Klaviyo’s Autonomous CRM: What Composer and Headless Mean for a Two-Person Team

Klaviyo calls itself an autonomous B2C CRM now. At its September event it did three things. It made 260+ of its tools callable from Claude, ChatGPT, or any agent. It put SQL behind a plain-language prompt in its data platform, in preview. And it showed Composer, the agent that builds a campaign from a brief and brings finished work for approval instead of waiting for a prompt. Klaviyo counts more than 205,000 paying customers, so these features reach a lot of small teams.

For a two-person CRM team on Klaviyo, the question isn’t whether this is real. It is what changes on Monday, what doesn’t, and what to do first.

What changes

Building is no longer the bottleneck. Composer takes a brief, proposes an audience and a message, and builds the flow or campaign with a person approving before launch. The hours a small team spent assembling a welcome series or a win-back go down. That is the headline, and it is true.

The data platform answers questions. SQL in Klaviyo Data Platform, behind a prompt, means “customers who bought twice in June and haven’t opened since” is a question, not a segment-building session. For a team with no analyst, this is the larger change.

Other tools can drive Klaviyo. Headless exposes Klaviyo to outside agents through the Model Context Protocol. A support tool, a custom assistant, or a data team’s agent can read profiles and trigger actions. The account is no longer only what the two people click.

The Customer Agent takes service load. Order tracking, returns, subscription management, in more than 100 languages. Klaviyo cites a customer whose agent resolves 86% of queries without a person. A team that was answering “where is my order” by hand gets those hours back.

What doesn’t change

Every one of those features reads the same three things: the metric list, the profile, and the consent per channel. Composer builds a segment on whichever purchase metric it finds; if there are two, it picks one. The data platform answers on the schema as named. Headless gives an outside agent the same view, faster.

So the account with Placed Order, placed_order, and Order Placed (Shopify) side by side gets three versions of the truth, now generated in seconds. The account where a customer has two profiles gets two welcome series from an agent that never gets tired. The account with email consent doing duty for SMS gets suppressed sends, or worse.

The features raised the price of a messy account. They didn’t clean it. That is the same finding as AI readiness on any platform, with Klaviyo’s names on it.

Three things to do before switching any of it on

Fig 1 · Three fixes first · Metrics, profiles, and consent before the agents

One: the metric audit. List every metric with its 30-day volume and which flows or segments read it. Merge duplicates. Retire the unread. This is an afternoon on most accounts and it is the difference between an agent that helps and one that confidently builds on the wrong event.

Two: one profile per person. Count profiles sharing an email or phone, and merge them, so the agent doesn’t treat one customer as two people.

Three: consent per channel, and Smart Sending on. Email consent, SMS consent, push consent as separate states with dates. Smart Sending switched on for anything promotional, so the agent-built flow and the Thursday campaign don’t land on the same afternoon. A Composer-built campaign respects the account’s settings; it doesn’t invent them.

Then, in order

Ask the data platform five questions you know the answers to, and compare. Let Composer rebuild a campaign you already run, and compare the audience count and the steps with yours; the difference is the rules you forgot to write down. Only then let it build something new, with a holdout, to a small audience. Headless comes last, once you know which outside tools should be allowed to act on the account and with what consent.

Sort what you hand over by the cost of being wrong.

The honest version for a small team

A two-person team on a clean Klaviyo account gets more from these features than a ten-person team on a messy one, because the features replace hours the small team didn’t have. A two-person team on a messy account gets the same messy campaigns, faster, and a service agent that can’t find the customer.

The three-item list above is a week. The features are switched on in an afternoon after it. That order is the whole advice.

The takeaway

Klaviyo’s agents build, answer, and serve on top of the metric list, the profile, and the consent state. Clean those three first, then switch on the data platform, then Composer on a campaign you already know, then the rest. The autonomy is real. It is autonomy over whatever it finds.

Read next

• AI readiness: what your customer data needs first. The same three checks, platform-neutral.

• Identity resolution: deterministic, probabilistic, and where each breaks. Item two, one profile per person.

• Event taxonomy pruning: fewer events, better data. The method behind the metric audit.