AI customer service belongs where the contacts are simple, repetitive, and have a clear answer the AI can ground in a system or knowledge base. That is most chat and most digital messaging, a meaningful share of email, a narrow slice of voice, and almost none of in-store or in-person. The principle is the same across channels; the share that fits varies a lot.
A customer support leader is asked where to start with AI. The honest first question is not which AI but which channel. The answer is not the same for chat, voice, email, social, and in-app. Each channel has a different mix of contacts, a different latency tolerance, and a different cost of getting the AI's answer wrong.
What people in the field are saying
Blake Morgan's "Experience is everything: the CX strategy..." connects the channel question to the wider CX strategy, arguing that AI placement should follow what the channel is best at. DCX Newsletter touches the same theme in "Is your AI strategy improving CX".
How do chat and digital messaging look?
They are the AI's strongest channels. The customer's input is text the AI can read. The reply can be drafted, edited, and grounded in retrieved content before sending. The customer's latency tolerance is high (a few seconds is fine). The cost of an off reply is a re-edit, not a hold time. Most of an AI rollout's volume should land here.
What about voice?
Voice AI is strong on a narrow band of calls: routine status checks, simple bookings, repeated short questions. It is weaker on long, emotional, or accented calls, where latency, audio quality, and ambiguity matter more. Deploy voice AI where the call profile fits, not as a blanket rollout.
What about email and social?
Email is well-suited to AI drafting with a human review step: the customer's latency tolerance is hours, and the cost of editing is low. Pure AI send-without-review on email looks tempting and usually loses to human-in-the-loop quality. Social messaging is closer to chat for AI fit, but the public stakes are higher; a wrong AI reply on social can become a screenshot.
How do you pick where to start?
Take last year's contacts by channel. Score each channel for AI fit on three axes: how often the contact has a clear answer, how high the latency tolerance is, how bad it is when the AI is wrong. The channel that scores highest on all three is the starting point. Chat is almost always that channel.
Related: the question on voice AI in the contact centre, the field note on the CX strategy gap, and the glossary explainer on conversational AI.