a sales manager. He laughed a little awkwardly when he finished. The production controller next to him said nothing and turned his laptop around. On the screen was an Excel file called "Delivery Tracking." The rightmost column was "Expected Completion Date." Several cells had been changed three or four times, and the traces were still there: the old date struck through, the new date written next to it, struck through again, written again. The owner, sitting at the head of the table, listened and said: "I'm thinking we pull delivery control out on its own and get a system to keep a tighter watch on it."
I'm Johnny Yang, Brightstream Technology in Taiwan. I've led my team through implementation for seven years, across 252+ projects in 32+ industries. I've seen this scene too many times, so I'll state the conclusion up front: A delivery date isn't promised. It's calculated. If you dare to tell a customer September 20 today, there has to be a formula behind it: when the materials will all be in, how long the work takes, and when the line gets around to this order. Those three parts map to three data sources: whether your inventory is accurate, whether your standard times are accurate, and how you calculate capacity. If even one of the three is a guess, the delivery date you calculate is a guess. It doesn't matter how accurate the other two are, because the guessed number drags the whole result off. Production delays become a chronic condition not because sales talks too loosely or production control isn't tough enough. It's because no one in the company holds the real version of these three data sources, so everyone uses their own inaccurate data to guess a date that sounds responsible and sends it out. The toughest salesperson can only shout an earlier date. He can't shout the materials in any faster.
First, you're right: sales quoting a date first is unavoidable
I know many owners' first reaction is "then tell sales to stop over-promising." That deserves a fair hearing first.
If sales doesn't give a date first, you don't get the order. The customer is comparing three suppliers at once. If you answer "let me check with production and get back to you," by the time you've asked and called back, the order has gone to someone else. So sales does something completely reasonable from where they sit: they pick a date from experience, and quietly pad a few days while they're at it. That isn't irresponsible. It's the best decision a person can make with incomplete information.
The real problem is that the buffer gets eaten all the way down the chain. Sales pads three days. Production control sees the date is tight and pushes the shop floor three days earlier. The shop floor knows production control always pushes, so they hold something back when reporting work. Everyone in the chain is fighting everyone else's inaccuracy, and not a single link is dealing with the actual question: what day this order should really ship.
On site I often run a test. I pick an order that's already late and ask sales, production control, and the floor supervisor separately: "Why is this one late?" I usually get three different answers that all sound reasonable. On the surface it looks like they're passing the blame. In reality, all three may be true at the same time. There's no shared record in the company to check against, so each person can only describe the part they can see.
Delivery dates are calculated: take the three data sources apart one by one
The formula is crude, crude enough that you can write it on a whiteboard without any system: the date you can start, plus the time to finish, plus the time waiting in line for the production line, equals the delivery date. Three terms, three data sources.
Data source one: inventory accuracy decides "when you can start"
This one gets skipped most often, because instinct says a delivery problem is a production problem. But look back at the formula. The first term is "when the materials will all be in," and that answer comes from inventory, not production.
Inventory goes wrong in two ways. The first is on the books but not on the shelf, and this one kills delivery dates outright: production control sees the system says there's enough stock and schedules the order, then on start day the warehouse can't find it. Emergency purchasing follows, the delivery date immediately slides one or two weeks, and the delay is completely outside the plan, with no time to renegotiate with the customer. The second is on the shelf but not on the books. It doesn't blow up on the spot, but it makes you buy material you don't need and ties up cash. Over time it breeds a very dangerous attitude: "The system numbers are wrong anyway, it's faster if I go look in the warehouse myself." Once the floor picks up that habit, whatever system you install later is hard to save, because everyone will route around it.
You don't have to wait for the year-end count to find out which one you have. Pick your ten most-used part numbers and go physically count them in the warehouse this week, then compare to the books. Why the differences happen usually points to the same few places: material issued but not logged, material returned but not logged, scrap thrown away with no one recording it, ad hoc borrowing. I wrote about how to trace these in more detail in "Your warehouse count never matches?". Here I'll only cover the link to delivery dates: every inventory error is a delivery date error, and the kind you can least recover from. If labor time was underestimated, you can still work overtime to catch up. If the material isn't there, overtime can't conjure it.
Data source two: standard time accuracy decides "how long the work takes"
The second term is manufacturing time, and its source is standard time.
In most companies, where this number came from is questionable. The usual story: when someone was working out a quotation, they needed a cost, so they asked a veteran machinist for an estimate and typed it in. From then on it lived in the system, for years. Machines were swapped, processes changed, a batch of new hires came in, and no one went back to touch it. You can test this right now: pick three part numbers at random, ask production control what the standard time is, then ask the machinist who actually does that step.
But more often than the number itself being wrong, whole segments are simply left out. Standard time usually only covers the time the machine is running. Between material going in and the finished goods going into stock, there's die changes, machine setup, first-piece inspection, waiting for QC, waiting for subcontractors to send parts back and forth, and rework. None of that appears on the time sheet, but all of it eats your delivery date. So you get a very typical pattern: standard times for every step add up to three days, but actual start-to-finish is twelve days. No one can say where the other nine days went, because they were never recorded. And the date you promised the customer was usually worked out from those three days.
Can you backfill using shop-floor work reports? In theory, yes, provided your work reporting is real. In many factories, reporting is entered all at once before clocking out, or caught up in one go at month-end to close the books, and neither the sequence nor the times match what actually happened. Reverse-engineering standard times from that data just locks the original errors in place, and makes them harder to challenge from then on, because now they're "backed by data." I went through a few ways to tell whether your own work reporting is usable in "Work reports are all filled in, but the numbers still can't be trusted?".
Data source three: how you calculate capacity decides "how long the wait is"
The third term is the most abstract, and of the three, the easiest to overestimate.
Most companies calculate capacity as number of machines times working hours. That looks great in a slide deck and doesn't survive a day on the floor. A few things actually eat capacity. People: the operator for that machine is off today, so the machine is there but the capacity isn't. Dies and fixtures: two orders need the same die, so the machine sits empty and still can't run. Inspection and rework: finished doesn't mean shippable. Whether the time stuck in inspection and rework counts as capacity is something you have to decide first, or that stretch turns into delivery time that vanishes into thin air.
There's one more that gets forgotten most easily: how many orders are already queued on the line right now. If you use theoretical capacity to say "this order takes three days" but don't count how many orders ahead of it have to finish first, those three days mean nothing. Most of the projects I've seen share one thing: the capacity number in the owner's head and what the floor can actually produce differ by a steady gap, and everyone from top to bottom knows it. That gap doesn't go away on its own. It turns into delivery delays, and every month it reappears at the review meeting as "we were busier this month."
| Data source | Which part of the delivery date it decides | Symptoms when it breaks | Check you can run this week |
|---|---|---|---|
| Inventory | When you can start | Shortage discovered on start day, emergency purchasing, delay with no warning | Physically count ten commonly used part numbers and compare to the system books |
| Standard time | How long the work takes | Steps add up to three days, actual is twelve, and no one can explain the difference | Pick a completed order and compare total standard time to actual days |
| Capacity | How long the wait is | Sales asks "can we take another one?" and everyone gives a different answer | Ask production control: how many days to finish every order currently queued |
When all three are guesses, here's what it grows into
Companies where all three data sources are inaccurate at once develop a very consistent course of illness. I see the same three stages almost every time.
Stage one is adding insurance. Deliveries are often late, so every link adds its own days: sales quotes a few extra days, production control pushes a few days earlier, purchasing sets lead times longer than they really are. The delivery dates you quote customers keep getting longer, you start losing to competitors at the quotation stage, and actual delays don't drop at all. None of the three causes of the delays has been touched.
Stage two is cutting in line to put out fires. Orders close to their due date get special attention, and the owner or the sales manager steps in to demand priority. That works in itself, but it's zero-sum: moving this order up means pushing several others back, and in a few days those pushed-back orders become rush orders too, so they cut in line again. At this point your schedule isn't planned anymore. It's argued out every morning, and whoever wins is usually whoever talks loudest.
Stage three is total distrust. No one looks at the expected completion dates in the system anymore, because everyone knows they're fake. Sales doesn't check the system for delivery dates; they call the supervisor they know. Production control doesn't look at system capacity when scheduling; they look at their own Excel. When the owner wants numbers, someone builds them by hand. The company has gone back to running entirely on personal relationships. There's just still a system sitting there on the surface.
The most expensive cost in this process isn't the money paid out for delays. It's the floor losing trust in "the numbers." Same group of people: three years later, anything data-related you try to push will be met with "that's what they said last time too."
So should you buy a system or not
Someone will ask: if all three data sources are inaccurate, won't buying a system make them accurate? It's the other way around.
What a system does is calculate the numbers you feed it faster and print them more neatly. If your inventory books are wrong, it will use wrong inventory to produce a very confident start date. If your standard times are estimates, it will use estimated times to produce a very professional-looking schedule. What you've bought is a faster guessing machine, and because the printout looks so official, it gets harder for anyone to say it's a guess.
So when should you buy? When your inventory books are close enough to reality, your standard times are maintained and cover changeovers and waiting, and your capacity definition subtracts people, dies, and inspection. At that point you'll hit a different kind of pain: too many constraints for a human to schedule. A dozen-plus machines, seven or eight process steps, a different routing for every order, and trade-offs between delivery dates and changeover costs. That kind of combinatorial explosion really is beyond Excel. The other case where you should buy: these three data sources exist only in the head of one senior production controller, and that person is about to retire. The purpose of implementation then isn't to calculate better than him. It's to force you to write the rules in his head down as parameters that can be checked.
So what should you do tomorrow
Don't hold another delivery review meeting. You've held that meeting many times. Everyone leaves in agreement, and next month nothing changes.
Tomorrow, do this instead: pick five recently late orders and, one by one, assign each delay to one of the three data sources. Did it start later than planned (inventory), did the work take longer than estimated once it started (standard time), or did it sit stuck in the queue (capacity)? Set aside any that don't fit, or that involve two or more. Five orders will take you about an afternoon, but you'll get something you have none of right now: a direction.
In most of the projects I've seen, a clear skew shows up: three or four of the five are stuck on the same data source. That's what you fix first. Leave the other two for now. Companies that fix all three at once usually finish none of them, because each one touches a different department and different people's habits, and no one can hold three fronts at once. If there's no clear skew, start with inventory, for a practical reason: it's the easiest to verify. Go count and you'll know. And it's the only one of the three where an error can't be recovered from.
If after sorting the five orders you find all three have problems and don't know where to start, or you're not sure your standard times are good enough to feed into any system, you can book 30 minutes with me and I'll go through the three questions with you one at a time.




