Session Notes: Megan — Production Labor Numbers

2026-06-01 · Megan Calloway + Nate St. Pierre

Overview

This session covered Merrick Awards & Engraving's efforts to capture and make sense of production floor labor data — specifically, understanding how long jobs actually take and using that to improve scheduling, capacity planning, and pricing decisions. Megan walked Nate through the current system: a Microsoft Forms–based checklist that workers fill out during jobs, feeding into a master Excel spreadsheet that they've been building out for about a year. The conversation moved from the current state to the data challenges to the business stakes underneath, including the fact that the person who currently makes most of these judgment calls intuitively (Sandra, the founder) is trying to step back.

What They're Trying to Do

The stated goal is to document and analyze production floor output — to know how long jobs actually take (from the job itself), and then use that data to predict how long future jobs will take. The intended applications are practical and specific: scheduling ("do we need overtime this week? can we take this rush job? when are we going to be done?"), capacity utilization management, and pricing (translating production time to cost). There's also an underlying accountability use case — being able to see when a job took significantly longer than expected, and understanding why.

The Forms + Excel system is in use and the team has adopted it — this is working, not failing. The issue is that the data it produces is too sparse across the wide SKU variety to generate reliable estimates yet. Scheduling and pricing decisions still rely heavily on Sandra's experience — Megan described her as someone who "can just look at the pile and be, like, okay, I know what to do, and it's going to take me a week." That works when Sandra is in the building, but she's trying to step back, and that kind of deep intuition isn't teachable to a new person in any direct way.

The Current System: Microsoft Forms Checklists

About a year ago, Merrick implemented a production checklist system built in Microsoft Forms, connected to a master Excel spreadsheet. Workers navigate through the checklist step by step as they complete a job. The checklists have branching logic — depending on the job type, certain sections are skipped or expanded — and they include links to SOPs and training images for newer operators. Megan said this is the best version they've had; previous systems kept documentation and output recording separate, which meant people didn't know where to find the SOPs and didn't use the recording spreadsheet.

The master spreadsheet receives a new row each time a form is submitted. There are dozens of different checklists, each covering a different production process (the one Nate observed was a sublimation checklist). The spreadsheet output includes item number, process steps completed, and other per-job data.

Megan and others have been using Claude on an ad hoc basis to analyze the data — pasting data in and asking questions like "make sense of this" or "tell me about what happened today." Each analysis creates a new tab in the spreadsheet. There's no systematized reporting; it's exploratory at this point.

The Data Problem

The central challenge is one of volume and variety. Merrick handles a very large number of SKUs across many different processes, and the "matrix of combinations" — product type, process, quantity, operator — is enormous. Because the same exact combination doesn't repeat frequently enough, individual use cases don't accumulate enough data points to produce statistically meaningful estimates. Megan pulled up one analysis where units-per-hour swung across a roughly tenfold range from one job to the next, with no clear pattern. That range is too wide to be useful.

Megan and Nate described this as "an existential problem" for this project. It doesn't prevent them from building something — they still have to make scheduling, pricing, and capacity decisions regardless — but it does mean that broad statistical models will take years to mature.

The more tractable starting point that Megan identified: roughly half of the business is the wholesale channel, which has a much smaller product set — a few hundred SKUs. Within those products, there's an 80/20 distribution where a smaller number of high-volume items account for the majority of production runs. Those high-mover products repeat often enough that historical data may be dense enough to show reliable patterns. Megan noted that if any part of this project is going to yield usable estimates quickly, that's the piece.

Technology and Tools

The Microsoft Forms and Excel setup runs on the company's existing Microsoft 365 subscription, which covers all employees. Megan mentioned they've experimented with Copilot for the analysis work but found Claude produces better results and is easier to work with. She's interested in potentially expanding Claude access to more people ($20/person/month, which she described as "peanuts for what we could potentially get") but wants to be more deliberate about how they actually structure that before rolling it out. She also noted that she doesn't really want a separate Claude structure for everyone without a plan for how it's actually going to be used.

Janet was mentioned as spending significant time trying to understand variability in the production data — why jobs of nominally the same type produce such different numbers.

Business Stakes

The underlying urgency is Sandra's eventual departure. The pricing and scheduling intuition that currently lives in her head — built over decades — isn't documented anywhere in a form that can be transferred. The labor data project is, at its core, an attempt to systematize enough of that judgment that someone new could eventually take it over. Megan acknowledged this is a long project with a long data-collection runway before it generates really reliable estimates, but said the alternative — continuing to run entirely on one person's intuition — is not a viable long-term model.

Where Things Stand

Nate said he'd explore this further — research approaches, talk to his technical partner Pete if needed, and come back with a perspective. He was honest that he couldn't immediately say "here's how you fix this" because the data sparsity problem introduces real uncertainty about what an AI-driven analysis layer could actually deliver in the near term. The 80/20 wholesale channel insight Megan shared was the clearest thread to pull: start with the high-mover SKUs, see if the historical data is dense enough to produce useful patterns, and build from there.

Follow-Ups

• Pull a sample of the master spreadsheet data for the top wholesale channel SKUs — specifically to see how many repeat-job entries exist and how tight the variance is on time-per-unit.
• Clarify what role Janet plays — she was mentioned in the context of variability analysis.
• Understand Sandra's actual step-back timeline (affects urgency and whether there's a hard deadline to work toward).
• Confirm: is job duration actually being captured in the Forms data, or just the checklist steps? (If there's no time/duration field, the data may not be useful for estimation at all without adding one.)