Forecasting Inventory for a Subscription Business

The great advantage of a subscription business is that you know roughly what you will ship before you have to ship it. The great risk is that you must commit to inventory before the number is final.

The equation

For the next ship date:

Units needed
  = active subscribers at last ship
  − expected churn before next ship
  + expected new signups that qualify
  − expected skips
  + safety stock

Four of those five are commonly estimated. The one that is commonly omitted is skips — and it is often 5–15% of a cohort.

Forecasting without skips means over-ordering every cycle, permanently. For a curated box where the inventory cannot be reused next month, that is pure loss.

Qualifying new signups

Not every signup between now and ship date receives this cycle’s box. It depends on your cut-off rule: does someone signing up two days before ship get the current box or the next one?

Whichever rule you choose, the forecast has to use the same one. A mismatch here is a reliable source of small, recurring shortfalls.

Lead time drives how much you have to guess

If your supplier lead time is two weeks and your cycle is four, you commit with two weeks of visibility. If lead time is eight weeks, you are committing before the previous cycle has even shipped, and your forecast error compounds.

Shortening lead time is worth more to a subscription business than to a conventional retailer, because every week of lead time is a week of extra uncertainty applied to a known demand curve.

Replenishment is easy; curation is not

Replenishment ships the same SKUs continuously. Leftover stock is next month’s stock. Normal reorder points and safety stock work.

Curation commits to a specific selection for a specific date. You cannot backorder — the box ships on the 1st regardless. Unsold units are dead, because next month is a different box.

That asymmetry is why curation carries much more working-capital risk, and why the forecast needs to be tighter for exactly the business where it is hardest. The comparison.

Build-a-box forecasting

The hybrid has a genuinely useful property: choices lock at a deadline, and after that you know precisely what to pack.

Before the deadline you hold safety stock across the eligible range; after it you have certainty. Narrowing the eligible list reduces how much breadth you must carry. That is the real operational argument for a curated eligible list rather than the whole catalogue. Build-a-box.

Safety stock, sized by consequence

Size it by what a stockout costs, not by a uniform percentage.

A stockout on a replenishment SKU is a delayed delivery — annoying, recoverable. A stockout on a curated box’s hero item means the box does not ship, or ships visibly incomplete, to the entire cohort at once.

Hero items deserve much more safety stock than filler items.

Track forecast accuracy

Record forecast versus actual every cycle and watch the bias.

Consistently over-forecasting usually means skips are not in the model. Consistently under-forecasting usually means new signups are growing faster than the model assumes.

Both are correctable once you can see the direction, and invisible if you only look at whether you ran out.