Sequence architecture
- Define entry trigger and suppression rules (customers, competitors, unsubscribes).
- Map each email to a single CTA aligned with lifecycle stage.
- Set branch paths for opens, clicks, and negative signals.
- Exit to sales or a longer nurture when scoring threshold hits.
Deliverability basics
- Warm new domains and IPs gradually.
- Authenticate SPF, DKIM, and DMARC on sending domains.
- Monitor bounce and complaint rates weekly.
Where AI fits
Draft personalized opening lines from CRM fields, generate subject-line tests, summarize call notes into follow-ups. Always keep a human approval step for net-new campaigns.
We implement these patterns for clients—see services.
Consent is an architecture decision, not a checkbox
The consent rules your flows must respect differ by regime — the US CAN-SPAM Act works on an opt-out model with identification and unsubscribe requirements, while GDPR and Canada's CASL require a lawful basis or consent before sending — and your automation should encode the strictest regime any recipient falls under. Practically, that means consent status and jurisdiction live on the contact record as first-class fields, every entry trigger checks them, and unsubscribes propagate to all connected systems within the legal window rather than living only in the email platform. Verify specifics with counsel; the architecture point stands regardless.
Where LLMs help email automation, honestly
Not subject-line roulette. The durable wins are unglamorous: summarizing long reply threads into CRM notes, classifying inbound replies (interested, unsubscribe request, out-of-office, angry) so routing stops depending on keyword rules, and drafting variant copy that a human then edits and approves. Full generative sends without review keep producing the failure everyone has seen — confidently wrong details in a customer's inbox, at scale.