Chapter 1
Meet Dale. Then check the calendar.
Dale Kowalski, 58. Senior Process Tech at Ridgeline Instruments since 1995. He owns the 1998 wave-solder line and the five test fixtures nobody else can keep in calibration. Three days ago, he made it official: 90 days.
Nobody's worried about the machines. The machines have manuals. What they don't have manuals for is what Dale knows about them — and the plant already knows, from experience, exactly what his absence costs. Tap the evidence:
The plant has already run this experiment.
Last July, Dale took two weeks of vacation. The reject rate on his line went 2.1% → 5.8%. The M-217 test-fixture changeover went 45 minutes → 2.5 hours. A 22,000-piece OP-450 lot shipped late and got rejected — $9,400 plus an expedited rerun. That was two weeks. Retirement is forever.
HR's plan was a form: "Please document your critical procedures." Three weeks later it had produced three pages. Dale doesn't write. Dale talks.
Chapter 2 · Playable
Play the night shift
It's 2:04 a.m. You're Marcus Webb, process engineer, hired last year. The wave line is throwing cold joints on every OP-450 connector row, the shipment cutoff is 6:00 a.m., and you are the most senior person in the building. What do you do?
Chapter 3
So Ridgeline put a tablet in the tool crib
Not a form. Not a "documentation initiative." An AI interviewer that talks to Dale like a sharp apprentice with no ego — and asks the follow-up questions a junior would be embarrassed to ask. Twelve sessions over five weeks — most of them 15 to 45 minutes, a couple that ran long once Dale got going. Total cost to Dale: 9.2 hours of talking.
Scripted re-enactment — on a real engagement this is a live voice conversation (first session facilitated by a human, ours), and nothing is recorded without a signed consent form. We never run live AI on a marketing page.
Notice what the interviewer did.
It didn't transcribe a meeting. It interrogated a claim: how do you know the zone runs cold? What breaks if someone follows the sheet? How would a new tech catch it before running scrap? Which products does it touch? That's the difference between a recording of what Dale happened to say and a structured capture of what Dale knows. A Teams transcript gives you the first one. Copilot can summarize it beautifully. It's still the first one.
Chapter 4
Talk becomes system
Every session gets processed into artifacts an engineer can actually use — and nothing goes live until a human signs it. Dale approves every card that carries his name. Watch session #4 become four of them:
From the transcript
"Runs cold — has since '99… Gauge says you're at temp; the board isn't. You run the sheet number, you get cold joints on anything with the big connector row."
The best line item is the one that deletes itself.
Eleven of Dale's quirks didn't get documented — they got engineered out. The zone-3 offset became a work order: recalibrate the sensor, correct the profile sheet, add an alarm limit. That's poka-yoke — mistake-proofing. The goal isn't a binder about the trap. It's no trap.
Chapter 5 · Playable
Two weeks after Dale's last day
Marcus, 2 a.m. again, cold joints again. But this time there's a search box on the shop-floor terminal. Try it — ask what Marcus asked, or pick a question:
In the real system this searches every approved SOP, quirk card, and transcript in your Microsoft tenant — from the same search box your team already uses. This demo answers three canned questions; it makes the point.
Chapter 6
What changed at Ridgeline
On his last day, Dale signed the quirk-card wall in the tool crib — 61 cards, his name on every one. Nobody replaced Dale. They kept him.
What this costs in the real world
Pilot Interview
Two sessions, one machine, ~10 approved SOPs. Proves the method — and whether your Dale will talk to it — in one week. Flat fee, nothing to cancel.
Knowledge Capture Sprint
One expert, one domain. 8–12 AI-led sessions over 3–4 weeks, first one human-facilitated. 25–40 SME-approved SOPs + quirk cards, a searchable knowledge base in your M365 tenant, and a mistake-proofing punch list.
Two experts / full department
Same sprint, doubled scope. Most plants have more than one Dale — and they know different halves of the same machine.
The questions worth asking us first
"Where does Dale's audio actually go?"
Straight answer: the live interview runs on a specialized voice-AI service, so audio passes through that service during the session — that's the one step that isn't inside your tenant, and we put it in writing (data-flow one-pager + consent form, signed before anything records). Everything produced — recordings, transcripts, SOPs, quirk cards — lands in your Microsoft tenant, owned by you. Blue Star retains nothing.
"My Dale won't talk to a robot."
Some won't — at first. That's why session one is run by a person (ours), with the AI listening in, and why the framing is legacy, not replacement: this is Dale's name on the wall, his voice in the answers, his 31 years becoming the onboarding track. The $1,500 pilot exists precisely to find out how your Dale takes to it before you commit to anything.
"We have Copilot. I'll just record him in Teams."
Do it — it's better than nothing. But a meeting transcript captures what Dale happened to say that hour. The interviewer probes: it asks how he knows, what fails, how to verify, what else it touches — then structures the answers into SOPs and routes them through his sign-off. Copilot summarizing hearsay is still hearsay. And the knowledge base we leave behind makes your Copilot better, because it finally has clean, approved source material.
"I've bought SOP binders before. They're shelfware."
So have we — that's why the deliverable that matters is the punch list. When a quirk can be engineered out, we write the work order to kill it, not a page about living with it. The binder is the byproduct. The mistake-proofing is the product.
Who's your Dale?
Every plant has one. Most have two. Thirty minutes, no deck: tell us who they are and when they're leaving, and we'll tell you exactly what a pilot would capture first — and what it can't.
Book the 30 minutesIf it looks like a fit, the next step is the $1,500 pilot — two sessions, one machine, one week, and you'll know whether your Dale takes to it. If it doesn't look like a fit, we'll say so.
Or just email john@bluestarcentral.com. And if you haven't seen how Ridgeline got its audit data under control first, start with Case File 01: The Daily Audit Problem.