Personalisation is a means, not a goal
Few marketing ideas are as universally endorsed and as poorly executed as personalisation. Teams invest in data platforms and dynamic content, then use them to put a first name in a subject line and swap a hero image by gender. The customer receives the same message with cosmetic changes, and nobody can show it made a difference.
Our view is that personalisation should be judged by one question: does it change what the customer receives in a way they would value? A replenishment reminder timed to when they are likely to run out passes. Their first name on a generic sale email does not.
The levels of personalisation
Fig. 01 · Hierarchy
Tap to explore
From cosmetic to genuinely personal
01 · Individual prediction
Next best product, offer or message chosen per person by a model
02 · Behavioural
Content driven by what the person browsed, bought or used
03 · Lifecycle
Messages matched to stage: new, repeat, loyal, at risk
04 · Segment
Different versions for groups such as region or category interest
05 · Cosmetic
Name, city or other merge fields on a shared message
There is no need to reach the top. For many businesses the biggest gains come from the middle levels: matching messages to lifecycle stage and to recent behaviour. These are understandable, testable and achievable with ordinary email and CRM tools.
What scale actually requires
Personalising for ten customers is a matter of attention. Personalising for a hundred thousand is a matter of systems. Three things have to be in place, and each is usually harder than the technology.
Compare scenarios
The three requirements
Reliable, connected information about each person, available when the message is assembled.
- Identity resolved across devices and channels where lawful
- Behaviour and purchase events flowing into the email or CRM platform
- Declared preferences captured and kept current
- Clean records; see CRM data hygiene
The logic that chooses what each person receives.
- Rules: if browsed category X recently, show X
- Models: predicted next purchase or churn risk
- Priorities when several rules apply
- Fallbacks when data is missing
Enough modular content to fill the variations the logic calls for.
- Content blocks by category, stage and offer
- Product feeds for dynamic recommendations
- Copy written to work in many combinations
- A process to keep blocks current
Content is the most underestimated requirement. A rule that shows different content to eight interest groups needs eight sets of content, kept up to date. Teams that plan the logic without planning the content production end up with personalisation that falls back to the default most of the time.
Where to personalise first
Fig. 02 · Matrix
Tap to explore
Prioritising personalisation opportunities
- Timing: send replenishment, renewal and reminder messages when they are relevant to the individual rather than on a fixed date.
- Recently viewed and related items: show products from the categories a person actually browsed.
- Lifecycle-aware content: welcome a new customer differently from a loyal one in the same campaign.
- Suppression: do not advertise products a customer has just bought or has a complaint about. This is personalisation by omission and often the most valued.
- Language and location: where customers have chosen a language or live somewhere with different delivery options, festivals or payment methods.
Rules or models?
Rules are transparent: anyone can read 'if a customer bought running shoes, recommend running socks' and understand it. They are limited by the imagination of the person writing them. Models, such as recommendation engines and propensity scores, can find patterns humans miss but need volume, clean history and monitoring, and their choices can be hard to explain.
A practical approach is to start with rules for the high-value moments, then add models where the volume of choices is too large for rules, such as product recommendations from a big catalogue. Keep a human-readable fallback for every model. Our guides to predictive analytics and AI in the CRM go further.
The creepiness line
There is a line between helpful and unsettling, and it is not always where marketers expect. Customers generally welcome personalisation based on what they have done with you directly and knowingly: their purchases, their stated preferences, items they saved. They are often uneasy when a message reveals inferences they did not expect, especially about sensitive matters such as health, finances or family circumstances.
Myth vs reality
Personalisation myths
Building the content system
Personalisation at scale is, in the end, a content operation. The logic can pick from options only if the options exist, are current and work in combination. Teams that succeed treat content as a library of modules rather than a series of finished emails.
- 01Define the dimensions you will personalise on. Two or three, such as lifecycle stage and category interest, are plenty to start.
- 02Create a module for each value of each dimension. A hero block per category, an intro paragraph per stage, a call to action per relationship.
- 03Write modules to stand alone. Each must read well next to any other module, without assuming what came before.
- 04Always write the default. Every slot needs a version for people whose data is missing.
- 05Tag and date every module. So stale content can be found and refreshed.
- 06Review combinations, not just modules. Preview real recipient profiles to catch awkward pairings.
Product feeds deserve special care. Dynamic recommendations draw on catalogue data, so missing images, wrong prices or out-of-stock items appear directly in customers' inboxes. Good product feed hygiene is a personalisation requirement, not just an advertising one.
Proving that it pays
Personalisation adds complexity, so it should prove its value. For each personalised element, keep a control group that receives the default version and compare outcomes such as clicks, conversions and revenue per recipient. Measure over several sends; one result is noise. Retire elements that do not beat the default, and redirect the effort to those that do. The email KPIs framework covers the measures.
Personalisation builds on email segmentation and depends on first-party data gathered with clear consent. Done well, it makes a large programme feel like a series of individual conversations.
Personalisation is not knowing a lot about the customer; it is using a little of what they told you, well.
Key takeaways
- 01Judge personalisation by whether it changes what the customer receives in a way they would value.
- 02The middle levels, lifecycle and behaviour-based content, usually deliver most of the value.
- 03Scale needs data, decisioning and modular content; content is the most underestimated of the three.
- 04Start with rules for high-value moments and add models where choices outgrow rules.
- 05Stay on the right side of the creepiness line and test every personalised element against a control.
Frequently asked
- What is personalisation at scale?
- It is the practice of tailoring marketing messages, offers and timing to individual customers' behaviour, needs and context across a large audience, using data, rules or models and modular content rather than manual work. The goal is consistent relevance for many people without proportionally more effort.
- What are examples of email personalisation?
- Useful examples include replenishment reminders timed to an individual's purchase cycle, content blocks matched to categories they browsed, lifecycle-specific messages within one campaign, product recommendations, language or location-specific details, and suppressing promotions for items they have just bought or have a complaint about.
- What data do I need for personalisation?
- Start with reliable basics: purchases, recent browsing or product use, lifecycle stage, stated preferences and consented contact details. These first-party signals support most valuable personalisation. Add more only when you have a specific use that the customer would value and the consent to support it.
- How do I avoid personalisation feeling creepy?
- Personalise visibly on information customers knowingly gave you, such as purchases and preferences, and be cautious with inferences, especially about sensitive matters like health or finances. Explain how you use data, give customers control and follow data protection laws in each market.
- How do I measure whether personalisation works?
- Keep a control group that receives the default version of each personalised element and compare clicks, conversions and revenue per recipient over several sends. Retire variants that do not outperform the default, since each one carries ongoing maintenance cost.
Published by Fabulous.Media, a network of specialist marketing agencies. Updated 9 October 2026. Platform features change often; check current official documentation before acting on platform-specific detail.
