AI Growth Hub
← AI Email & CRM

Klaviyo Customer Agent Setup Guide: Omnichannel AI Support for E-Commerce

klaviyo ai-customer-service email-automation klaviyo

What Customer Agent Actually Does and Where It Stops

Klaviyo Customer Agent launched with chat and SMS support, then added WhatsApp and email in Q1-Q2 2026, bringing the total to four channels running on the same AI engine. A customer reaching out via WhatsApp and a different customer emailing about the same return request both get handled by the same underlying logic — same order data, same return policy rules, same Agent Guidance configuration.

The model is trained on interaction data from 193,000+ brands across 14 years on the Klaviyo platform. That scope matters because it means the system has seen a very wide range of post-purchase scenarios, not just a general-purpose language model with a retail prompt layered on top. For standard query types — tracking, returns, subscription changes — the handling accuracy reflects that specialization.

Customer Agent handles inbound service conversations only. It does not initiate outbound messages. If you want proactive WhatsApp marketing messages, that is Klaviyo Flows, not Customer Agent. The two systems are separate and do not share configuration.

Enabling the Four Channels

The recommended order for enabling channels is chat first, then SMS, then email, then WhatsApp. Chat and SMS require the least setup and have limited blast radius when you are still tuning Agent Guidance. Get the AI’s responses stable on the simpler channels before expanding.

Chat opens directly in the Klaviyo settings panel with no external account verification. It is the fastest way to test how the retail skills respond to real customer questions. Because chat conversations are synchronous, problems show up quickly — you will see gaps in the return policy coverage or awkward tone issues within the first few days.

SMS requires that your 10DLC or Toll-Free registration is already complete. Customer Agent takes over inbound SMS service conversations; outbound campaign and flow SMS messages are unaffected and run through separate logic.

Email requires a dedicated support address (for example, [email protected]) with DNS verification completed inside Klaviyo. Emails sent to that address route to the Customer Agent queue. Other sender addresses used for marketing campaigns are not affected.

WhatsApp has the most involved setup: you need to connect a Meta-verified WhatsApp Business Account to Klaviyo, which typically takes 3-7 business days for Meta’s review. Have your business verification documents ready before starting. Once connected, Customer Agent on WhatsApp supports rich media — images, documents — so you can send return shipping labels or tracking screenshots directly in the conversation thread.

Configuring the Retail Skills

Five retail skills are available out of the box: order tracking, order editing, returns processing, subscription management, and loyalty lookup. Each skill can be enabled or disabled independently, and you can configure them per channel. A common setup is enabling order editing on chat and WhatsApp (where the conversation is synchronous) but limiting SMS to tracking and returns only.

SkillTypical triggerDependency
Order tracking”Where is my package?”Shopify or carrier API connected
Order editingAddress or variant change before shipmentMust be within your cutoff window
Returns processingReturn request, label generationReturn policy document uploaded
Subscription managementPause, skip, frequency changeRecharge or Shopify Subscriptions integrated
Loyalty lookupPoints balance, redemption rulesLoyaltyLion or Yotpo connected

Returns processing requires the most preparation. You need to upload a return policy document in the Customer Agent settings — which return reasons are accepted, which products are excluded, the return window in days. The AI follows this document literally. It does not infer exceptions or exercise judgment about edge cases. If your policy has seasonal variations, like an extended holiday return window, write them into the document explicitly.

Order editing has a cutoff window: modifications are only possible before the order enters your fulfillment system. You set this window in Klaviyo. Once a customer’s edit request falls outside that window, Customer Agent tells them the order cannot be modified and then follows the Agent Guidance escalation rules to decide whether to hand off to a human agent.

Agent Guidance: Tone and Escalation Rules

Agent Guidance is where most of the meaningful configuration happens, and it is also the section teams most often skip. The defaults are generic. Getting the AI to sound like your brand and escalate at the right moments requires writing specific instructions.

Guidance has two parts: tone and escalation rules.

Tone instructions are plain text. Write them the way you would brief a new support hire: “Use short sentences. Avoid formal language. Never use ‘Dear Customer.’ When a shipment is delayed, give a concrete next step rather than apologizing.” Brand-specific restrictions go here too — competitor name avoidance, any regulatory language around certain product categories, and instructions for how to handle complaints about specific issues you know come up frequently.

Escalation rules determine when Customer Agent should stop attempting to resolve an issue and transfer to a human. Some configurations worth including:

  • Transfer when a customer has expressed frustration two or more times in the same conversation
  • Transfer when a return request exceeds a set dollar threshold (adjust to your average order value)
  • Transfer immediately when a customer asks for a human agent
  • Transfer immediately for fraud reports or payment disputes — do not let the AI attempt to handle these

Too-loose escalation rules mean the AI circles back on problems it cannot actually solve, eroding trust. Too-tight rules and you pay for Customer Agent without getting the volume reduction you expected. Review the escalation log after the first two to three weeks and adjust based on which escalations were necessary versus which ones the AI could have handled.

Multi-Email Profiles and Channel Identity Resolution

Multi-Email Profiles, also released in 2026, allows up to 5 email addresses per customer record with independent consent tracking for each. This directly affects how Customer Agent handles identity resolution on the email channel.

When an email arrives from an address not yet in a customer’s profile, Customer Agent searches for a matching record. If it finds one — for example, the customer’s Shopify order has the same name and a closely matching phone number — it can access the full order history and loyalty data without requiring the customer to provide an order number. If no match is found, the AI prompts for a verification credential (order number or the email address used at checkout), then associates the new address with the matched profile.

For cross-border e-commerce specifically, this matters because international customers frequently use multiple email addresses, and WhatsApp numbers may not match the email on file from their original order. Having Customer Agent configured to attempt identity resolution before asking for credentials reduces friction in the first message exchange, which is where a lot of customers drop off in traditional support flows.

Channel priority configuration determines how Customer Agent handles simultaneous contacts across channels. If a customer sends the same question via email and WhatsApp within a short window, the system responds through the highest-priority channel first and sends a brief acknowledgment on the other, avoiding duplicate full responses. For cross-border brands where WhatsApp is the dominant async channel, setting WhatsApp as the top-priority channel is usually the right call.

Measuring Whether the Setup Is Working

The Customer Agent analytics section in Klaviyo tracks a few numbers that tell you whether the configuration is actually performing.

Autonomous resolution rate is the share of conversations the AI completes without human intervention. For standard query types — tracking, returns, subscriptions — a well-configured setup should land between 75% and 85%. If you are seeing under 60%, check the escalation rules first. Overly conservative escalation thresholds are the most common reason for low autonomous rates. After that, look for recurring query types the AI is failing to classify correctly.

First response time should be near-instant on chat and WhatsApp. If email first response time exceeds 5 minutes, the issue is usually DNS configuration or a mail forwarding delay, not the AI itself.

Customer satisfaction scores are available if you enable the post-conversation survey in Klaviyo. Comparing satisfaction scores between AI-handled conversations and escalated conversations gives you a read on whether the AI is genuinely resolving issues or just responding quickly to questions that still require follow-up. A large gap between the two numbers points to gaps in the retail skills configuration or the return policy document.

Pull the low-score conversations every two weeks and read through them. The patterns that appear there are more actionable than aggregate metrics. Most issues trace back to either the Agent Guidance tone instructions being too vague or a retail skill’s policy document missing coverage for a scenario that is common in your actual order stream.

阅读本文中文版: Klaviyo Customer Agent 实战：四渠道全开，零人工处理售后问题

Related Articles