The 1.9% Nobody Was Looking For: Finding Hidden Waste in a Bakery's Packaging Line

When I started as a Production Data Analyst Intern at Atticus Bakery, one of the first numbers anyone mentioned to me was the 2% buffer — the accepted margin for overpacking finished product. It's the kind of number that gets set once, gets nodded at in meetings, and then quietly stops getting questioned. Nobody had recently gone back and asked: are we actually hitting it?

I decided to find out.

Starting with the data nobody had assembled yet

The production and packaging numbers existed, but they lived in scattered places — logs, shift reports, the kind of raw data that's technically available but not actually usable without someone sitting down and stitching it together. So that's where I started: pulling production and packaging data into Excel and building a structure that could actually answer the question, rather than just gesture at it.

I wasn't looking for a single dramatic number. I was looking for a pattern — was overage clustered around specific SKUs, specific shifts, specific lines? Or was it just noise?

What the numbers actually said

Once the model was built, the answer wasn't subtle. Actual overage was running 1.9 percentage points above the 2% target buffer — which doesn't sound catastrophic until you multiply it across every unit the line produces. That's not a rounding error. That's material walking out the door that doesn't need to.

The harder part wasn't finding the number — it was making sure it was defensible. I cross-checked it against multiple time windows and product lines before I trusted it enough to bring to anyone.

Turning a one-time finding into something that sticks

A single report answers a single question once. I didn't want this to be something that got discovered, discussed, and then quietly forgotten in six months — which is what tends to happen to one-off audits. So I built recurring KPI reporting files tracking yield, scrap, and line performance, and worked with the DMAT and operations team to fold better data collection into how the line already ran, rather than bolting on something extra.

Alongside that, I also reviewed facility layouts and blueprint changes to make sure production changes stayed compliant with Connecticut regulations — a reminder that in food manufacturing, "improve the process" and "stay within the rules" aren't separate conversations.

What this taught me

The biggest lesson wasn't really about Excel formulas or KPI dashboards. It was that "acceptable" thresholds are only as good as the last time someone actually checked them. A 2% buffer had been treated as settled fact — and it took someone actually pulling the data to notice it wasn't holding. That's most of what data analysis is, in practice: not finding some hidden secret, but being willing to check the thing everyone assumed was fine.