How reorder recommendations get made

Inventory forecasting is a lead-time problem wearing a math costume.

Strip away the dashboards and the core question is simple: how much will sell before the next order arrives, and how much buffer do you need in case you're wrong? Everything else — moving averages, seasonality adjustments, safety stock — exists to answer that one question more precisely.

The three inputs every forecast needs

  1. Sales velocity — how many units of a SKU sell per day, typically averaged over a trailing window (7, 30, or 90 days depending on how stable demand is).
  2. Supplier lead time — how many days between placing a purchase order and having stock in hand, including production time, shipping, and receiving.
  3. Current on-hand quantity — what's actually sellable right now, which sounds trivial but is where most forecasting errors actually originate (see below).

From these three, the standard reorder-point formula is: (average daily sales × lead time in days) + safety stock. When stock drops to that level, it's time to reorder so the replacement arrives before you sell out.

Where forecasts actually break

In our experience building this, the failure modes aren't exotic — they're mundane data problems:

Incoming stock not counted

A forecast that only looks at current on-hand, ignoring purchase orders already in transit, will recommend reordering stock that's already coming.

New SKUs with no history

Under ~30 days of sales data, a velocity average is mostly noise. Forecasts for new products are inherently weaker until real history accumulates.

Seasonality without a prior cycle

A model needs to have seen at least one full seasonal cycle to adjust for it. First-year seasonal spikes catch every forecasting tool off guard, not just weaker ones.

How Foreshelf handles this

Foreshelf computes reorder quantities from real Shopify order history and current inventory levels, factoring in supplier lead time and in-transit purchase orders so it doesn't double-count stock that's already on the way. Where a SKU has expiry or batch data attached, the recommendation also accounts for spoilage risk — see expiry-aware inventory for that specific mechanism. Marcus explains each recommendation in plain language and proposes the action; you approve or reject it — the calculation itself is deterministic math, not a black box. Full breakdown at how forecasting works.

What Foreshelf does not claim: perfect forecasts for brand-new SKUs, or fully automatic seasonality detection before a first full cycle has been observed. Those are real limits of any forecasting approach, not something we've solved that others haven't.

See your own numbers

Free, no signup — the same reorder-point math, run on numbers you enter.