Klaviyo shipped Composer yesterday. One plain-language prompt delivers a complete, multi-channel campaign: audience segments, copy, flow logic, timing, channel affinity routing. All of it. Autonomously generated, queued for human approval, ready to send to 193,000 brands' customers.
Co-CEO Andrew Bialecki was direct about what this means: "The execution layer in software is moving from humans to agents. What matters now is having both the agents that do the work, and the infrastructure that gives them the full picture of the customer."
Most marketing leaders will read the first sentence and stop there. The second sentence is the one that matters.
Because Composer is only as good as the customer data it runs on. And for the vast majority of those 193,000 brands, the infrastructure that gives Composer "the full picture" does not yet exist.
This Is a Different Kind of AI
The context matters here. Klaviyo Composer is not a smarter chatbot or an AI-assisted drafting tool. It represents a step up in the autonomy hierarchy that changes what failure looks like.
Two weeks ago, ActiveCampaign announced agent-to-user AI: a system that monitors your campaigns, diagnoses problems, and surfaces recommendations for what to do. The AI informs. You decide. You execute.
Composer operates at the next tier. It does not surface recommendations. From a single prompt, it builds the audience segment, writes the copy, configures the multi-channel flow, sets the timing, selects the channel affinity routing, and queues the entire campaign for your review. The AI executes. You authorize.
This is the distinction Bialecki is marking: at Level 3, the AI recommends and humans execute. At Level 4, the AI executes and humans authorize. That shift changes the blast radius of every data gap in your stack.
When an AI recommends a campaign based on incomplete data, a human reviews it and can catch the error. When an AI builds and queues a campaign based on incomplete data, that campaign arrives for approval looking complete and coherent. The error is inside the logic, not visible in the output.
What Composer Actually Builds
To understand why data completeness is so critical, it helps to understand exactly what Composer generates.
Take a scenario: you prompt Composer with "Build a spring re-activation campaign for lapsed customers across email and text." In a traditional setup, this requires a human to define what "lapsed" means for your brand, build the segment, write the copy for each channel, configure the multi-step flow, set the time delays, and schedule the send.
Composer interprets the intent and autonomously executes all of it. Using Klaviyo's data engine (trained on billions of interactions across 14 years of platform history), it identifies the optimal "lapsed" criteria for your specific brand based on historical purchase frequency, average order value, and engagement decay rates. It writes channel-specific copy optimized for email subject lines, SMS, RCS, and WhatsApp. It configures conditional splits based on predicted customer lifetime value. It determines whether a specific user responds better to SMS or email and routes accordingly.
Nothing goes live without your approval. That governance layer is hard-coded. But the logic, the segments, the copy, the flow structure: all of it was built by the AI based on the data it could see.
Which brings us to the problem.
The Three Failure Modes of Autonomous Execution
When Composer operates on incomplete data, the failures are not random. They follow three predictable patterns, each one costly in a different way.
Failure Mode 1: Tone-Deaf Engagement
A customer files a Zendesk support ticket at 2pm about a damaged product they received. They are frustrated. Their ticket status: Open. Their CSAT signal: Negative.
In a standard Klaviyo deployment, Zendesk data is not connected to the Klaviyo Data Platform. The marketing system views this customer as a "recent purchaser who is eligible for cross-sell campaigns."
Composer builds a "How did we do? Leave a review" campaign targeting recent purchasers. It goes out at 9am the next morning, directly to the customer with the open damage complaint.
The AI executed correctly given the data it had. The data was incomplete.
Failure Mode 2: Margin Cannibalization
A B2B prospect is in active contract negotiations with your account executive. Deal stage in Salesforce: Negotiation. Contract value: enterprise tier.
In a standard Klaviyo deployment, Salesforce deal stages are not synced to the marketing platform. Klaviyo views this prospect as a "lead who has not yet converted to a paid plan."
Composer identifies the segment, generates a "Start today: 20% off your first year" campaign. It queues for approval, looking perfectly reasonable for an unconverted lead. The human reviewer approves it. The discount code lands in the prospect's inbox while the sales rep is negotiating full price.
The AI executed correctly. The sales context was invisible to it.
Failure Mode 3: Logistical Hallucination
A product variant sells out at the warehouse at 3pm. The inventory system records the stockout. But SKU-level inventory data is not synced in real time to Klaviyo.
Composer is prompted to build a spring clearance campaign. It includes the variant based on the last catalog data it received. The campaign goes out at 6pm to a high-intent segment. The promoted product is a dead end.
The AI executed against the data it had. The data was three hours stale.
These three failures map directly to three missing data layers. And those missing layers are not exceptions, they are the standard configuration for the majority of Klaviyo's 193,000 customers.
The Scale of the Gap
The 2026 Salesforce State of Marketing report (4,450 marketers, 26 countries) documented this precisely. Despite 75% of marketers adopting AI, 84% still run generic campaigns. The stated reason from the data: only 58% of marketing teams have complete access to service data, 56% to sales data, and 51% to commerce data.
At Klaviyo specifically, 92.9% of the 193,000 paying customers are small and medium-sized businesses. Most of these deployments rely on out-of-the-box native integrations that capture the basics: orders, email opens, cart events. The service, sales, and deep commerce data layers require custom engineering connections that the native integrations do not provide.
This is not Klaviyo's failure. Composer is a genuine advance in autonomous marketing capability. But it is built on the assumption that the data feeding it is complete. For most brands, it is not.
As IDC Research Director Gerry Murray observed about the Composer launch: "We're seeing a shift from AI assisting discrete tasks to taking on more complete workflows across marketing and customer engagement." That shift makes the underlying infrastructure more important, not less.
What "The Full Picture" Actually Requires
Building genuine execution-ready infrastructure for Klaviyo Composer means connecting five specific data layers. Each requires custom engineering. None are provided by standard native integrations.
1. Server-Side Behavioral Tracking
Standard client-side JavaScript pixels are limited by ad blockers and Apple's Intelligent Tracking Prevention, which caps cookie persistence at 24-hour windows. Server-side tracking captures behavioral events continuously, providing up to a year of uninterrupted intent data. When a user authenticates, anonymous browsing history is backfilled to their unified profile. Composer's segmentation logic becomes dramatically more accurate because the behavioral signal is complete rather than intermittent.
2. Cross-Channel Identity Resolution
A single customer may use different email addresses for mobile browsing, work purchases, and subscriptions. Without identity resolution, they appear as three separate records, each receiving independent campaigns. Klaviyo's Q1 2026 update handles up to five email addresses per profile, automatically identifying the optimal primary address for sends while preserving consent status and engagement history across all inboxes. But that resolution only works when incoming data is properly normalized and identifier fields are mapped correctly.
3. Service Interaction History
This is the highest-risk gap in most deployments. Zendesk or Gorgias ticket status, CSAT scores, and inquiry topics must be mapped in real time to Klaviyo profile properties via webhook integrations. Without this layer, Composer has no mechanism to suppress promotional messages to customers with open, unresolved support issues. The AI cannot apply the contextual judgment it was never given data to exercise.
4. Sales Deal Context
For B2B SaaS and high-ticket direct-to-consumer brands, Salesforce deal stages, pipeline values, and account owner assignments must sync bidirectionally with Klaviyo in real time. When a human sales representative is in active negotiation with an account, the autonomous marketing system must treat that account accordingly. StackSync and direct API connections make this real-time pipeline sync possible. Without it, marketing automation and human-led sales cycles operate in permanent conflict.
5. Deep Commerce and Custom Objects
SKU-level inventory status, subscription billing intervals, loyalty tier balances, and offline point-of-sale data. Klaviyo's Custom Objects API (revision 2026-01-15) handles complex data structures through a dedicated ingestion endpoint supporting 75 requests per second for real-time single-record ingestion, or up to 30,000 records per minute for batch migrations. This is the layer that transforms a product catalog from a static list into an agent-readable, contextually complete view of your commercial reality.
The Five-Stage Build Sequence
At DozalDevs, we execute this infrastructure build in five sequential stages. Sequence matters because each stage is a prerequisite for the one that follows.
Stage 1: Foundational Tracking and Identity Resolution
Migrate from client-side to server-side tracking. Configure advanced identity resolution to handle multi-email and multi-device profiles. Establish anonymous activity backfilling so browsing history maps to unified profiles upon authentication. This is the foundation that every downstream data layer depends on. Without it, the entire infrastructure is built on incomplete behavioral context.
Stage 2: Custom Object Commerce Integration
Move beyond default Shopify or BigCommerce webhooks. Map subscription intervals, localized inventory levels by fulfillment center, loyalty tier structures, and multi-currency transaction histories into Custom Objects schemas using the ingestion API. This gives Composer real-time commercial context for product and timing decisions: it knows what is in stock, who is subscribed, and what each customer's value tier is before it builds a single campaign.
Stage 3: Service Context Synchronization
Integrate the primary helpdesk (Zendesk or Gorgias) bidirectionally with the Klaviyo Data Platform. Map ticket status, CSAT scores, and inquiry categories to Klaviyo profile properties via webhook triggers. When a support ticket opens, Klaviyo receives that signal in real time and can use it as a programmatic suppression filter. Composer cannot send "Review your purchase" campaigns to customers with open damage complaints because the data tells it not to.
Stage 4: Sales and Pipeline Integration
For B2B SaaS and high-ticket direct-to-consumer scenarios: connect Salesforce via StackSync or custom API to sync deal stages, lead qualification status, and account owner assignments in real time. Composer now respects active human sales cycles. Marketing automation operates parallel to direct sales, not in competition with it. High-value accounts in late-stage negotiations are suppressed from generic lead-nurture sequences.
Stage 5: Governance and Guardrail Configuration
Before activating autonomous execution, the control architecture must be in place. This involves two components.
For the Customer Agent: configure Agent Guidance with a precise Brand Summary (500-character limit) that calibrates tone and context. Define up to 20 specific escalation rules, in natural language, dictating exactly when the AI must perform a human handoff. These rules are evaluated before each response is generated: if a customer expresses frustration twice, mentions a credit card dispute, or asks to speak to a manager, the AI stops, notifies the customer, and passes the full conversation history to your support team.
For Composer: establish role-based approval workflows that ensure human review before any AI-generated campaign goes live. The technology enforces that nothing activates without your authorization. Your team's job is to actually validate the logic, not rubber-stamp the output.
The Opportunity Is Measurable
The difference between brands operating with complete data infrastructure and those operating without it is not abstract. The Salesforce State of Marketing report documents it precisely: marketers with unified customer data are 42% more likely to respond to customers in real time and 2.2x more likely to be among the top performers in their category.
The 13% of marketers who have reached the agentic AI tier show 19% higher conversion rates, 19% lower costs, and 20% better ROI than those using standard generative AI tools.
The path from the 87% to the 13% runs through infrastructure, not through additional tools.
Composer is ready. It is a genuine advancement in autonomous marketing capability. Brands that build the underlying data architecture will access something most of the market cannot: campaigns that are personalized because the AI actually understands each customer's complete context at the moment of execution.
The question is not whether to activate autonomous execution. The question is whether the infrastructure is ready to make it work.
If you are a Marketing Operations Director or VP of Marketing evaluating what it would take to build genuine autonomous execution capability on your current Klaviyo deployment, the 20-point Autonomous Execution Readiness Checklist in the full research below maps exactly where your current deployment sits against the five-layer architecture. Use it to identify the specific gaps before you activate Composer.
