Guides › Setting a threshold

Why a round number fails

A shared threshold of 10 applied across a catalogue produces two failures at once.

• Fast movers alert far too late. Something selling 20 a day is gone within hours of crossing 10. The alert arrives after the decision point.
• Slow movers alert constantly. Something selling one a month sits under 10 permanently and emails you forever about a situation that is fine.

The second failure is the one that kills the system. Alerts that are usually noise train people to ignore all of them, including the one that mattered. An alerting setup everybody filters is worse than none, because it produces false confidence.

Working it out

Step 1: average units sold per day

Take units sold over a recent window and divide by the days in it. Ninety days is usually a reasonable window: long enough to smooth out noise, short enough to reflect current demand. Use a shorter window for anything trending quickly.

Step 2: days to restock

The honest elapsed time from noticing to having sellable stock. Not the supplier's quoted lead time. It includes your own approval time, their production time, shipping, customs where relevant and receiving. This is the number people underestimate most.

Step 3: a safety buffer

Cover for the days demand spikes or the supplier slips. Roughly a quarter to a half of your lead time demand is a workable starting point, higher for anything erratic. The formal version is in the safety stock formula.

Two worked examples

A fast moving basic

• Sells 12 a day
• Restocks in 14 days
• Lead time demand: 12 × 14 = 168
• Buffer at a third: 56
• Threshold: about 224

Set this product to 10 and you would be reordering with roughly twenty hours of cover left against a two week lead time. It would be out for almost a fortnight.

A slow moving accessory

• Sells 0.2 a day, about six a month
• Restocks in 10 days
• Lead time demand: 2
• Buffer: 2
• Threshold: about 4

Set this one to 10 and it alerts permanently while never actually being at risk.

A practical shortcut

Doing this properly for a thousand SKUs by hand is not realistic. It is also not necessary.

1. Calculate properly for your top 20 products by revenue. They are most of your exposure.
2. Group the rest into three or four bands by rough sales rate and lead time and give each band one sensible threshold.
3. Fix individual products when an alert proves wrong. An alert that arrived too late means raise it, one that has been noise for a month means lower it.

This converges on a good configuration within a couple of months with a fraction of the work.

Review it periodically

Thresholds go stale because the inputs change. Sales rates move with seasons, suppliers change lead times and products shift between fast and slow.

A quarterly pass over the top sellers is enough for most stores, plus a review before any period where demand changes shape. Also raise thresholds ahead of a known peak rather than during it, because that is when lead times stretch at exactly the moment demand rises.