Every seller I know has a version of the same story: a bestseller goes viral on TikTok Shop, and by the time the reorder lands from the supplier, three weeks of sales have gone to a competitor. Or the opposite happens — you overorder "just in case" and end up with cash sitting on a shelf as unsold stock, quietly eating your margin every month it doesn't move.
Running listings across Amazon FBA, eBay, Shopify and TikTok Shop, I used to plan reorders by gut feeling and a spreadsheet I updated when I remembered to. That doesn't scale past a handful of SKUs. Here's the AI-assisted forecasting workflow that replaced it — no expensive inventory software required, just the AI tools you're probably already paying for.
Why Spreadsheet Guessing Breaks Down Fast
Manual forecasting usually means eyeballing last month's sales and adding a bit of a buffer. That works fine when you're selling one product on one platform. It falls apart the moment you're multi-channel, because Amazon's Buy Box behaviour, eBay's slower organic search, and TikTok Shop's viral spikes all move on completely different rhythms. A method that treats every platform the same will always be wrong for at least one of them — usually the one about to sell out.
Step 1: Get Your Sales History Into One Place
Before AI can forecast anything, it needs clean data. Export the last 3–6 months of unit sales per SKU from each platform — Amazon Seller Central, eBay Seller Hub, Shopify's analytics, and TikTok Shop's dashboard all let you download CSVs. Paste that data (or upload the file directly) into Claude or ChatGPT along with your current stock level and typical supplier lead time. The goal isn't a perfect dataset; it's enough real numbers that the forecast is grounded in your actual sales pattern, not a generic guess.
Step 2: The Prompt That Turns History Into a Forecast
This is where most sellers ask too vague a question and get a useless answer back. Don't just ask "how much should I order" — give the AI the structure to reason through it properly.
That last part matters more than people expect. AI is very good at spotting a one-off spike caused by a promotion or a viral post and telling you not to bake it into your baseline — something a simple spreadsheet average will happily get wrong.
Step 3: Set Real Reorder Points, Not Round Numbers
Once you've got a forecast, turn it into a reorder point: the stock level that triggers a new order, calculated from lead-time demand plus your safety buffer. Ask the AI to build you a simple reorder-point table for your top 10–20 SKUs, ranked by how close each one currently is to triggering. Check that table weekly rather than trying to remember every product's status in your head — five minutes on a Monday beats an emergency reorder on a Friday.
Step 4: Layer In Seasonality and Platform Spikes
Generic forecasting tools often miss what actually drives e-commerce demand: a TikTok Shop product going viral, a Black Friday spike, or a slow patch every summer. Feed the AI last year's data for the same period if you have it, and specifically ask it to flag which weeks look seasonal versus which look like noise. For newer products with no history, ask it to benchmark against a similar SKU you've sold before instead of guessing from zero.
What AI Still Can't Do For You
AI is excellent at pattern-spotting across messy multi-platform data, but it can't call your supplier, doesn't know about a factory delay nobody's told it about, and won't catch a listing error that's tanked your visibility overnight. Treat every forecast as a strong starting point that you sense-check against what you actually know about your supply chain — and re-run it monthly, not once and forget it.
If you're spending more time babysitting spreadsheets than actually growing your store, that's usually a sign the whole operation needs a proper systems review, not just a better forecast. That's the kind of automation and workflow audit I help sellers with through Tafuma Web Services — building the dashboards and processes that make this run itself.