It's 6:52 p.m. on a Tuesday. An order sits three days behind schedule in a warehouse. Nobody has told the customer yet, not because anyone decided to stay quiet, but because no one owns the moment when "we know" becomes "they know." Multiply that single order by 340 identical ones moving through the same delay, and you have a decision that will shape thousands of dollars in support cost and churn before midnight.
In my work with companies building customer-facing AI, I keep returning to a claim that sounds almost too simple: the highest-value message a company sends is often the one the customer never had to ask for. Not the fastest reply. Not the most polished chatbot. The message that arrives before the question does.
Most organizations get this backwards. They build faster response systems for inbound complaints and leave the outbound side, the proactive warning, as an afterthought, or worse, as a blast that goes out on everything and trains customers to ignore it. Both mistakes are expensive. The fix is not more AI. It's a test that tells you which events earn a proactive message and which ones don't.
The Silence Tax
Every operation has a gap between the moment it learns something has gone wrong and the moment the customer learns it. I call that gap the silence tax, because it behaves like one: it accrues quietly, it compounds, and it falls disproportionately on the people who can least absorb it.
Here is what the silence tax looks like in practice. A shipment slips. A warehouse system flags it internally. Support has no idea until a customer calls. The customer has no idea until the delivery date passes. In that gap, the customer's mental model of what's happening (did they forget me, did something break, is my order lost) is almost always worse than the truth. Silence doesn't read as neutral. It reads as evidence.
The tax is paid twice: once in the support contacts that a proactive message would have prevented, and once in the cancellations that anxiety produces on top of the ones the delay itself would have caused anyway. I'll walk through both with worked arithmetic later in this piece, but the direction is worth stating plainly first: waiting to be asked is rarely the cheaper option, even though it feels like the safer one.
The PACA Test
Not every bad or uncertain event deserves a proactive message. Send too many and you get the second mistake: a channel customers learn to skim past, which is worse than silence because it burns the trust you need for the messages that actually matter. So before any message goes out, I run it through four questions. All four have to be true, or I don't send it.
Predictable — do I know about this before the customer could reasonably find out on their own? If the customer would discover the problem at the same moment I would, a proactive message adds nothing.
Anxiety-weighted — does silence on this topic read worse than the truth would? Some events are minor enough that nobody spends emotional energy wondering about them. Those don't qualify, no matter how predictable they are.
Chase-inevitable — would this customer contact me anyway, given enough time? If the answer is yes, I'm not creating a new conversation by messaging first. I'm just choosing who starts it and on what terms.
Actionable — is there a real next step I can offer? Reschedule, substitute, wait, refund: something concrete, not a status update with nowhere to go.
A delayed order, an outage on a service someone is actively paying for, a slipped appointment, an out-of-stock item on a standing order: these tend to pass all four. A five-minute processing lag on a routine transaction almost never does. The question the test asks is whether staying quiet actually protects anyone.
Inside One Delay Event
I want to make the silence tax concrete rather than rhetorical, so here is a worked example. The numbers are illustrative: a composite built to keep the arithmetic clean and consistent, not a single client's live report, but every figure downstream follows from the inputs, and I've checked the math twice.
Start with 340 delayed orders, averaging $180 each, at a 35 percent margin. That's $63 of margin per order.
Run the reactive scenario first: the company says nothing and waits. In this example, 58 percent of affected customers contact support to ask what's happening: 197 contacts, at roughly $3.80 per contact to handle, for $749. Of those 197 who contacted support, 22 percent cancel (43 customers) at $63 of lost margin each, for $2,709. Total cost of staying silent on this one delay event: $3,458.
Now run the proactive scenario: a message goes out to all 340 customers as soon as the delay is known, naming the new delivery date and offering a choice. Sending it costs about $0.05 per message, or $17 total. Of the 340 who receive it, 9 percent still reply with a question (31 people) at $3.80 to handle, for $118. And cancellations drop to about 6 percent of the full 340, not 22 percent of those who complained: 20 customers, at $63, for $1,260. Total cost, including the send: $1,395.
The difference is $2,063 on a single delay event, about $6.07 per affected order. Nothing about the delay changed. The order still arrived three days late. What changed is who controlled the first conversation about it.
What the Message Must Say — and Must Not Become
The content of the message matters as much as the decision to send it. A proactive message that says "we're aware of the issue and are working on it" fails almost every purpose it's meant to serve, because it gives the customer nothing to do with the information. I hold every message to five requirements:
- Name the new time, not "we're aware." A specific date does the work that reassurance can't.
- Offer one thing the customer can do: reschedule, substitute, wait, or get a refund. Not a menu. One clear option, sometimes two.
- Provide one reply path, watched by an actual human, for the exceptions that don't fit the standard offer.
- Never send it on an event that fails any part of the PACA test. Passing three out of four is not enough.
- Send one update per state change, not one per hour. A delay that moves from three days to five days is one more message. A delay that hasn't changed is silence, correctly.
That last point deserves emphasis, because it's the one companies get wrong once they've seen the arithmetic above and decide proactive messaging is unambiguously good. Proactive messaging is good only when it clears that test, not by default.
The Noise Trap
The moment a proactive message pipeline exists, the temptation is to widen it: to flag more events, cover more edge cases, catch more of the "we should have told them" moments after the fact. This is how a genuinely useful system turns into an inbox customers learn to ignore.
The PACA test exists specifically to resist that pressure. I treat it as a gate. If an event is predictable but not anxiety-weighted, I don't send. If it's anxiety-weighted but the customer would never have known to ask, so there's nothing chase-inevitable about it, I don't send either. The discipline is the same discipline that made the delay example above worth $2,063: the message is only worth sending when the customer is genuinely better off receiving it than not.
The best customer message, in the end, is the one that made the phone call unnecessary. Not because the news was good (often it wasn't), but because the customer never had to sit with the not-knowing. That's the actual product. The warehouse delay, the price change, the appointment slip: those are the operational facts. Whether the customer experiences them as a betrayal or as a company that keeps its word is a separate decision, made entirely by whether someone spoke up before being asked.
Related evidence: Google's SRE book sets the bar for when a system should speak up unasked: an alerting rule should fire only where the condition it detects is urgent, actionable and actively or imminently user-visible, and it warns that where pages arrive too often people skim or ignore them outright, so that a real alert gets lost in the noise. (Google's SRE guidance on alerting only where the condition is actionable)
The FTC's CAN-SPAM compliance guide sets the legal floor under unsolicited commercial messages: every such message must offer a way to stop all marketing mail from the sender, and that opt-out must actually be honoured rather than filtered away. (the FTC's compliance guide on unsolicited commercial messages)
