Email marketing with AI that gives you better segments and mail that lands
Segmentation driven by what people actually did, drafting that gives a human three good options instead of a blank page, and the deliverability fundamentals that decide whether any of it reaches an inbox.
Segments built from behaviour rather than from the fields people filled in
A job title tells you who someone was when they signed up. What they read, clicked, used and ignored tells you what they want now. We build segments from events joined across web, product, CRM and email, and we let a model propose segments you had not thought of.
Lifecycle segments
New, activating, engaged, at risk, dormant, customer, advocate. Defined by rules everyone can read, recalculated nightly, and the basis for every journey.
Behavioural segments
Read three pricing pages this month. Opened nothing in 90 days. Used feature X but never Y. These are where relevance comes from, and where most programmes have nothing.
Predicted segments
Likely to convert, likely to churn, likely to upgrade, from a predictive model trained on your outcomes. Shown as a score beside the rules so a person can see why.
Drafting with review
The model drafts and a person decides
The brief states the segment, the goal, the one thing the reader should do and the facts that may be used. The model produces three drafts in your voice, with subject line options. A marketer picks one, edits it, and runs it through a claims check that flags anything not in the brief. Then it is sent, or scheduled, with a record of who approved it.
This is faster than writing from nothing and safer than sending what a model wrote. It also makes testing cheap: subject lines, openers and calls to action can be tested properly because producing variants no longer costs an afternoon.
Voice guide used as model context and by the editor
Facts limited to the brief; the claims check flags the rest
Subject line and send-time tests with sample sizes set in advance
Reply classification: interested, not now, wrong person, complaint, routed accordingly
Approval and audit trail per send
What we measure instead of opens
Clicks to a meaningful page, replies, conversions attributed in your analytics, unsubscribes and spam complaints per campaign, and inbox placement by mailbox provider. Opens are reported as a health indicator only, because privacy features in some mail clients inflate them.
Consent and compliance
Consent recorded with its source and date. A lawful basis per message type. Unsubscribe honoured promptly and respected by every system that can send. Suppression lists shared across platforms. The baseline is UK GDPR and PECR; we adapt it to the markets you send to. More in AI data privacy.
Deliverability
How a receiving server decides where your email goes
Every message is checked against three DNS records before reputation is even considered. Failing authentication means the spam folder or rejection regardless of how good the email is. Passing it is the entry ticket; reputation and engagement then decide the folder.
Record
What it does
Common mistake
Our check
SPF
Lists the servers allowed to send for your domain
Too many includes; the lookup limit is exceeded and SPF fails silently
Flatten or prune includes; one record only
DKIM
Cryptographically signs messages so receivers can verify origin and integrity
Platform signs with its own domain, not yours; keys never rotated
Sign with your domain; rotate keys on a schedule
DMARC
Tells receivers how to treat failures and sends you reports
Policy left at "none" forever; reports go to nobody
Move to quarantine then reject; route reports to a dashboard
List hygiene
Removing bounces, dormant contacts and traps
Sending to a five-year-old list on day one
Warm up, suppress dormant, verify before import
Unsubscribe
One-click unsubscribe in headers and body
Link to a login page; delays of days
List-Unsubscribe headers; processed immediately
Process
Deliverability first, then segments, then drafting
Deliverability audit
DNS records, platform configuration, list quality, complaint and bounce history, inbox placement by provider. One week.
Fix and warm
Records corrected, DMARC moved towards enforcement, dormant contacts suppressed, sending volume ramped. Two to four weeks.
Segments and data
Events joined, lifecycle and behavioural segments defined, predictive scores added where history allows. Alongside automation.
Drafting workflow
Voice guide, brief template, model drafting, claims check, approval trail. Your team trained in one session.
Test and report
Monthly: placement, engagement by segment, test results and the next three tests.
AI will draft them, several ways, from a brief and your voice guide. A person chooses, edits and approves. Claims are checked against the product and the offer. We never send unreviewed generated email to customers; the reputational cost of one wrong promise outweighs any time saved.
What are SPF, DKIM and DMARC in one sentence each?
SPF lists which servers may send mail for your domain. DKIM signs each message so the receiver can verify it was not altered and came from you. DMARC tells receivers what to do when SPF or DKIM fail and sends you reports. Major mailbox providers now expect all three from anyone sending in volume.
Why are our open rates unreliable?
Some mail clients pre-load images for privacy, which registers an open whether or not the person read the message. Opens are now a weak signal. We measure clicks, replies, conversions and complaints, and use opens only as a rough health indicator.
How do you keep us compliant?
Consent recorded with source and date, a lawful basis documented per message type, a working unsubscribe honoured promptly, sender identity and address in every message, and suppression lists that the sync respects. The detail depends on the jurisdictions you send to; UK GDPR and PECR are the baseline we start from.
Which email platform do you work with?
The one you have, as long as it supports authentication, segmentation and an API. We implement Salesforce Marketing Cloud often, and have integrated most common platforms with CRMs and warehouses. We recommend a change only when the current platform blocks the plan.
Talk to us about your AI project
Tell us what you are trying to do, which systems are involved and what is in the way. We reply within one business day with an opinion and a suggested next step.