← All posts

Your alert was set for a smaller shop

Most rules you set about your shop are a number and a comparison: tell me if sales fall below this, reorder when stock drops under that, we should have at least this many reviews. The comparison keeps working for years. The number was true on the day you typed it, and a growing shop walks away from it — quietly, while every report stays green.

Published — written and checked by hand.

There is a kind of rule that fails by succeeding. You set it when things were smaller, it was a sensible number then, and it has passed every check since. It is still passing now. The reason it is passing has changed, though: it used to pass because the shop was healthy, and now it passes because the shop has grown so far past the line that almost nothing could push it back under.

Nothing about it looks wrong. The rule runs, it reports, the report is green. The number is simply describing a shop you no longer have.

A rule per item gets stronger; a rule about the total gets weaker

This is the distinction worth taking away, because it tells you which of your rules need re-reading and which ones you can leave alone for a decade.

A rule applied to every item strengthens as you grow. 'Every product needs at least three photos' is checked once per product: ten products means ten checks, four hundred products means four hundred. Adding stock adds assertions. You never have to touch the number, because the number was never about the size of the shop — it was about what one product owes.

A rule applied to the total weakens as you grow, and it does so without any help. 'We should have at least fifty reviews' was a real bar at sixty reviews. At two thousand it is not a bar at all; you could lose nineteen hundred and forty of them and the rule would still report that everything is fine. Same sentence, same comparison, same green tick, and it has stopped being a safeguard. It decayed by arithmetic rather than by neglect.

So the question to ask of any threshold is not 'is this number right?' but 'what happens to this number when the shop doubles?'. Per item, it holds. Against a total, it halves in value and nothing tells you.

Where this sits in a real shop

The pattern turns up wherever a rule was written as one number about everything you have:

  • A sales or traffic alert. 'Tell me if weekly revenue drops below X' set when X was most of a good week. Three good years later the same X is a catastrophe alarm, and the ordinary bad week it was meant to catch sails past it.
  • A reorder point. Twenty units was two weeks of cover once. If sales tripled and the number did not, it is now three days of cover wearing the label of two weeks.
  • A review or rating target for the shop as a whole. A total that grows on its own makes the floor easier to clear every month, which is the opposite of what a target is for.
  • A refund or complaint budget stated as a count. 'Fewer than ten complaints a month' is strict at two hundred orders and generous at four thousand, and the sentence does not change when the orders do.
  • An ad budget guard expressed in money rather than in return. A daily cap set against a small spend becomes a formality once the account is larger, and a cap that never binds is not a control.
  • A stock-breadth rule. 'At least thirty products listed with full descriptions' is a real standard in a catalogue of forty and a rounding error in a catalogue of six hundred.

Every one of those is a number that was measured honestly, once, by somebody who was paying attention. That is what makes the decay hard to see: there is no mistake in the history to find.

This is a different failure from a check that opened only some of your products and reported on the shop. There, the rule was fine and the list was short. Here the list is complete, every item was looked at, and the threshold the results are compared against is the part that aged.

Our own version of it, this week

We run an automated check over every public page of this site before anything ships, and it holds a few dozen numbered thresholds — minimum words on a page, maximum length for a title, how many internal links a page owes. This week we read the thresholds themselves instead of the results, and sorted them by that per-item-or-total question.

Twenty of them are per-item, and all twenty are fine for the reason given above: they are asserted once per page, so every page we publish brings its own copy of the check along with it.

Four are totals, and three of those had stopped doing much. One demanded a minimum number of links between posts; the site had grown to seventy of them against a floor of thirty-one, so the rule would have shrugged at losing more than half. Another counted a particular kind of heading across the whole site, with a floor of ninety against a hundred and fifty-five actually there. The comment beside that one recorded the day it was measured and the figure it was measured at — a hundred and fifty-two — and the very run that read the comment printed a hundred and fifty-five. It had drifted in a day.

The fix was not to type bigger numbers, because a bigger number ages exactly the same way. Each of the four is now written as a rate against something that grows on its own — so many links per post, so many headings per page — and the threshold recalculates itself every time the site gets larger. We kept the old hand-set figures as a floor underneath, so nothing we had already earned can be given back.

Then we added one more check whose only subject is the other checks: every numbered threshold in that file now has to declare whether it is a per-item rule or a total, and any total that has not been written as a rate fails the build. It caught a threshold on its first run — one the new check had itself introduced a few lines earlier, which is a tidier proof than we would have arranged on purpose. That is a relative of the habit of making sure a stated figure is something a check can actually verify.

What to do with this in an hour

You do not need tooling for this. You need to find your numbers and ask one question of each.

  • List every rule you have that contains a number — alerts, reorder points, targets, budget caps, anything you or a tool checks on a schedule. Most shops have between five and twenty and have never seen them in one place.
  • Beside each, write what it is compared against: one product, one order, one week — or the total of everything you have.
  • For the totals, work out roughly when you set it and how much bigger the shop is now. If the shop has doubled and the number has not moved, the rule is now half a rule.
  • Rewrite those as a rate: per product, per order, per hundred orders, as a share rather than a count. 'Complaints under one in two hundred orders' survives growth; 'fewer than ten complaints' does not.
  • Keep the old number as a minimum beneath the rate, so a quiet month cannot lower a standard you have already met.
  • Write the date and the measurement next to each one. Not for tidiness — so the next person can see at a glance how old the number is, which is the only clue that it might have aged.

The uncomfortable part is that none of this shows up as a failure. A decayed threshold does not break anything or alert anybody; it reports good news slightly more often each month until it reports nothing else. The only way to find one is to go and read the number instead of the result.

Lights up

Read enough? Watch your own product.

One photo in — a cinematic product video and a full set of marketing photos out. Your first one is free.

Free preview · no cardSee it filmed — free