Lifecycle Messaging

The problem

Lifecycle flows are set-and-forget — and generic

Most stores run the same three templated emails for every customer. They ignore what someone actually browsed or bought, so they under-convert and slowly train customers to tune them out.

This solution writes each message to the individual — their products, their timing, their objections — while keeping brand voice and compliance intact, with a human approving the flow.

Primary objective

Recover more revenue with per-customer messaging that stays on-brand and on-policy.

How it works

Trigger → personalise → approve flow → send via ESP

Trigger

Cart, purchase, inactivity

→

Personalise

Per-customer copy & offer

→

Approve

Sign off the flow & rules

→

Send

Schedule through your ESP

What it does

✓ Abandoned-cart recovery
✓ Post-purchase & replenishment
✓ Win-back & re-engagement
✓ Per-customer personalisation
✓ Subject-line variants
✓ Send-time optimisation

Built with

✓ Claude for copy & subjects
✓ ESP (Klaviyo/Braze) API
✓ Behavioural event stream
✓ Offer / discount guardrails
✓ Approval queue per flow

Guardrails

✓ Discount caps & rules
✓ Frequency / fatigue limits
✓ Brand & legal tone checks
✓ Unsubscribe compliance
✓ Human approves each flow

Under the hood

Behaviour-grounded message personalisation

Each message is written from the customer's real events and constrained by offer rules — so it's relevant without ever over-discounting.

lifecycle/personalise.py PYTHON

from anthropic import Anthropic
client = Anthropic()

def cart_recovery(customer: dict, max_discount: int = 10):
    prompt = f"""Write an abandoned-cart email.
Items left: {customer['cart']}
Past purchases: {customer['history']}
You MAY offer up to {max_discount}% — only if it helps. Brand voice: friendly, concise.
Return JSON: {{"subject": str, "body": str, "discount": int}}"""
    msg = client.messages.create(model="claude-sonnet-4-20250514",
        max_tokens=500, temperature=0.6,
        messages=[{"role": "user", "content": prompt}])
    out = parse_json(msg.content[0].text)
    out["discount"] = min(out["discount"], max_discount)  # never exceed policy
    return queue_for_review(customer, out)

Outcomes

Recovered

Abandoned revenue

Per-customer

Not templated

Higher

Open & click rates

On-policy

Discounts capped

Make every lifecycle message feel 1:1

Feeds on signals from Review Insights and works with the Content Generator.

Start a conversation →