Blue Star Labs · Case File 01 · Interactive

The Daily Audit Problem

How five months of quality data got trapped in 900 Excel files — and what it took to set it free.

Ridgeline Molding Co. is fictional. Every number on this page is synthetic, generated in your browser from a fixed seed. No client data exists anywhere in this demo.
Scroll to start

Chapter 1

Meet Ridgeline Molding Co.

A precision injection molder. Fourteen presses, three shifts, six product families shipping to automotive and consumer customers. Like almost every plant we've walked, their quality system runs on two things: good engineers and Excel.

Every shift, a QA engineer pulls sample parts off the presses and runs the daily product audit: inspect each unit, mark any defects, disposition it PASS or REJECT. The record of that audit is an Excel workbook. One template, used by everyone. That template is where this story starts.

Chapter 2

The template everyone uses

It's a good template. Somebody put real thought into it. You type your name, the date, and your shift into the header block once — then the sheet does the rest for you: it carries your info down every row and computes the disposition automatically.

Daily Audit_MTorres_6-29.xlsx
L9fxPASS

Chapter 3

Now multiply by every engineer, every day

Eight QA engineers. Six production days a week. Each one does the right thing: fills out the template and hits Save As. Five months later, the audit folder on SharePoint looks like this.

None of this is anyone's fault. Every engineer did their job — the audits are done, and done well. But nobody's job was to make nine hundred workbooks act like one dataset. So nobody did.

Chapter 4

The Monday it hit the wall

Ridgeline's quality manager gets a reasonable request from a customer: "Show us your defect trends for the last quarter." Easy — the data exists. He points Power BI at the audit folder.

Power BI — refresh log (dramatization)
  • ERROR Column 'Result' contains mixed types — formula cells evaluated at last save; 41 files show stale values
  • ERROR Header block A3:B5 breaks table detection — expected column headers on row 1, found "Engineer:"
  • WARN 'Date' parsed as text in 5 distinct formats — "6/29/26", "June 29, 2026", "2026-06-29", "29-Jun", "March 14"
  • WARN Column count mismatch — 3 workbooks contain user-added columns ("Notes for Dan")
  • ERROR =SUM() defect counts return 0 where defects were marked "x" instead of 1 — rejects reported as PASS

So he does what every quality manager in America does. He opens each workbook, selects the data, copies, and pastes-as-values into a master sheet. File by file.

And here's the part that stings: the master sheet is a snapshot. Next Monday there are 48 new workbooks, and the whole exercise starts over. The audits keep getting done. The learning keeps not happening.

Chapter 5

Run the fix yourself

This is the part we build. A pipeline that lives in Ridgeline's own Microsoft tenant, watches the audit folder, and turns every workbook — past and future — into one clean, analysis-ready table. Press the button. This is a simulation of the real run, at about 400× speed.

0
0
0
0

Chapter 6

What the data was hiding

Five months of audits, visible for the first time as one dataset. Within an hour of the first refresh, three findings surfaced that nobody could see one workbook at a time. Every chart below is drawn live from the synthetic data.

Finding 1 — the plant is slipping, and it started in April

Weekly first-pass yield, all products. Each daily audit looked normal. The trend is only visible when 900 files become one line.

Finding 2a — two defects are most of the problem

Defect Pareto, full period. Flash and splay together account for the majority of all defects logged. Focus follows.

Finding 2b — the splay is a night shift problem

RX-340 bezel, weekly splay rate by shift. Shifts 1 and 2 are clean. Shift 3 breaks out the week of June 8 — the same week a new regrind lot started running nights. One material call fixes it.

Finding 3 — a mold asking for maintenance, in writing

p-chart: CL-208 wire-harness clip, weekly reject proportion, control limits set from the Feb–Mar baseline. This is flash from progressive tool wear — no single day's audit ever looked alarming, but the weekly proportion walks straight through the ceiling. The chart flags it weeks before it becomes a scrap bill and a line-down PM.

Where to walk first — June reject rate, product × shift

The whole plant, one glance. Hover any cell. This is the view a quality manager opens with coffee on Monday.

The point isn't the charts.

Power BI drew these. JMP could take the same clean table and run capability studies, DOE, regression against process parameters. The point is that every one of these signals was already sitting in the daily audits — paid for, inspected, recorded — and structurally invisible. The expensive part wasn't collecting the data. It was the last inch: making it usable.

Chapter 7

What the manual version costs

Put your own numbers in. These sliders start at Ridgeline's; adjust them to your plant and watch what copy-paste-as-values actually costs per year.

8
12
4
$65

Straight time only. Doesn't count the decisions made late, the customer report that took two weeks, or the splay running 7% on nights for a month because nobody could see it.

Chapter 8

How this works in your plant

Everything above ran on synthetic data in your browser. The real version is built around one principle that matters to any plant with proprietary product data:

Your data never leaves your building.We build the plumbing. We don't host, store, or even see production data.

Where your files already live

SharePoint · Teams · network drives · ERP/MES exports

The pipeline — runs in your tenant

Power Automate / Office Scripts / Fabric dataflow, or a small tool we deploy inside your network. Materializes formulas, normalizes, validates, flags exceptions.

Where your team already works

Power BI dashboards · JMP-ready extracts · Teams alerts when a control limit trips

"We already have Copilot."

Keep it — it gets better with what we build. Copilot answers questions about a document. It doesn't crawl 900 workbooks, materialize formulas into values, enforce a schema, maintain a refreshable Power BI model, or run SPC. That's data engineering, and it's the layer Copilot sits on top of. Clean data in, smarter Copilot out.

"Our data is proprietary."

Good — ours is too. That's why the architecture is tenant-native: the pipeline runs on your Microsoft 365 subscription or on a box inside your firewall, under your access controls and audit logs. There is no Blue Star server with your data on it. Nothing to breach on our side, nothing to subpoena, nothing to trust but your own tenant.

"We're an Excel shop, not a software shop."

Stay one. Your engineers keep the template they know — or we tighten it up with validation so the "x vs. 1" problem dies at the source. The pipeline meets your data where it is. No new software to learn on the floor, no six-month ERP project.

Who's behind this

Two manufacturing engineers with 30+ combined years on plant floors — daily audits, PPAPs, SPC, JMP, Power BI — who also build software. We've lived the Friday copy-paste. That's why this demo exists.

Chapter 9

The audit problem has cousins

Same disease, different spreadsheet. If any of these live in your plant, the treatment is the same: clean data, in your tenant, visible as a trend.

02

Downtime & OEE logs

Operator downtime sheets from every line, unified into real OEE — availability, performance, quality — instead of arguments about whose number is right.

03

CMM & measurement data

Dimensional reports trapped in PDFs and per-part files, pulled into capability trends. Cp/Cpk by feature, by cavity, by month — JMP-ready.

04

Maintenance logs

Work orders and PM records mined for repeat offenders and early-warning patterns. The mold that's about to start flashing? It usually says so first.

05

Tribal knowledge

Twenty years of setup sheets, deviation memos, and "ask Gary" — made searchable inside your tenant, so the answer survives Gary's retirement.

Bring us one ugly spreadsheet.

Thirty minutes, screen share, no deck. Show us the workbook your team fights with every week and we'll map out — live — what it would take to turn it into the kind of picture you just scrolled through. If it's not a fit, we'll tell you that too.

Book the 30 minutes

Or forward this page to the person who does your Friday copy-paste. They'll know exactly what it's about.