Docs › Forecasting

How forecasting works

Skucast looks at how each part has actually sold, figures out the pattern, and tells you what to reorder and when — so you stop running out of fast movers and stop drowning cash in slow ones.

The 10-second version

For each part, Skucast figures out how fast it really sells, how long your supplier takes, and what's already on hand or inbound — then tells you what to buy so you don't run out or over-buy.

Why parts are hard to forecast

Most tools assume steady demand: "you sold 30 last month, so order one a day." That works for milk. It does not work for parts. A coupler sells 6 one week and zero for the next three. A roof vent sells all summer and stops in October. A discontinued fender hasn't moved since spring.

Average that out and you get a number that's wrong in both directions — you over-order the slow stuff (cash stuck on a shelf) and under-order the bursty stuff (a stockout the day a customer actually needs it). This pattern is called intermittent demand, and it's the normal case for a parts business.

Skucast uses Croston's method — a forecasting approach built specifically for intermittent, lumpy demand. Instead of one naive average, it separates how often something sells from how much sells when it does, then combines them into a realistic demand rate. You never touch that math — you just read the result.

What Skucast calculates for every SKU

  1. Demand rate (velocity) — roughly how many units a day this part really moves, smoothed across its history.
  2. Demand type — the shape of that demand (below). This decides how aggressively Skucast plans.
  3. Days of supply — at the current sell rate, how long your on-hand stock will last.
  4. Lead time — how long your vendor takes to deliver, learned from your received POs (or set per supplier).
  5. Reorder recommendation — putting it together: should you order now, and how many, so new stock arrives before you run out, with a safety buffer.

The 5 demand types

Skucast tags every part with one of five demand types. The tag tells you why it's planning the way it is — the same on-hand quantity gets a completely different recommendation depending on the type.

Smooth
Sells steadily and predictably — a popular solenoid that moves most weeks.
Skucast: forecasts confidently; reorder amounts are tight and reliable.
Lumpy
Sells in occasional bursts then goes quiet — big-ticket, low-frequency parts like an axle or a winch.
Skucast: plans conservatively — won't over-stock off one busy week.
Seasonal
Demand rises and falls with the time of year — vents, propane, A/C parts.
Skucast: weights recent in-season demand so you're stocked for the season, not the off-season.
New
Too little sales history to trust a pattern yet.
Skucast: watches it, holds off on aggressive recommendations, and re-classifies once a pattern forms.
Dead
No meaningful sales for a long stretch.
Skucast: does not recommend reordering — surfaces it so you can clear it out, return it, or discontinue it.

Reading the recommendation

stock time → reorder point lead time stock arrives safety buffer
Skucast orders at the reorder point — early enough that the new stock lands before you'd hit your safety buffer, given the vendor's lead time.

Why a forecast changes day to day

A forecast isn't a fixed number — it learns. It moves when:

This is a feature, not a bug: the more it watches your business, the better it gets.

"That looks wrong" — FAQ

Why is this part "lumpy" and not "smooth"?
Because it sells in bursts with quiet gaps, rather than a steady trickle. That's not a problem — it's just a different demand shape, and Skucast plans for it more conservatively so you don't over-buy off one good week.
It says to order, but I just got a shipment / I have plenty on the shelf.
Check two things: (1) Did the receive get logged in Skucast (or sync from Shopify)? Days-of-supply is based on the on-hand it knows about. (2) Is the inbound on a PO in Skucast? If a PO isn't recorded, Skucast can't credit the incoming stock. Once it sees the stock, the recommendation drops.
It's NOT recommending something I clearly need to reorder.
It's likely classified Dead (no recent sales) or New (not enough history). Dead items are skipped on purpose; new items are watched until a pattern forms. Also confirm it isn't excluded from forecasting (kits, services, and one-offs often are).
The recommended quantity looks too high / too low.
The number is driven by three inputs — velocity, lead time, and current/inbound stock. A long vendor lead time pushes the quantity up (you have to cover more days); a short one pulls it down. If a lead time looks off, fix it on the supplier and the recommendation corrects itself.
How much sales history does Skucast need before I can trust it?
It starts working immediately, but the more history it has, the sharper it gets — especially for lumpy and seasonal items. Brand-new parts stay New until they've sold enough to establish a pattern.
A kit / service / labor charge is showing up in the forecaster.
Exclude it. Skucast forecasts physical, re-orderable parts; anything that isn't (services, labor, bundles, one-offs) can be flagged out so it stops cluttering your recommendations.

Next