Why do custom order mistakes get past a review?

If you order custom work from a supplier, you already know there are no returns. When windows, a cabinet run or a printed sign show up built to the wrong spec, you pay to have them made again, and your customer waits on a job you thought was finished.

What makes these mistakes hard to catch is that the order usually looks fine. Every field on the supplier’s form is filled in, every option on it is a real option, and the supplier builds exactly what you asked for. The problem tends to be one choice left on the normal setting when this particular job needed something else, and a review that checks whether the order is complete won’t see it.

I’ll cover why a complete order can still be wrong, why “double-check it” doesn’t catch the problem, what to check the order against instead, where AI helps and where it doesn’t, and when the review you already have is enough.

Why does a wrong custom order look fine on paper?

A wrong custom order looks fine because the mistake is usually a choice the form considers perfectly valid, just not for this job.

Say most of your window jobs use standard glass, and then one job has a window right beside a bathtub. In most of the U.S., the residential code treats glass within five feet of a tub, with its bottom edge less than five feet off the floor, as a hazardous location. That usually means safety glass, typically tempered, though local codes vary. (It’s section R308.4.5 of the International Residential Code if you want to look it up.)

The person placing the order has filled out that form a hundred times, and standard glass is what they always pick. Nothing on the screen tells them this window is different, so the order goes out as standard. Nobody did anything careless along the way. They checked that the order was complete, and it was. What they didn’t have in front of them was the one detail that made this job unlike the last hundred.

Why doesn’t “double-check the order” catch it?

I don’t think double-checking does much on its own, because the person checking needs something to compare the order against. Without it, they compare the order to their sense of a normal job, and an order full of normal options will usually pass.

This is an easy habit to fall into, and it isn’t a sign of a careless team. Most of the time the normal options are correct, so a review that looks for blanks and obvious typos feels thorough. The trouble is that the expensive mistakes often sit in the places where the normal answer is wrong.

What should you check the order against?

Check it against a short list of the ways this job is different from your usual one, and write that list down before the order goes in.

For most jobs that list is two or three lines long, things like a glass type, a finish color, an unusual size or which side the door hinges on. Put the list next to the order and go down it line by line. If the list says tempered, the order should say tempered, and if something on the list isn’t on the order, stop and fix it before you submit. If you aren’t sure what the customer agreed to, ask now, while asking is still cheap.

I’d keep the list short on purpose. A sheet that covers everything about the job turns into one more form people skim, while a list of only what’s unusual tells the person checking exactly where to look.

The list also makes the work easier to hand off. Someone new to the job can’t yet tell which parts of an order are unusual, because they haven’t seen enough jobs to know what normal looks like, and the list tells them.

Where does AI help with custom orders?

AI helps most at the start, while you’re still pulling the list together from everything the customer told you.

What a customer wants is usually spread across the quote, the contract, measurement notes, a few texts and a phone call where they changed their mind about the color. Pulling all of that together is usually the hard part of making the list, and it’s the kind of reading AI is good at. You can hand it a call transcript or a pile of notes and ask it to pull out anything that sounds like a requirement or a change, and it can find the spot forty minutes into a call where the customer says they’d rather go with bronze.

I’d still treat everything it finds as a question rather than an answer. Somebody who knows the job should confirm each item before it goes on the list. If the call says bronze while the contract says black, that’s a question for the customer, and the AI shouldn’t settle it.

Why shouldn’t AI do the final check?

AI shouldn’t do the final check because, once you know what the job requires, checking the order is a yes-or-no question, and a plain rule answers it the same way every time.

If the job needs tempered glass, the only question is whether the order says tempered. That check can be a line in a spreadsheet, a box on a printed sheet or a rule in the software you already use. If you ask an AI tool whether an order looks right, you’ll get a confident answer, but you might get a different one if you ask again, and it won’t always tell you what it actually compared.

We build software for a living, and I still wouldn’t put AI on that last step. I’d let it help read the messy material and point out what might matter, have a person confirm what it found, and leave the final comparison to a rule that can’t change its mind.

When is your current review enough?

Your current review is probably enough when your jobs rarely differ from each other. If you sell a small catalog with few options, or the same experienced person talks to every customer and places every order, a separate list probably adds little.

The list starts to matter when jobs vary, when the person placing the order isn’t the person who talked to the customer, or when someone on the team is still learning which jobs are unusual. That’s when a normal choice can slip through without anyone noticing it.

Signs your order check is missing what’s different

You can check these before the next order goes out:

• The person placing the order didn’t talk to the customer.
• Your usual options get picked by habit or come pre-filled on the form.
• The review checks that every field is filled in, not what each field should say.
• The customer changed something by phone or text after the quote.
• Someone placing orders is still learning which jobs are unusual.

If none of these apply, the review you have is probably doing its job. If any of them sound familiar, write down what’s different about the next job and check the order against that list before it goes in. If you’ve had a costly mistake before, add whatever went wrong to the list for every job like it.