What happens today
The patchwork
- A reorder spreadsheet one buyer keeps in their head
- Last year plus a fudge factor for the holidays
- Promo spikes nobody planned the inventory for
- Safety stock set high on everything, just in case
Forecast playbook
Plan inventory per product
Overstock down 31%. Stockouts down 44%. Q4 saved.
- 5 min read
- Intermediate
- 4 moves
- Updated weekly
The number
Overstock down 31%, stockouts down 44%, and Q4 planned to the numbers instead of survived.
−31% / −44%
The situation
From a guess that burns cash to a forecast you can plan around.
Upload eighteen months of sales and add your holiday and promotion dates. Forecast predicts the next twelve weeks of demand for every product (with a confidence range, not a single guess) so you order to the numbers instead of gut feel.
With Forecast
The workflow
- One demand forecast per product, refreshed weekly
- A confidence band you can size safety stock against
- Holiday and promo dates baked into the numbers
- Your buyer opens Monday to a twelve-week demand plan for every SKU.
How the work moves
Four moves.
Forecast runs each move with a preview attached, so you see the demand curve before you commit a purchase order. Skip freely once you know which parts carry the weight.
Forecast at work
One brief travels through the whole job. Context stays attached until the finished work lands.
Kickoff
Upload your sales history. A CSV export or straight from your store backend.
Forecast starts with the history you have and asks only for what is missing. No data-science setup, no spreadsheet formulas.
Add context
Add your holiday and promotion dates.
Those dates go in as exogenous series, so the model learns your real spikes instead of smoothing them away. Weather and launch dates fit here too.
Forecast
Get a twelve-week demand forecast for every product.
Each SKU comes back with a confidence band, not a point estimate: the range you actually size safety stock and reorder points against.
Plan
Schedule it weekly. Review the forecast. Place the orders.
A refreshed plan lands in the format your buyers already use. You adjust the last 10%, not the first 90%, and rerun the moment reality drifts.
Prompts worth keeping
Paste these into Forecast.
Three prompts, a kickoff, a full run, and a packaging pass. Copy the one that matches the phase you're in. Rewrite any detail to fit your catalog.
Kick it off
You are Forecast. I want to plan inventory per product: I will upload my sales history and add my holiday and promotion dates, and you predict the next twelve weeks of demand for every product with a confidence range. My goal: cut overstock and stockouts so Q4 is planned, not survived. Walk me through the first move and tell me what you need from me.
Run it in one shot
Run the full playbook end-to-end: Upload my sales history: a CSV or straight from the store backend. Add my holiday and promotion dates as exogenous series. Return a twelve-week demand forecast per product with confidence bands. Schedule it weekly, review the forecast, place the orders. Ask before skipping any step. Show work as you go.
Package the output
Deliver the output as a single plan I can share with the buying team: lead with "−31% / −44%" (overstock down 31%, stockouts down 44%), then the per-product forecasts and their bands. Call out any product where the confidence range is wide enough to warrant a hedge.
What goes in. What comes back.
Inputs in, outputs out.
Forecast runs on the inputs on the left and hands back the artifacts on the right. Skip any input, the agent will ask for it the first time it needs it.
Give it
- Your sales history (CSV, database, or store backend)
- Holiday and promotion dates for the season
- The horizon you plan against: twelve weeks by default
Get back
- A twelve-week demand forecast per product, with bands
- A structured file for your reorder and safety-stock math
- An alert when actuals drift outside the predicted range
The finished artifact
A demand plan, not a spreadsheet to wrangle.
Every run ends the same way, a packaged forecast in the channel your team already reads. Here's a preview of what shows up.
Forecast → your team
Plan inventory per product, ready for review
playbooks
Mon 7:03 AM
Here's this week's demand plan. I ran the playbook end-to-end, refreshed the twelve-week forecast for every product against the latest sales, and flagged anything drifting outside its band.
- Upload your sales history: a CSV or straight from the store backend.
- Add your holiday and promotion dates.
- Get a twelve-week demand forecast for every product.
- Schedule it weekly. Review the forecast. Place the orders.
What moved
−31% / −44%
Overstock down 31%, stockouts down 44%, and Q4 planned to the numbers instead of survived.
forecast.pdf · demand.csv · bands.xlsx
Before you run it
Where teams stall.
Three ways we see this go sideways and how to avoid each one.
Keep judgment with the team. The playbook should make the work clearer, not hide the assumptions behind it.
- Uploading history without the holiday and promo dates. Add them once, and the model stops smoothing away the spikes that actually cost you.
- Ordering to the point estimate and ignoring the band. Size safety stock against the range: that is where the overstock and stockout wins live.
- Running it once before the season and forgetting. Put it on a weekly cadence so the forecast tracks reality instead of last month.
Questions
Before you start.
Enough history for the pattern to be visible: roughly a year and a half of sales lets the model learn your seasonality and holiday spikes. Forecast runs on a plain CSV export for the first pass; connect the store backend once you want it refreshing on its own.
Demand is never a single number, so a point estimate quietly hides the risk you are planning against. Forecast returns a confidence band per product, and that range is exactly what you size safety stock and reorder points against: narrow bands you can plan tight, wide ones you hedge.
You add them as dates and Forecast treats them as exogenous series alongside the sales history. That means a Black Friday promo shows up as a learned spike instead of noise the model averages out, so the twelve-week forecast reflects the calendar you actually run.
Weekly is the sweet spot: schedule it so a refreshed twelve-week plan is waiting Monday morning. You can trigger an ad-hoc run whenever a new promo lands or sales swing hard, but the overstock and stockout numbers only move once it is on a standing cadence.
