Two things wearing the same clothes
Most store teams have one notification channel and two completely different kinds of information flowing through it. The first kind needs a human to do something now. The second kind is useful to know and useful to review, but nothing changes in the next hour because of it.
When you mix them, people learn to skim. Once people skim, the urgent ones get missed, and the usual response is to make the urgent ones louder, which trains everyone to skim harder.
The test we use: if a competent person cannot take a useful action within five minutes of seeing this, it is not an alert.
What passes the test
- Stock about to run out on something you are actively advertising. Action: pause the ad set, or pull stock from another location.
- A high-risk order over your value floor. Action: verify or hold before it ships.
- A fulfilment that has not moved in 48 hours. Action: chase the carrier before the customer chases you.
- A return request from a repeat customer sitting unapproved. Action: approve it, or decide not to and say why.
What fails it, and belongs in the daily summary instead
- Every new order. Nobody acts on an order simply existing.
- Daily revenue against target. Important, reviewable, not actionable at 11:04am.
- A product dropping below threshold that you are not advertising and can restock in a week.
- Refund rate creeping up. This needs a conversation, not an interruption.
An alert interrupts someone. If you are not asking for an action, you have spent their attention and bought nothing.
The demotion exercise
Take your current alert rules and, for each one, write the action in the imperative. "Pause the ad set." "Call the customer." "Move stock from the shop." If you cannot write the sentence, demote the rule to the daily summary.
Most teams demote between a third and half of their rules the first time they do this, and every single one reports that the remaining alerts started getting read.
What to do with the demoted ones
Do not delete them. The information was never the problem. The interruption was. Put them in a once-a-day digest that lands before your team starts work, and let people read it in the order they choose.
Store Huddle takes a position here: only two things interrupt anyone (low-stock products and high-risk orders) and everything else waits in Store pulse until someone looks. The point is not fewer facts. It is fewer interruptions carrying the same facts.
Applying the rule to what you already have
Take your current alerts and sort each into one of three piles. Most teams find the piles are wildly uneven.
Actionable in five minutes. A flagged order that can be held. Stock crossing a reorder point where somebody can place an order. A failed payment that can be retried. These earn their interruption.
Actionable, but not in five minutes. A supplier is late, a product is trending, returns on one line are climbing. Real information, genuinely useful, and nothing changes if you read it four hours later. These belong in a daily summary, not a notification.
Not actionable at all. Anything that reports a fact nobody can change. These should be deleted rather than rerouted, and deleting them is the single fastest way to make the remaining alerts matter again.
The exercise usually reveals that the first pile is much smaller than expected. That is the finding. A team receiving forty notifications a day where four are actionable has trained itself to skim all forty, and the four that mattered were skimmed with the rest.
The test that settles arguments
When somebody wants to add a new alert, ask one question: what will the person receiving this do differently in the next five minutes?
If the answer is a specific action, add it. If the answer is "be aware", that is a report, and reports belong somewhere you go rather than something that comes to you.
The question also works in reverse. If an alert has been firing for months and nobody can describe what they do when it arrives, it has already failed the test and everyone has quietly been ignoring it.
An alert is a request for someone to stop what they are doing. If nothing stops, it was never an alert.
What changes when you get this right
The measurable difference is not fewer notifications, though that happens. It is that people start reading them again.
A channel where every message demands action gets read carefully. A channel where one message in ten demands action gets skimmed, and skimming is indiscriminate: it does not distinguish between the nine that did not matter and the one that did. That is why alert fatigue is dangerous rather than merely annoying.
The related decision is who receives each one, which is a separate problem with its own failure modes, covered in who should see inventory alerts.
Where the five minutes actually goes
The rule assumes the recipient can act, which quietly depends on three things being true. When an alert fails in practice, it is usually one of these rather than the alert itself.
They have the authority. An alert about a flagged order reaching someone who is not allowed to hold orders has not saved any time; it has added a hop. Authority and alerting have to be designed together.
They have the context attached. If acting requires opening admin and finding the record, the five minutes becomes fifteen and the alert competes with whatever they were already doing. An alert carrying the order itself is a different object from one carrying an order number.
They are actually available. An alert routed to one person who is on a day off is an alert nobody receives. Name a backup for anything genuinely time-critical.
The exception worth allowing
One category earns an interruption despite failing the five-minute test: things that are cheap to act on now and very expensive to act on later.
A supplier confirming they can still fit your order into this week's dispatch. A customer asking to change an address before a label is printed. Neither is an emergency, and both have a window that closes.
The rule for these is not urgency but reversibility. If waiting removes an option that currently exists, treat it as an alert even when nothing is on fire.
Rebuilding your alerts from zero
Editing an existing set rarely works, because every alert has somebody who will defend it. Starting from nothing and adding back is faster and produces a much shorter list.
Turn everything off. Then add back only alerts where you can name the person, the action, and the window, in that order. If you cannot fill all three fields, it does not go back on.
Run for a fortnight and note what you wished you had known sooner. That list, which is usually two or three items, is your real alert set. Most teams doing this end up with between three and six alerts, having started with twenty or more, and report that they read all of them.
The uncomfortable part is that nothing breaks during the fortnight. The alerts nobody missed were the ones proving the point.
The one that should wake someone up
Almost nothing in a store genuinely warrants an out-of-hours interruption, and pretending otherwise means staff either ignore everything or lose their evenings.
The honest list for most stores is short: payments failing, the storefront being down, or a safety issue. Everything else waits, and saying so explicitly is what allows people to stop checking.
Whatever your version is, write it down and tell the team it is exhaustive. The absence of that statement is why conscientious people check messages at eleven at night about things that could have waited until Tuesday.
Common questions
What makes something an alert rather than a report?
Whether someone can act on it within about five minutes of it arriving. If the useful response is "note that and carry on", it is a report and belongs in a summary rather than a notification.
How many alerts is too many?
There is no fixed number. The signal is behavioural: once people skim the channel rather than read it, you are past the limit, and adding more makes the existing ones less effective.
What should we do with useful but non-urgent information?
Put it in a daily or weekly summary that people read at a fixed point. It keeps the value without spending an interruption on it.
Should alerts go to everyone?
No. Route to whoever can act. Sending alerts to people who cannot act does not add a safety net, it removes one, because everyone assumes somebody else is handling it.
How do we cut alerts without losing something important?
Ask what each one has caused anyone to do in the last three months. Anything with no answer has already been ignored, so removing it changes nothing except the noise level.