Klaviyo K:AI Spring 2026: Composer Builds Campaigns in One Prompt, Customer Agent Handles Returns
Klaviyo’s Spring 2026 release moved two previously beta features into general availability: Composer, which generates a complete automation flow from a single text prompt, and Customer Agent, a virtual support agent that handles routine tickets without human involvement. Both fall under the K:AI umbrella, but they solve different problems and suit different team sizes. This is a feature-by-feature breakdown of what each does, how to set it up, and which one to prioritize.
Composer: how one-prompt flow generation actually works
Composer sits inside the Flows builder. You describe the flow you want in plain text, and it generates trigger conditions, wait steps, email templates, and SMS copy in one pass. The underlying model was trained on flow structure data from 193,000 brands on the Klaviyo platform, which is why it performs reasonably well on common patterns like cart abandonment, welcome sequences, and win-back series.
To access it: go to Flows, click “Create Flow” in the top right, then select “Build with AI (Composer).” If you do not see the option, go to Account Settings, find “K:AI Features,” and request early access. Approval typically takes 3 to 5 business days.
Generation takes 10 to 15 seconds. The output is a draft flow with all steps populated, which you review, edit, and activate. The catch is that Composer only reads what you put in the prompt box. It has no access to your account’s historical data, existing flows, or brand guidelines. If your prompt is vague, the output fills in every unknown with generic defaults.
Writing prompts that get usable output
The difference between a prompt that produces a workable draft and one that produces something you throw away comes down to four inputs: audience filter, number of messages with timing, promotion mechanics, and brand tone. Leave any of these out and Composer guesses.
| Scenario | Weak prompt | Strong prompt |
|---|---|---|
| Cart abandonment | ”Send emails to cart abandoners" | "3 emails + 1 SMS to users who added to cart but did not purchase within 24 hours. Email 1: no discount, just ask if they have questions. Email 2 (48h later): 10% off. Email 3 (72h later): 15% off, mention limited stock. SMS on day 4. Tone: direct, not soft. Brand sells outdoor camping gear.” |
| Welcome series | ”Welcome new subscribers" | "5-email welcome series for new subscribers, no discounts, focus on brand story and product curation philosophy, 2-day intervals, final email drives sign-up for loyalty program. Audience: women 25-35 interested in sustainable consumption.” |
| VIP win-back | ”Win back lapsed VIPs" | "Users who spent over $500 in the past 12 months but have not ordered in 90 days. 2 emails: Email 1 shows new arrivals in categories they previously bought. Email 2 (10 days later) gives an exclusive 20% code, valid 7 days, framed as VIP-only.” |
Strong prompts specify who gets into the flow, how many messages, the gap between each, discount values, and at least one tone signal. All four, every time.
What Composer does not handle
Composer does not check your existing flows for conflicts. If a user is already in a cart abandonment flow and Composer generates a new one with overlapping triggers, they can enter both. The fix is to add an “Is in Flow” filter to the new flow’s trigger before activating. Also, template copy comes out in English regardless of prompt language. Non-English brands need to replace all copy manually after generation.
Customer Agent: setup and what it can actually do
Customer Agent is a virtual support agent you attach to a Klaviyo form entry point or your own customer-facing interface via API. It handles four skill categories: return requests, order modifications (address changes, SKU swaps), subscription changes (pause, cancel, frequency), and loyalty point inquiries. According to Klaviyo, those four categories cover 60% to 70% of routine support tickets for most brands.
Setup steps:
- In your Klaviyo account, go to “Customers” and open “Customer Agent (Beta)”
- Connect your order management system — Shopify and BigCommerce work without issues; Magento is supported; WooCommerce is still in beta
- Configure escalation rules: define which situations transfer to a human agent (examples: refund amount over $100, sentiment detection flags “angry,” two consecutive unresolved interactions)
- Choose where the Agent appears: Klaviyo form embed, website widget, or API integration with your existing helpdesk
- Run the Agent in test mode through at least 50 simulated conversations before going live, specifically checking that escalation triggers fire when expected
Beyond escalation rules, you can configure whether the Agent can proactively offer a discount to retain a customer requesting a refund (for example, offer 10% off before processing the return), and which operations it must refuse outright (for example, no address changes after an order ships).
Pricing and which team size should go first
Customer Agent bills per conversation, not per seat. A conversation is one complete interaction between a user and the Agent, regardless of how many messages it contains. The rate runs from $0.08 to $0.12 per conversation depending on your plan tier. At 5,000 monthly support tickets with 60% handled by the Agent, you’re looking at roughly $240 to $360 per month.
| Monthly ticket volume | Prioritize | Reason |
|---|---|---|
| Under 500 | Composer | Support load is manageable; automation quality has more revenue impact |
| 500 to 2,000 | Either; start with Composer | Support is a growing problem but still coverable by a small team |
| 2,000 to 5,000 | Customer Agent first | Support costs are already meaningful; Agent ROI is clear |
| Over 5,000 | Customer Agent, no question | The cost of human agents handling routine tickets outweighs setup friction |
One thing that gets skipped in most setup guides: Customer Agent’s actual resolution rate depends heavily on the cleanliness of your order data. If your Shopify instance has duplicate SKUs, inconsistent address formats, or unmapped order statuses, the Agent escalates more than it resolves. Before going live, spend a day auditing your Shopify data quality. The 60-70% deflection rate assumes reasonably clean data.
Next Best Product expanded to SMS, RCS, and WhatsApp
This release also extended Next Best Product recommendations beyond email. In Flows, when you insert a recommendation block, you now have a “Multi-channel NBP Block” option that adapts format based on the channel: image and text in email, text-only link in SMS, rich media card in RCS (if your carrier plan supports RCS sends), and WhatsApp-formatted content for WhatsApp Business integrations.
The recommendation logic itself did not change — it still runs on purchase history and browse behavior. The expansion is purely about channel coverage. For brands in replenishment categories (skincare, pet food, coffee subscriptions), adding SMS-delivered recommendations to post-purchase flows is worth testing. The format constraint in SMS means the product name and a short descriptor need to fit in roughly 100 characters before the link, so product naming matters.
Pre-launch checklist before enabling either feature
For Composer: audit your active flows before activating anything generated by Composer. Identify any flows with overlapping trigger conditions and add entry filters to the new Composer-generated flows. If you run multiple Klaviyo accounts (for example, separate accounts per market), note that Composer is account-scoped — it does not learn from your other accounts.
For Customer Agent: write out your escalation scenarios explicitly before touching the configuration UI. Do not rely on default thresholds — they are set conservatively and will not match your brand’s policies. Before launch, run manual tests on two scenarios specifically: a large refund request (over your escalation threshold) and a user who expresses repeated frustration across multiple messages. Confirm the Agent escalates at the right moment, not one message too late.
After both features go live, check Flows analytics and Customer Agent conversation logs daily for the first two weeks. Look for unexpected escalation spikes (signals your data or rules need adjustment) and Composer-generated flows with abnormal unsubscribe rates (signals the template copy needs revision). After two weeks, weekly reviews are enough.
Related Articles
Email Deliverability in the Gmail Gemini AI Era: How Content Format Affects Inbox Placement
After Google I/O 2026, Gemini AI inside Gmail now summarizes, prioritizes, and filters emails before users see them. For cross-border e-commerce brands, this means email content quality—especially structural clarity and front-loaded value—directly affects inbox placement. Here's how to adjust your email writing and formatting.
Mailchimp Analytics AI Is Live: Ask Your Email Campaigns Anything in Plain Language
Mailchimp launched Analytics AI on May 28, 2026 — a conversational analytics agent that answers natural language questions across Shopify/WooCommerce/Wix data, such as which flow drives the most revenue or why open rates dropped last week, with specific next-step suggestions. Now live for all paid users.
Klaviyo vs Mailchimp vs Attentive 2026: Which One Is Right for Your DTC Cross-Border Brand
All three tools updated heavily in May-June 2026: Klaviyo launched Composer and Customer Agent, Mailchimp shipped Analytics AI and WooCommerce/Wix integration, Attentive released Brand Voice 2.0 and Reporting Agent. Comparison matrix from a DTC cross-border perspective with recommendations by brand GMV size.