When the AI has a specific, time-sensitive piece of information the customer would want to know, it reaches out: a delivery delay, an unusual sign-in, a maintenance window, a refund that has cleared. The new ingredient on this rung is that the AI is initiating the contact, not reacting to one. Setting the bar for what counts as worth reaching out for matters more than the technical capability to reach out.
A customer's parcel is now arriving two days late. The system knows this before the customer does. Before AI, that fact sat in a logistics dashboard until someone scheduled a batch message to everyone affected. The customer learned by checking the tracking page, or by waiting and wondering. The AI version sends one short message to the one customer whose parcel is delayed, with the specifics and a next step.
How this used to be a decision tree
The old approach was a campaign tree. Marketing or CS ops picked a segment ("all customers in zone X with orders in the last seven days") and scheduled a generic notification. Branches were segments, not situations. The customer who was affected got the same message as customers in the segment whose orders happened to be on time. The customer treated the message accordingly: marketing, ignore.
Why AI doesn't make this a decision tree anymore
The AI watches the system state for each customer and reaches out only when that specific customer has a specific reason to be reached. The customer's situation drives the contact, not a campaign calendar. The message says "your order #12345 is now arriving Friday; here is the tracking link" because that customer's order is now arriving Friday. It does not say "some orders in your region may be delayed."
What people in the field are saying
DCX Newsletter's "What CX leaders need to fix before..." argues that the proactive-contact question is a litmus test for whether the CX function is serving the customer or serving its campaign metrics. Good proactive contact is the customer thanking you; bad proactive contact is marketing under the support brand.
How does proactive AI contact look today?
The AI consumes event streams from the back-end systems (logistics, account, billing) and a policy that defines what events warrant a contact and to whom. When a qualifying event lands, the AI composes the message grounded in the specifics, picks the right channel for that customer, and sends. Most of the work is the policy. The reach is the easy part.
What does it take to make this work?
Event streams the AI can subscribe to (delivery status, account events, billing, system status). A policy for what events justify proactive contact and which channel to use. A clear distinction between service messages (the customer would thank you) and marketing messages (the business would thank you). An opt-out the AI respects. Audit logs of every outreach for compliance.
Where does this go wrong?
Service brand used to push marketing dressed as service. Over-contact ("we noticed you have not used the app in two days") that trains customers to mute the channel. False alarms (the AI sends a delay notice that was later resolved before the customer read it). Timing: messages that arrive at 3am because the system event landed at 3am.
Which tools handle proactive contact?
- Sierra: workflow-aware outreach across channels.
- Decagon: event-driven action-taking including outbound.
- Cognigy: enterprise platform with multi-channel proactive flows.
- Parloa: voice-first outbound including service contacts.
- Gladly: one conversation thread per customer makes proactive natural.
How would I start doing this?
Pick the most useful event the customer would unambiguously thank you for being told about (a clearly material delivery delay is the canonical one). Write the policy on one page: which event, which message, which channel, which exceptions. Send to a small cohort first. Read the responses. Tighten before adding the next event.
Next: the AI handles cases where its own earlier action is being disputed. Recovering from an AI mistake the customer is disputing.