Why I Built This Prototype
In a Supply Chain Data & Analytics team, rolling out Microsoft Copilot isn't about writing clever text—it's about helping colleagues see where language models add speed (screening unstructured contracts, summarizing cross-regional meetings, writing Excel/DAX formulas) and where they fail (doing mental arithmetic on tables or guessing missing data).
Supply Chain Prompt Library & Failure-Mode Lab
Select one of the 4 Supply Chain scenarios below to inspect the synthetic input dataset, see where a naive prompt or LLM calculation fails, and compare the structured prompt.
1. Synthetic Input Data Given to Copilot
4 Verification Rules for Supply Chain Copilot Users
A one-page quick-reference guide I would use in workshops and office hours with Lyngby and Bangalore colleagues.
Ask for the Formula, Not the Number
Language models predict tokens; they do not reliably sum columns, count time intervals, or calculate weighted safety stock across rows. I'd train colleagues to prompt: "Write the Excel SUMIFS / XLOOKUP or Power BI DAX formula to calculate X, and explain how to audit it"—letting Excel's deterministic engine compute the value.
Give the Model Permission to Say "Not Found"
When screening supplier contracts or batch logs, an unguarded prompt pressures the model to guess missing lead times or Scope 3 clauses. Every extraction prompt in a shared library should end with a strict guardrail: "If a metric or owner is not explicitly stated in the text, output [MISSING IN SOURCE]—never infer."
Force Section & Timestamp Anchors
A summary is only useful if a colleague can verify it in 10 seconds. Always require an extra column in output tables for Source Section / Page / Timestamp and require verbatim preservation of Batch IDs, Purchase Order numbers, and SKU codes.
Separate Decisions from Unassigned Actions
Cross-regional Teams transcripts often blur casual brainstorming with agreed S&OP actions. Meeting prompts should separate Confirmed Decisions from Unassigned Open Tasks and tag every action item by hub (Lyngby vs. Bangalore GCC) so nothing falls through the time-zone gap.
How I Would Structure Weekly "Copilot Office Hours" as a Student Assistant
A colleague brings a recurring pain point (an Excel tracker, a recurring S&OP slide update, or a supplier PDF). Together we check whether it belongs in Copilot, Power Automate, or standard Excel formulas.
I'd help test the prompt on edge cases—missing sensor rows, ambiguous dates, conflicting units—and add verification guardrails before the team relies on it.
Every validated workflow becomes a 1-page template in the Supply Chain AI Community hub, crediting the colleague who brought the use case.
Hypothetical Workshop & Time-Reclaimed Estimator
No black-box multipliers. Adjust the sliders below to test conservative vs. ambitious assumptions for how many net hours a structured prompt library and workshop program could save across Supply Chain teams.