I-Connect007 Magazine

I007-Nov2026

IPC International Community magazine an association member publication

Issue link: https://iconnect007.uberflip.com/i/1547617

Contents of this Issue

Navigation

Page 105 of 129

106 I-CONNECT007 MAGAZINE I NOVEMBER 2026 N obody in a PCB shop changes a plating chemistry because a vendor says it "feels better." You run a sample, compare it with what you had, and look at the numbers. If they move in the right direction, you write the change into the process. If they don't, you change it back. That discipline is so normal in our industry that most of us barely notice it anymore, yet it is missing from almost every AI effort I see. AI usually enters a shop when someone tries a tool, gets an impressive answer, and tells everyone it's great. A few people start using it, some eventu- ally stop, and six months later the question comes up in a meeting: Is this actually helping? Nobody knows, because there was never a "before" to compare it with. The whole effort runs on feelings, which is the one thing we would never accept on the floor. That is why I keep saying don't buy AI, learn AI. You can buy a license, but you cannot buy a baseline. That baseline has to come from your process, your people, and your numbers, and only you can collect it. Treat Every AI Habit Like a Process Change The good news is that fabricators already know how to do this. We run SPC, use change control, and understand the difference between a trend and a one-off, so the next step is to take the habits we already trust and point them at AI. Start with one piece of recurring work, not some- thing broad like "AI for quality," but one specific task. Maybe it is the RFQ review that takes up the first half of every Monday for your sales engineer, the shift handoff summary that costs the outgoing lead 20 minutes, or the first draft of a corrective action report. Before anyone touches AI, write down what that work looks like today. Track how long it takes, how often something gets missed or sent back, and how many times a customer comes back with a ques- tion the RFQ review should have caught. You do not need a complicated system to do this. Then change one thing by bringing AI into that task in the same way each time for the same number of weeks. When the trial is over, compare the results. Did the review get faster, did fewer questions come back, or did the handoff catch the things that used to slip through? If the numbers move in the right direction, keep the change and write it down so the next person does it the same way. Stop if they don't. That is not a failure, but rather the experiment doing its job, and it keeps you from building a habit around something that does not work. Measure What the Work Is For Speed is one of the easiest numbers to measure, but it is also the one most likely to fool people. AI will almost always make a first draft faster, though BY S E A N PAT T E RS O N , C RO S S G E N A I You Can't Buy a Baseline D O N ' T B U Y A I , L E A R N A I

Articles in this issue

view archives of I-Connect007 Magazine - I007-Nov2026