The short answer
- Four customer messages is the right number: confirmed, proof, in production, shipped
- Every one should carry the delivery date, even when it has not changed
- Internal alerts should be about exceptions, not about everything happening
- An alert that fires constantly gets ignored, which is worse than none
- Review what you send every few months and delete what nobody acts on
Notifications are the cheapest customer service improvement available and the easiest to get wrong in both directions. Send too few and people phone to ask where things are. Send too many and they stop reading, which produces the same phone call with more effort behind it.
The two failure modes
Silence. A customer orders, receives one confirmation, and hears nothing for eight days. Silence is ambiguous, and people fill ambiguity with the worst plausible explanation — which is why the majority of status calls come from the gap between approval and shipping.
Noise. A message for every state change — order received, artwork received, artwork checked, scheduled, screens made, on press, cured, packed — trains the customer that most of your messages do not matter. When one does, it is skimmed with the rest.
The judgement is straightforward: send a message when the customer's mental picture of the job would otherwise be wrong or stale. Everything else is noise dressed as service.
Every message you send that did not need sending makes the next one less likely to be read.
Customer messages
| Moment | Must include |
|---|---|
| Order confirmed | What was ordered, and the delivery date |
| Proof ready | Approve link, and the approval deadline |
| In production | Date restated — nothing else needed |
| Shipped or ready | Tracking, and arrival date |
Add one more only when something changes: a delay, a substitution, a short count. Those are genuinely news and should arrive quickly, with an option attached rather than an apology alone — the shape that works when a style has been discontinued mid-order.
The date repeating in all four feels excessive when you write them and is the most valuable element. The date is what the customer cares about and the thing they are least likely to have retained from the last message.
Sending these from the job's own status changes rather than by hand is the only way it stays consistent. That is what our back-office management does with order status — the message goes because the work happened, not because somebody remembered.
Internal alerts
Different purpose entirely. Customer messages report normal progress; internal alerts should report exceptions — things that need somebody to do something.
Worth alerting on
- A proof unapproved and the deadline approaching
- A job at risk of missing its date
- A supplier delivery short or late
- An order placed for a style now out of stock
- A payment overdue past your threshold
- An integration that has stopped syncing
Not worth alerting on
- Every new order — that is what the order list is for
- Normal status progression
- Anything nobody would act on
- Anything already visible on a screen people look at
The last exclusion is the one shops get wrong. If the production board already shows what is on press, an alert saying the same thing is duplication. Alerts are for things you would otherwise not see.
The integration line deserves emphasis. An integration that fails silently is worse than no integration, because you have stopped doing the manual check that would have caught the problem — so it must alert, loudly.
Choosing a channel
| Message | Channel | Why |
|---|---|---|
| Order confirmation | Referred back to; needs detail | |
| Proof ready | They need to see and click | |
| Approval overdue | Text or call | Time-sensitive |
| Shipped | Email, or text if urgent | Tracking is a link |
| Something has gone wrong | Call | Bad news is not an email |
| Internal exception | Wherever the team already looks | Not a new inbox |
The bad news row is a rule worth holding to. A delay or a substitution delivered by email reads as avoidance; the same information by phone, with an option ready, is handled in two minutes and usually strengthens the relationship. Follow the call with an email confirming what was agreed.
For internal alerts, use whatever your team already has open. An alerting system that requires people to check a new place will be checked for a fortnight.
Writing them
Short, specific, and never generic. The most common failing is a message that describes activity without committing to anything.
Instead of / send
- "Your order is being processed" → "48 navy tees, front print. Proof by Wednesday, delivery Friday the 14th."
- "Please review the attached proof" → "Proof attached — approve by Thursday and we hold the 14th."
- "Your order has progressed" → "On the press today. Still on for the 14th."
- "Your order has shipped" → "Shipped, tracking 1Z…, arriving Thursday — a day early."
Each of those replaces a status with a fact and a date. The second also carries a consequence, which is what makes approval times fall — a request becomes a decision with a cost attached.
Test deliverability to a free webmail account, not to your own inbox. Transactional email from a small domain lands in spam more often than people expect, and a confirmation nobody receives is worse than none because the customer assumes the order failed.
Tuning it
Review every few months and delete what nobody acts on. Notification sets accumulate — somebody adds one for a specific situation, it stays forever, and after two years the team ignores the lot.
The review
- List every alert currently firing, internal and external
- For each, ask: when did anyone last act on this?
- Delete anything that fires more than a few times a week without action
- Check status calls are still falling; if not, the gap has moved
- Confirm customer emails are still being delivered
The second question is the whole test. An alert that fires daily and is dismissed daily is not information, it is background noise — and it is actively harmful, because it teaches people to dismiss the category. Cutting the list back is usually more valuable than adding to it. The rest of the software archive assumes you send what people act on.


