When should AI customer service make proactive contact with customers?

AI should make proactive contact when it has a specific, time-sensitive piece of information the customer would want to know before they contact you, and when the cost of getting the timing or content wrong is low. That is mostly delivery updates, appointment confirmations, expiring offers the customer opted in for, and remediation messages after a known outage. It is rarely "we noticed you might want to know..." messages, which read as marketing and erode trust.

A customer's parcel is delayed by two days. The AI knows this before the customer does. A message that says so plainly, and offers a sensible next step, is service. The same channel used to push an unrelated promotional offer ten minutes later is not service. The first builds trust; the second loses it.

What people in the field are saying

DCX Newsletter has a piece on what CX leaders should fix before they get too ambitious with AI: "What CX leaders need to fix before...". The proactive question shows up there as a litmus test for whether the CX function is in service of the customer or in service of campaign metrics.

What counts as good proactive?

Four kinds, mostly. Delivery and appointment updates ("your parcel is now arriving Friday"). Account events the customer should know about ("a new sign-in from a new device"). Remediation after a known issue ("our system had an outage that affected you; here is what we are doing"). Renewals or expirations the customer asked to be reminded of. All four share the property that the customer would have wanted to know.

What counts as bad proactive?

Anything where the AI is reaching out because the company would like the customer to do something, dressed up as service. Win-back messages, upsells, "we noticed you abandoned a cart" prompts, and personalised promotions all live here. They can be legitimate marketing, but they should not be sent under the support brand, and they should not be the AI's idea.

What about silent messages that just inform?

A surprising amount of value lives there. A short message confirming an action was completed, an update when a long-running request progresses, an alert that something the customer was waiting for has arrived. These rarely require a reply and rarely get a complaint. They count as proactive too, and AI handles them well.

What is the test before any proactive AI message goes out?

Two questions. Would a competent human agent send this, today, unprompted? And would the customer thank you if you sent it? If both answers are yes, the message earns its place. If either is no, the message is being sent for the business and dressed as service. Customers can tell.