Handle False Rejects With a Controlled Review of Vision Inspection

A quality manager calls you about false rejects on a vision inspection station right after a production line move. She wants the numbers to drop today, and she wants you to change the settings remotely, now, while she's on the phone. The line is down for reinspection, the pressure is real, and you have almost none of the information you'd need to act responsibly. This is where technical sales engineers earn or lose trust in a single call.

Why this conversation goes wrong

When a vision inspection station starts flagging good parts as defects, the plant sees it as a settings problem. Someone changed a threshold, someone moved a camera, and now the numbers look wrong compared to last week. The instinct is to treat it like a dial you can just turn back. A sales engineer who wants to help, with a support clock running, feels pulled to say yes and start adjusting parameters over a screen share.

That instinct causes real damage. A vision system change involves lighting, camera position, part presentation, and the labeled sample set the algorithm was trained against. Loosen a threshold to clear false rejects and you also raise the odds that a real defect gets shipped. You can't see that risk from a phone call. Only the plant's quality function, with the documented acceptance criteria and change control record in hand, can say what tolerance is acceptable. Your job is to slow the moment down without making the quality manager feel dismissed. That's the hard part. Say no too bluntly and she hears you as unhelpful. Agree too fast and you own a shipped defect that traces back to a change you made on a service call with no engineering authorization behind it.

How to prepare before you respond

Before you touch anything, or promise to, get four things in front of you. First, the current approved quality criteria for that station, not a verbal description of it. Second, sample images from before and after the line move, sorted into good parts and defective parts if the plant has done that work. Third, the change control record for the move itself, covering what physically changed, lighting, camera angle, part fixturing, conveyor speed. Fourth, the name of the person authorized to approve a change to inspection settings, because it is rarely the person calling you.

Ask about the move itself before you ask about the software. A station that moved twenty feet across the floor can pick up new ambient light, a different angle on its mounting bracket, or vibration it never saw before. False rejects that started right after a move point at the physical environment first and the algorithm second. Jump straight to remote parameter changes and you skip the step that usually explains the problem.

Also check whether the plant has a labeled defect set for this station. Many plants have plenty of images and no organized record of which ones are true defects and which are false rejects caused by the new environment. Without that labeled set, nobody, including you, has a basis for judging what a corrected threshold should look like. Building it isn't busywork. It's the only way to validate a change instead of guessing at one.

What to say

Nia, the quality manager, opens with pressure. She says, "False rejects jumped after we moved the line last week. We're drowning in reinspection. Can you just tune the station remotely today?"

You don't open with a settings conversation. You say, "I want to fix this as much as you do, and changing a threshold without the right information could make outgoing quality worse, not better. Before anything changes, I need the current quality criteria for this station, sample images from before and after the move, and the change control record for the move. Did the lighting or camera position change when you moved the line?"

Nia confirms that the lighting changed and that the team has plenty of images but nothing sorted into a proper defect set. You respond, "That gives us a real place to start. Have your controls engineer and your quality lead pull a representative set of good parts and defective parts from what you already have. Once we define validation conditions against that set, we can run a controlled test before anyone changes a production setting. I can turn that around fast once the samples are ready."

Notice what this exchange does. It never says no to helping. It refuses to guess, names exactly what is missing, and hands ownership of the acceptance criteria back to the quality function where it belongs. Nia gets a plan and a timeline instead of a promise you can't actually keep.

Mistakes that make it worse

The most common mistake is changing a parameter over a screen share because the quality manager sounds confident and the deadline sounds real. Confidence on the phone is not the same as documented authority. A second mistake is asking for information but accepting a verbal summary of it instead of the actual record. Someone's memory of what the acceptance criteria say is not the acceptance criteria. A third mistake is promising a specific outcome, telling Nia the false reject rate will drop to some number once you adjust the station. You don't control the labeled data or the validation process, so you can't promise its result.

A quieter mistake is treating the physical move as irrelevant and going straight to the vision algorithm. Plants that just moved a line rarely think to mention that the camera now sits a few degrees off its old angle, or that a new overhead light falls across the inspection field at a different time of day. Ask about the move directly, every time, because nobody volunteers it on their own.

How to practice and score this conversation

Run a coaching drill built around five short descriptions of sample images, some clearly good parts and some clearly defective, with a couple ambiguous on purpose. Have the rep play the technical sales engineer while a manager or peer plays a quality manager under deadline pressure. Score three things. Did the rep ask for the documented criteria instead of accepting a description of them. Did the rep ask about the history of physical changes to the line, not just the software. Did the rep end the call with a scheduled data review and a clear list of what evidence is needed to authorize a controlled test, instead of a vague promise to look into it.

A rep who ends the call having changed a setting, or having promised a result, fails the drill regardless of how calm they sounded doing it. A rep who ends the call with a plan, a named owner for the labeled sample set, and a next meeting on the calendar has handled it correctly. This kind of drill works well in DealSpeak, where you can run the same pressured opening line repeatedly and compare how different reps handle the handoff between sympathy and authority. For more scenarios built around this kind of pressure, see Manufacturing practice conversationsincluding Set Vendor Managed Inventory Criteria Before Revising an Agreement and Plan a Welding Fixture Wear Review Before Scheduling Service.

The plant doesn't need a fast settings change. It needs a controlled review that protects outgoing quality while it clears the reinspection backlog. Handle it that way and Nia gets a real answer instead of a quick one, and you get a defensible record if anyone ever asks why the station's behavior changed.