Klaviyo Marketing Agent Launches as Customer Agent Goes GA: 14-Day Autonomous CRM Pilot

Breaking: Klaviyo flips the switch on autonomous marketing

On September 25, 2025, Klaviyo introduced Marketing Agent and released Customer Agent to general availability at its K:BOS event. The company positioned both as built-in teammates inside its B2C CRM so growth and CX teams can plan, create, and launch campaigns in minutes while service conversations convert into revenue. Read the official details in the Klaviyo newsroom announcement.

Why this is a real shift for growth teams

Most AI add-ons live outside the workflows that decide what gets sent, to whom, and when. Klaviyo’s agents operate inside the CRM that already holds identity, consent, and purchase history. That placement unlocks two practical upgrades for teams under pressure to drive revenue with fewer hands on keyboards:

• Autonomous assembly, not just assisted writing. Marketing Agent assembles a strategy, builds flows and campaigns, and drafts forms that are ready for approval. You provide guardrails, it generates assets and schedules them.
• Closed-loop learning tied to actual sales. Customer Agent handles real customer conversations, posts back outcomes, and exposes signals that can shape future campaigns and segmentation.

The workflow shifts from starting with a blank brief to editing an AI-proposed plan. Routine chats resolve automatically, and escalations go to humans with full context.

What Marketing Agent does today

Klaviyo positions Marketing Agent as an autonomous teammate focused on speed to launch and on-brand execution. In practice:

• Start with a URL. Enter your site, and the agent pulls catalog data, brand voice cues, and performance history. It returns a multi-week plan, not just a single email.
• Ready-to-send assets. Drafts email and SMS, activates must-have flows like welcome and abandoned cart, and proposes an on-site form with targeting rules. One-click approve or request edits.
• Guardrails and approvals. Define tone, legal statements, offer limits, and excluded SKUs. The agent respects constraints and routes exceptions to human review.
• Continuous idea generation. Weekly proposals tied to seasonality, promotions, and observed trends keep calendars fresh.

Because the agent sits inside the CRM, it can evaluate impact using the same attribution and revenue models you already trust.

What Customer Agent adds across the journey

Customer Agent is a 24-7 AI assistant for service and shopping across web chat, SMS, email in beta, and WhatsApp coming soon. It answers questions, recommends products, adds to cart, and escalates to a human with full conversation history when needed. Klaviyo outlines channels and setup on the Customer Agent capabilities page.

Key unlocks for growth teams:

• Support signals enrich marketing. Conversation topics and objections become inputs for your next send.
• Cart actions from chat. Adding products and applying promotions turns reactive service into proactive commerce.
• Faster handoffs. Escalations include transcript and customer context, reducing time to first response while protecting satisfaction.

Run a 14-day pilot to prove lift before you scale

You do not need a long transformation project. You need two weeks, an approval playbook, and a clean experiment.

Pilot goals and success criteria

Pick three headline metrics and put them in the brief so every stakeholder is aligned:

1. Time to launch from brief to send for a net-new campaign or form.
2. Revenue per send attributed revenue divided by messages delivered for pilot campaigns.
3. Unsubscribe rate total opt-outs divided by delivered, monitored per message and by segment.

Optionally track assisted resolution rate for Customer Agent and list growth from the new on-site form.

For adjacent examples and discipline, see our 14-day GTM pilot playbook.

Day 0 setup, guardrails before go-time

• Connect data and content: product catalog, help center or FAQs, shipping and returns policies, discount rules, seasonal calendars.
• Define brand and legal guardrails: tone examples, offer ceilings, sensitive categories to exclude, required disclaimers.
• Create roles and approvals: who can auto-approve, who can approve with edits, and who must review promotions or legal language.
• Baseline benchmarks: current time to launch, typical revenue per send by tier, unsubscribe rates for similar sends.

Days 1 to 3, establish the control

• Choose two upcoming moments that matter, such as a seasonal drop and a content newsletter.
• Ship the first moment with your current process and timing as the control. Record time to launch.
• Publish your current on-site form variant to a 50 percent holdout for the duration of the pilot.

Days 4 to 7, turn on Marketing Agent for assembly

• Feed the agent a single URL and weekly priorities. Ask for a 2-week plan with one email, one SMS, and one form.
• Approve with edits. Require human proof for subject lines, discounts, and any claims. Log edit count to track creative quality over time.
• Launch to a matched segment mirroring the control audience. Record time to launch and revenue per send.

Days 8 to 10, iterate and expand to SMS

• Use the agent’s weekly ideas to produce one incremental send without spiking frequency.
• Add an SMS companion. Apply quiet hours, frequency caps, and country-specific compliance. Keep opt-out instructions clear.
• Replace the on-site control form for half of traffic with the agent-generated form. Use conservative targeting on high-intent pages.

Days 11 to 14, activate Customer Agent and capture insights

• Deploy Customer Agent in web chat with conservative scope: product Q&A, order status, and policy answers. Route high-risk topics to humans.
• Turn on transcript logging and tag common questions. Push the top objections into your next send via dynamic blocks or a short FAQ.
• Measure list growth from the new form and compare against the holdout. For Customer Agent, measure resolved conversations and escalation rate without optimizing for speed alone.

At the end of Day 14, you will have a like-for-like comparison on time to launch, revenue per send, and unsubscribes, plus directional reads on resolution and form performance.

The approval playbook to keep you safe while you scale

• Ownership and roles. CRM owner approves segmentation and scheduling. Brand or creative lead approves copy and design. Legal approves promotions and claims.
• Content tiers and gates. Tier 1 evergreen copy can auto-publish with daily sampling QA. Tier 2 campaigns require two-person review. Tier 3 promotions or claims require legal.
• Offer guardrails. Set absolute discount and spend thresholds plus excluded SKUs. The agent cannot exceed them.
• Compliance. Require explicit opt-in for SMS, apply quiet hours, and enforce country-specific rules. Agents can propose sends but not override compliance toggles.
• Handoff standards. Escalations must pass full transcript, customer ID, and any cart actions so support can reply in the same channel.

Bake these into platform permissions and templated disclaimers rather than relying on policy decks.

Implementation checklist, what to connect and who to involve

• Data sources. Product catalog feed, returns and warranty pages, shipping tables, promo rules, and any trusted review content.
• Channels. Email, SMS, and web chat on day one. Consider enabling RCS if your audience warrants it. WhatsApp is on the near-term roadmap, so stage approval logic for that channel.
• People. CRM owner, lifecycle marketer, copy lead, CX lead, and an analyst for measurement. Loop legal and security in during Day 0 so they are not a bottleneck at Day 13.

How to measure the pilot with statistical discipline

• Matching. Build pilot and control segments with identical eligibility rules and similar historical engagement. Exclude recent purchasers if that is your norm. Confirm list size and average order value are within 5 to 10 percent.
• Attribution. Use the same attribution window and settings for control and pilot sends.
• Unsubscribe context. Monitor opt-outs at the message level and aggregate by subscriber tenure so volatility among new subscribers does not mask copy or frequency issues.
• Time to launch. Measure from brief to scheduled to capture assembly lift.

If volume allows, add a small holdout to validate incrementality. Otherwise, match segments tightly and use directional reads.

What good looks like after two weeks

• Time to launch drops from multiple days to same-day for at least one email and one SMS send.
• Revenue per send for the agent-assembled campaign is within 5 percent of control or better, with equal or lower unsubscribes.
• The agent-generated on-site form performs within 10 percent of control or better on submit rate, with equal or lower bounce rate.
• Customer Agent resolves a meaningful share of routine conversations without lowering satisfaction, and escalations include full context.

Treat these as thresholds to move from pilot to production. If you miss one, tighten guardrails, add more brand voice examples, and repeat a short run.

Extend the win beyond email

Once you clear thresholds, widen the aperture:

• Email to SMS. Use SMS for high-intent reminders, shipping updates with cross-sell, and short promos. Keep unified frequency caps across channels.
• Forms. Deploy agent-generated forms for seasonal capture on high-traffic pages, then route new subscribers into an agent-assembled welcome series.
• CX surface. Enable Customer Agent deeper in the journey with proactive prompts tied to stock alerts or size questions. Keep handoff logic simple and fast.

For adjacent experiments, consider our 30-day direct sales test and this visual, conversational shopping playbook.

Vendor risk checklist before you hand over the keys

• Data governance. Confirm how storefront, knowledge base, and conversation data are ingested, stored, and used for model improvement. Require clear retention and deletion paths.
• Access controls. Apply least privilege for who can publish, who can edit guardrails, and who can view conversation transcripts.
• Safety and quality. Run a red team exercise with tricky queries about policies, returns, or regulated claims. Document responses and time to human intervention.

How teams are organizing the work

High-performing teams reduce tool switching and make approvals visible. Many Upcite.ai customers capture the pilot brief, guardrails, and weekly agent output in one workspace, then publish an internal weekly recap that shows time to launch, revenue per send, and unsubscribe trends alongside the copy that went live.

Final word and next steps

Klaviyo’s agents bring autonomous assembly and conversational commerce into the system that already powers email and SMS. That is a meaningful shift for lean growth teams juggling calendars, creative, and service queues. You still need human judgment. You no longer need human assembly for every task.

Actionable next steps:

1. Write the 14-day pilot brief with goals, guardrails, roles, and the two campaign moments you will test.
2. Connect your catalog and help center, and add five on-brand copy examples so the agent can learn tone and structure.
3. Ship the control send by Day 3, then approve and launch the agent-assembled send by Day 7.
4. Turn on Customer Agent with a narrow scope by Day 11 and tag common questions to feed back into campaigns.
5. Decide on Day 14: scale to more sends and channels, or tune and run one more short cycle.

For more structured marketing sprints, you can also reference our 14-day marketing pilot guide.