An agent that turns your sales history into a 90-day demand forecast for every SKU — how much you'll sell, what to ship to FBA/WFS and when, and the cash you'll need to fund it. The planning layer a reorder point can't give you.
For: Amazon FBA · WFS sellersAmazon FBA · Walmart WFS↗ Fund the right stock
The workflow
How the agent runs
1
Pull your sales history
12+ months of unit sales per SKU from your own account — exports, or the Amazon SP-API / Walmart Marketplace API integration the prompt sets up with you.
2
Build the forecast
Trend, seasonality, and day-of-week patterns per SKU become a weekly demand curve for the next 90 days, with an honest high/low band.
3
Layer market signals
FreshCrawl checks the public side: competitor stock-outs, price moves, and listing changes that bend demand — flagged on the SKUs they touch.
4
Plan the shipments
Forecast minus on-hand and in-transit becomes an FBA/WFS inbound plan: what to send, how many units, and the latest ship date that still avoids a stock-out.
5
Project the cash
Units × landed cost, week by week — the inventory cash-out curve for the quarter, so purchasing never ambushes the bank account.
6
Grade the forecast
Monthly, forecast vs actual per SKU, the biggest misses explained (promo? stock-out? competitor move?) and the model adjusted. A forecast that never gets graded never gets better.
What it needs
Inputs & tools
Sales historyYour reports / API
On-hand + in-transitYour inventory data
Market signalsFreshCrawl MCP
Landed costsYour cost sheet
OutputForecast + ship plan + cash curve
Your time≈ 20 min/week
A reorder point answers 'when do I buy more?' A forecast answers the questions that come first: how much cash does next quarter's inventory need, and what goes on the next FBA shipment? Sellers who plan shipments from a forecast stop paying for both stock-outs and panic freight.
Run this agent yourself
📑
1 · Bring your account reports
This agent's source of truth is your own account data — the inventory, shipment, returns and fee reports from Seller Central (or Walmart Seller Center). Export them — or integrate with the marketplace APIs directly (Amazon SP-API, Walmart Marketplace API): the LLM you're using can write that integration with you, and then the reports pull themselves on schedule. No public crawl can see them; they're behind your login, as they should be.
🔗
2 · Optional: link the FreshCrawl MCP
Add the FreshCrawl MCP connector — server crawl.onewally.com/mcp — and the agent can cross-check what your public listing pages state (dimensions, weight, current price) against your reports. Supporting evidence only: disputes rest on your measured actuals and your report lines.
📋
3 · Paste the prompt
Copy the prompt below, attach your reports, name your store. The agent reconciles everything and hands you the claim file — no code.
Copy the prompt
Starter prompt
Step 3 · Paste this into your LLM
# Inventory Demand Forecaster
MY STORE (fill in once):
- Store name: [your store name]
- Seller ID or storefront URL: [paste it — the agent uses it to find your catalog]
- Marketplace: [Walmart US / Amazon US]
First, use the FreshCrawl MCP to look up my store, pull my catalog,
and work only on my listings. If you can't identify my store from
what I gave you, ask me for my storefront URL before anything else.
RUN SCHEDULE (set this where you run the prompt):
- Run this prompt: weekly forecast refresh Mondays 6:30 AM, plus a monthly accuracy audit on the 1st.
- If your LLM app supports scheduled/recurring tasks, set it there; otherwise re-paste the prompt on that cadence.
DATA SOURCES: my own sales and inventory data is the source of truth. Feed it to you two ways: (a) I attach the exports (sales history, inventory, in-transit), or (b) we integrate with the Amazon SP-API and/or Walmart Marketplace API directly — if no integration exists yet, write it with me first, then pull the data yourself on schedule. Use the FreshCrawl MCP, if connected, only for market signals (competitor stock, prices, listing changes) — never as my sales data.
Build a 90-day demand forecast for every SKU: weekly units, from trend, seasonality, and day-of-week patterns in at least 12 months of history. Show a high/low band per SKU, not a single number, and mark the SKUs where history is too thin to forecast honestly.
Turn the forecast into an FBA/WFS inbound shipment plan (units per SKU, latest ship date) using on-hand and in-transit counts. Then project the purchase cash the plan requires, week by week, at my landed costs.
On the first run of each month, grade last month's forecast against actual sales per SKU, explain the biggest misses, and adjust the model. Flag any SKU where a competitor stock-out or price move (FreshCrawl) likely bends demand.
EXECUTION — READ CAREFULLY:
You forecast and plan; you never place orders, create shipments, or move money. Every number is a range with its basis shown — never present a forecast as a certainty, and mark low-history SKUs as low-confidence instead of guessing.
The prompt gets you the shape. The reliable version runs on live data through the MCP — that's steps 1 and 2 above.
Questions sellers ask
How is this different from the Reorder Forecast agent?
That agent drafts purchase orders from velocity × lead time. This one plans further out: a 90-day demand curve, FBA/WFS shipment plans, and the cash the plan needs. Many sellers run both — forecast for the quarter, reorder for the week.
How accurate is it?
It shows a high/low band instead of pretending precision, and it grades itself against actuals every month. Expect the band to be honest before it is narrow.
Does it create shipments or place orders?
No. It produces the forecast, the shipment plan, and the cash projection. Creating shipments and placing orders stays with you.
What if I only have a few months of history?
It forecasts anyway but marks those SKUs low-confidence and widens the band — and says so, rather than dressing a guess up as a forecast.
Run this on your store.
Every agent here runs on live FreshCrawl data through the MCP. Start free, connect your LLM, paste the prompt — or get the free Buy Box Loss Report and see the data first.