"Hi {FirstName}" has not been personalisation for a decade. What people mean now is email that responds to behaviour: what somebody looked at, bought, ignored or abandoned.
AI has made two parts of this genuinely easier: producing many content variants, and deciding which to send to whom. It has not made the strategy easier, because the strategy is deciding which behaviours matter, and that is a business question.
Quick Info
Real personalisation
Based on behaviour, not on merge fields
Highest return
Abandonment and post-purchase triggers
Where AI helps most
Generating content variants at volume
Where it helps least
Deciding what to personalise on
Minimum data needed
Purchase or browse history, or engagement history
Rule
Personalise on things the recipient knows you know
The three levels, and where to start
Level one: segmentation
Different email to different groups. Customers versus prospects, by service interest, by engagement level. Crude, cheap, and it produces most of the available gain. Start here.
Level two: behavioural triggers
Email sent because somebody did something. Viewed a page, abandoned a basket, booked a call, did not open for ninety days. This is where the revenue is, and it is automation rather than AI.
Level three: dynamic content and prediction
The same send containing different blocks per recipient, chosen by rules or by a model. Genuinely useful at scale, and rarely the right place for a small business to begin.
Most businesses jump to level three and skip level one, then conclude personalisation does not work. Segmenting customers from prospects is worth more than any model.
The triggers worth building first
01Welcome sequence, on signup. Three emails. Highest return per hour of work in all of email.
02Abandoned basket or abandoned form, within an hour. The single highest-revenue automation in e-commerce.
03Post-purchase: delivery, usage guidance, a review request, then a complementary offer.
04Browse abandonment: viewed a service or product page repeatedly and did not act.
05Re-engagement: no opens or clicks in ninety days, one attempt, then remove.
06Anniversary or replenishment, where your product has a natural cycle.
Six automations, built once. In most accounts these produce more revenue than all broadcast sends combined, and they run while you do something else.
Where AI genuinely helps
—Generating variants. Five subject lines and three body versions per segment, in minutes rather than an afternoon.
—Adapting one email for several segments, changing the examples and emphasis rather than rewriting.
—Summarising your own content into email-length pieces.
—Drafting product descriptions and recommendation blocks at catalogue scale.
—Send-time optimisation, where the platform predicts per-recipient timing. Modest and free, so use it.
—Predictive scoring: who is likely to buy, who is likely to churn. Useful once you have enough history.
The pattern is the same as everywhere else: it is good at volume and variation, weak at judgement.
Where it does not help
—Deciding which behaviours matter to your business. That is knowing your customers.
—Writing the email people actually respond to, which is usually plain, specific and slightly personal in a way generated text is not.
—Anything based on data you do not have. No model invents purchase history.
—Small lists. Predictive features need volume; under a few thousand engaged subscribers they are guessing.
—Compensating for a bad offer, which is the most common thing personalisation gets asked to do.
Setting up a behavioural campaign
01Pick one behaviour that indicates intent. Viewed the pricing page twice in a week is a good first choice for a service business.
02Confirm your platform can see it. This usually needs site tracking connected to your email tool, which is a one-time setup.
03Write one email that acknowledges the behaviour without being creepy. "You were looking at our pricing, here is what usually comes next" is fine. "We noticed you spent 4 minutes on our pricing page" is not.
04Set the delay. An hour for abandonment, a day for browse behaviour, longer for considered purchases.
05Set exit conditions so somebody who converts stops receiving it. This is the step people forget and it is the one that annoys customers.
06Run it for a month, then use AI to generate variants of the parts that underperform.
One good trigger beats five half-configured ones. Build them individually and confirm each works before adding the next.
The creepiness line
Personalisation works when it feels like service and fails when it feels like surveillance. The line is roughly: personalise on things the recipient knows you know.
—Fine: what they bought, what they subscribed to, what they told you, what they clicked in your email.
—Uncomfortable: precise page-level browsing behaviour, time spent, repeated visits described back to them.
—Fine: "Since you bought the starter kit, this might be useful."
—Not fine: "You have visited this page three times this week."
—Always: an easy way to opt out of the specific type of email, not just of everything.
If the email would be embarrassing to explain to the recipient, do not send it, whatever the uplift.
What to measure
—Revenue or enquiries per automation, compared with what the segment did before it existed.
—Click rate on the specific personalised element, which tells you whether the personalisation is the reason.
—Unsubscribe and complaint rate per automation, which is your creepiness detector.
—Reply rate, which rises sharply on genuinely relevant email.
—Hold out a small control group where your platform allows it, so you know the automation is causing the result.
A realistic first month
01Week one: segment customers from prospects and stop sending them the same thing.
02Week two: build the three-email welcome sequence.
03Week three: build one abandonment or intent trigger.
04Week four: use AI to generate subject line and body variants for all of the above, and test them.
05Then leave it running and add one automation a month.
That sequence puts the money in before the machinery. The automation workflows piece covers the full set worth building.
Frequently Asked Questions
What is AI personalisation in email, practically?
Mostly generating content variants at volume and predicting things like send time or purchase likelihood. The triggers and segments themselves are ordinary automation.
Do I need AI to personalise email?
No. Segmentation and behavioural triggers deliver most of the available gain and require no AI at all.
How much data do I need?
For triggers, just the behaviour itself. For predictive features, a few thousand engaged subscribers and real purchase history, or the predictions are noise.
Is behavioural email creepy?
It becomes creepy when you describe browsing behaviour back to people. Personalise on what they know you know: purchases, signups, clicks.
Which automation should I build first?
The welcome sequence, then abandonment. Those two produce more revenue than everything else in most accounts.
Does send-time optimisation work?
Modestly, and it is usually free in your platform. Turn it on and do not expect it to change your results much.
Before You Go
Segment first, build the six triggers, and use AI for the variants rather than the strategy. That order puts revenue in early and machinery in later.
And keep the personalisation to things the recipient knows you know. The workflow set is what to build next.
Behaviour, not merge fields. Triggers before models.
WRITTEN BY TARIQ SALLAM
Marketing Consultant. Entrepreneur. Content Creator.
I'm a marketing consultant, entrepreneur and content creator. I help businesses grow through practical marketing, websites, SEO, content and AI.
More About Tariq →
Keep reading