I Haven't Built a Klaviyo Flow by Hand in Months. An AI Agent Drafts and Deploys Them for Me.

The before

What flow building used to cost

Every DTC operator knows the maths on flows. Automated emails generate around 41% of total email revenue from barely 5% of sends, and the revenue per recipient is roughly 18 times higher than campaigns. The work pays for itself many times over. And yet most brands run three or four flows when top accounts run twelve to sixteen, because building them properly is slow, expensive, and nobody's favourite job. You either pay an agency a retainer somewhere between £2,500 and £8,000 a month and wait four to six weeks for a welcome series to crawl through revisions, or you write the emails yourself at midnight after the actual business is done for the day.

So the flows don't get built. The browse abandonment flow stays on the someday list. The win-back flow never exists. The replenishment flow that should be quietly compounding repeat revenue sits at zero. If your flows generate less than 15% of your total Klaviyo revenue, the automation layer is underperforming, and the most common cause is simply missing coverage. Not bad strategy. Unbuilt flows.

The agent

What the flow builder actually does

The input is small: access to the Klaviyo account, the brand's context file, and the product catalogue. From there the agent runs a four-stage workflow. First, the audit. It pulls every existing flow, its status, its messages, and its performance metrics through the Klaviyo API, then maps what exists against the core stack a DTC brand should be running: welcome, abandoned cart, browse abandonment, post-purchase, replenishment, win-back, and sunset. The output of stage one is a gap list ranked by expected revenue impact, using the brand's own traffic and order data rather than generic benchmarks.

Second, the drafting. For each missing or underperforming flow it writes the complete sequence: subject lines, preview text, body copy, and the call to action for every email, plus the logic around them. Trigger conditions, time delays, conditional splits for first-time versus repeat buyers, and the smart sending exclusions that stop someone who just purchased from getting an abandoned cart nudge for the same product. Third, the build. The agent creates the flow directly in the live Klaviyo account, templates and all, set to draft status. Fourth, the handover: a short summary of what was built, why, and what it expects each flow to do, ready for a human review pass before anything goes live.

A full flow stack that would have been a six-week agency project lands in the account in a day, drafted, structured, and waiting for review. The review itself takes an hour.

Brand memory

Why it doesn't sound like AI

The difference between this and pasting a prompt into a chatbot is the context layer. The agent writes from a brand memory file that holds the voice rules, the banned phrases, the positioning, the proof points it is allowed to use, and the product economics: margins, replenishment cycles, AOV, which products lead and which attach. It knows a £24 supplement with a 45-day use-up cycle needs a replenishment email at day 35, not a generic check-in at day 14.

On top of that sits the voice of customer layer: a phrase bank mined from the brand's real reviews and support messages. When the agent writes the abandoned cart email, the objection handling comes from objections actual customers raised, in the words they used. That is the part that makes the output feel like an employee who has read every customer conversation the brand has ever had, because functionally, it has.

The output

What lands in the account

Here is what a deployed welcome series looks like for a wellness brand we work with. Five emails over eight days. Email one fires instantly: the founder story and the single strongest proof point, subject line drawn from the most repeated phrase in five-star reviews. Email two, day two: the ingredient or mechanism explained in plain language, answering the most common pre-purchase question found in support tickets. Email three, day four: social proof, three short customer quotes chosen by theme. Email four, day six: the objection email, built from the most common reason hesitant buyers gave in reviews. Email five, day eight: the first and only incentive, sized to the brand's margin rules rather than a reflex 10%.

Each email arrives in Klaviyo with its trigger logic, delay, and exclusions already configured. The benchmark for a welcome flow is around $2.35 revenue per recipient and an open rate above 80%, and the gap between an average flow and a great one is sequence design and language. Both of those are exactly what the agent is built to get right, because it works from the brand's own data on every draft.

Inside the system

How we build this for brands

The flow builder is one component of the lifecycle layer we deploy for DTC brands. It sits alongside a VOC engine that mines customer reviews and support messages into positioning and ad creative, so the language in your email flows and the language in your TikTok and Meta ads come from the same source: what customers actually say. The same context layer feeds both, which is why the brand sounds consistent everywhere a customer meets it.

Around the email work sit profit and cash-flow dashboards built from live Shopify and ad data, with a reporting agent that flags leakage or risk weekly, so we can see what each flow is contributing rather than guessing from Klaviyo's attribution alone. Part of this runs live for portfolio brands today; the full system is what we deploy when we take a brand on.

Frequently asked questions