Demand Sensing Router
Route each SKU to the planning cadence its evidence deserves. Build the first routing board with Claude, then make every recommendation prove its signal, response window, and operating decision.
See the operating artifact
The operating problem
Faster is only useful when the signal can still change a decision.
Demand sensing and monthly forecasting solve different planning-horizon problems. A volatile SKU does not automatically deserve a faster cycle, and a stable SKU does not automatically belong in a monthly lane.
This pack forces the routing decision through five tests: regularity, variability, volume and value, signal freshness, and the remaining response window. Intermittent, lumpy, new, and phase-out items receive their own review path instead of being forced into a false binary.
Claude prepares the evidence and the first-pass board. The planner owns the final policy and any system-of-record change.
How it runs
From portfolio data to a reviewable routing decision
- 01
Bring the evidence
Demand history, current forecast, recent signals, lead time, service priority, lifecycle, and planner comments.
- 02
Profile the demand
Run the included ADI and CV² script to separate regularity from order-size variability.
- 03
Route the cadence
Recommend frequent sensing, monthly planning, or special-method review using the supplied policy.
- 04
Verify every move
Reject stale signals, missing evidence, closed response windows, and unsupported route changes.
- 05
Start the review forward
Bring changed lanes, exceptions, and open operating decisions to the planner approval board.
Inside the ZIP
Everything needed for a safe first run
Four-page field guide
A compact Shetty's Desk PDF with the safe first run, exact file replacements, evidence fields, review order, and troubleshooting.
Sample portfolio data
Synthetic SKU-location history and context fields, ready for a safe first run.
Installable agent skill
A concise routing workflow with policy, schema, deterministic scripts, and output contract.
Three worker definitions
Demand profiling, cadence routing, and independent verification responsibilities.
Expected outputs
Example routing board, route-change log, data-quality exceptions, and review brief.
Validation controls
Checks that stop missing evidence and unsupported frequent-sensing recommendations.
Tool adapters
Setup paths for Claude Cowork, Claude Code, and a portable chat-first starter.
Expected output
The meeting starts with the rows that changed.
Stable rows remain available, but the review begins with changed lanes, failed validation, and response windows that are closing.
| SKU | Evidence | Proposed lane | Decision | Status |
|---|---|---|---|---|
| A100 / TOR | Daily POS, 3-day response | Sense frequently | Replenish | Ready to review |
| C310 / VAN | Stable, 45-day horizon | Plan monthly | Capacity | No lane change |
| D440 / MTL | ADI 3.8, long zero gaps | Special review | Method choice | Planner required |
Get the complete Demand Sensing Router Pack
The archive includes a four-page field guide, one-command sample run, the skill, deterministic profiling scripts, sample data, routing policy, worker definitions, expected outputs, and validation checklist.
- Run python3 START.py for the synthetic first-run check
- Synthetic data only; no production access required
- Runs as a skill or as a guided Claude project
- Designed for planner review before any policy change
Your first run
- 01Run
python3 START.pyto prove the package with synthetic data. - 02Copy the two CSVs in
skills/demand-sensing-router/assets/and replace the sample rows with your governed fields. - 03Build the five outputs, validate the board, and review changed or failed routes before any system update.
The complete pack is ready.
Download v1.2.0 as a ZIP. No email address, account, or newsletter signup is required.
Download v1.2.0 ZIPThe archive is served directly by Shetty's Desk and contains synthetic sample data only.
Before you run it
What the pack will not decide for you
It does not claim that demand sensing is automatically more accurate or that every volatile item should move faster.
It does not update planning master data, release orders, or change the approved forecast policy.
It does make the evidence, missing data, route changes, and remaining operating decision visible enough for a planner to challenge.