Key Takeaways
Behavioural segmentation crossed with value beats demographics. What someone buys and how recently predicts far more than their age bracket.
A segment is only real if you would treat it differently. Otherwise it is a label that adds reporting overhead and nothing else.
Keep it to five to eight segments. Twenty means nobody owns any and campaigns default to the broadest group.
Refresh membership continuously. A member who lapsed in week two should not sit in the active segment until the quarter ends.
High-value members need recognition and service, not discounts. Discounting them erodes margin on people who would pay full price.
A single-tier loyalty programme makes the same offer to the member who shops weekly and the member who joined eighteen months ago and never came back. One of them finds it irrelevant. The other finds it insufficient. Both cost you the same to send.
Segmentation is the fix, and it is less about demographics than most programmes assume. What someone buys, how often, how recently and what they redeem tells you far more than their age bracket ever will.
This guide covers the segmentation models worth using, how to build segments that stay useful, and the mistakes that produce segments nobody acts on. For the metrics underneath, see our guide to customer loyalty analytics.
Why segmentation decides whether a programme works
- Relevance drives engagement: A reward that matches what the member actually buys gets redeemed. One that does not becomes evidence the programme does not know them.
- Budget goes where it moves something: Blanket offers spend most of their value on members who were going to buy anyway. Segmentation concentrates spend on the members whose behaviour can still change.
- Different segments churn for different reasons: High-value members leave over service; occasional buyers leave over price. One retention offer cannot address both.
- It makes the programme legible: Segment-level reporting reveals the declining cohort that an average conceals.
- It is the prerequisite for personalisation: You cannot personalise what you have not first grouped.
5
Segmentation models
behavioural is the one that pays
3
Questions a segment must answer
who, why they differ, what changes
1
Test of a good segment
you would treat it differently
Five ways to segment, and which actually work
| Model | Splits members by | Worth using when |
|---|---|---|
| Behavioural | What they do: purchase, frequency, recency, redemption, channel | Almost always. The strongest predictor of future behaviour is past behaviour |
| Value-based | Spend, margin or predicted lifetime value | Deciding where retention budget goes, and for tiering |
| Lifecycle stage | New, active, lapsing, dormant, reactivated | Any programme large enough that one message cannot fit everyone |
| Needs and preference | Category affinity, stated preferences, redemption choices | Large catalogues, where relevance is the whole problem |
| Demographic | Age, location, household | Sparingly. Useful for compliance and logistics, weak for predicting behaviour |
The practical answer for most programmes is behavioural crossed with value: what they do, weighted by what they are worth. Demographics belong in the mix for operational reasons, not predictive ones. Our guide to tiered loyalty programs covers the value axis in more depth.
Group members by what they actually do
Behavioural and value segments that refresh themselves, with segment-specific earn and redeem rules.
Book a demoTrusted by 5,000+ companies
The test for a useful segment
Before building a segment, check it against four questions. If any answer is no, it is a label rather than a segment.
- Would you treat this group differently? If the offer, message and timing would be identical to another group, the split adds reporting overhead and nothing else.
- Is it big enough to matter? A segment of two hundred members in a base of a million is a case study, not a strategy.
- Can you identify members reliably? A segment you cannot query is a segment you cannot action.
- Will it still be true next month? Segments built on a one-off campaign response decay fast. Behaviour-based segments refresh themselves.
Building segments that stay useful
- Start from the decision, not the data: Decide what you would do differently for each group first. Segments built because the data allowed it rarely get used.
- Use recency, frequency and value as the spine: These three cover most of what matters and are available in any programme with transaction history.
- Layer in redemption behaviour: What a member chooses to redeem is a stated preference they volunteered, which makes it unusually reliable.
- Keep the number small: Five to eight segments is workable. Twenty means nobody owns any of them and campaigns default to the broadest.
- Refresh continuously, not quarterly: A member who lapsed in week two of the quarter should not stay in the active segment until week twelve.
- Hold a control group: Without one you cannot tell whether the segment-specific offer beat sending nothing.
What to send each segment
| Segment | What they need | What usually fails |
|---|---|---|
| New, first 90 days | Early value and a reason to return, fast | Waiting for them to qualify for the standard reward |
| Active high-value | Recognition, access, service that matches their spend | Discounts, which erode margin on members who would pay full price |
| Active low-frequency | A reason to come back sooner, basket-building offers | Premium rewards they cannot realistically reach |
| Lapsing | A timely, specific nudge while the habit still exists | A generic win-back sent months after the drop-off |
| Dormant | A genuine reason to return, or removal from active reporting | Repeated campaigns that keep them in the headline numbers |
| Advocates | Referral mechanics and recognition rather than discounts | Treating them as a sales channel with no reciprocity |
How different industries segment
The axes change by category, and copying another sector's model is a common mistake.
- Airlines: Frequency and route value. The distinction between a frequent business flyer and an annual leisure traveller drives almost everything, including whether status or price matters more.
- Retail: Spend and basket composition. High-spenders and occasional shoppers need genuinely different programmes, not the same one with different thresholds.
- Hospitality: Stay frequency and length, plus whether the booking is business or leisure, which changes what a 'reward' even means.
- Food and beverage: Visit frequency above all, since the category lives on habit rather than value per transaction.
- Ecommerce: Recency and category affinity. Browsers and buyers require different mechanics, and treating a frequent browser as disengaged misses the opportunity.
Where segmentation goes wrong
- Segments that never change: Built once, never refreshed, slowly describing a base that no longer exists.
- Too many to act on: Fifteen segments means campaigns revert to the biggest one and the rest become reporting furniture.
- Demographics standing in for behaviour: Age and location are easy to collect and weak at predicting what someone will buy.
- Segmenting without a control: If every segment gets a tailored offer and none gets nothing, you cannot prove the tailoring worked.
- Ignoring the middle: Programmes obsess over the top tier and the churn risks, and neglect the large group in between where most of the growth potential sits.
- Personalising beyond consent: Behavioural segmentation that drives automated offers needs its own lawful basis. Our guide to loyalty program regulations covers the requirement.
Segmentation on Loyalife
01 · Behavioural
Segments from what members do
Recency, frequency, value, category affinity and redemption behaviour, combined rather than picked one at a time.
02 · Dynamic
Membership that updates itself
Members move between segments as their behaviour changes, so a lapsing member is treated as lapsing this week, not next quarter.
03 · Rules
Segment-specific earn and redeem
Different earning rates, rewards and thresholds per segment, configured rather than run as separate programmes.
04 · Messaging
Multichannel by preference
Each segment reached on the channel it actually responds to, with the message matched to the treatment.
05 · Insight
Segment-level reporting
Retention, redemption and value by segment and cohort, which is where the declining group becomes visible.
06 · Governance
Consent-aware profiling
Separate profiling consent and audit logs, so behavioural targeting stays defensible.
You can also explore member profile and segmentation and the Loyalife platform.
The discipline is simpler than the tooling suggests: group members by what they do, keep the number of groups small enough that each has an owner, treat each group differently in a way you could defend out loud, and let membership change as behaviour changes.
Book your Loyalife demo
Frequently asked questions
Segmentation is grouping members so that each group can be treated differently: different rewards, messages, timing or earning rules. The useful models split members by what they do rather than who they are, using recency, purchase frequency, value and redemption behaviour, because past behaviour predicts future behaviour far better than demographics.
Because demographics describe who someone is, not what they will do. Two members of the same age living in the same city can have completely different purchase frequency, category preference and churn risk. Behavioural data captures the thing you are actually trying to predict. Demographics remain useful for compliance and logistics, just not for prediction.
Five to eight is workable for most programmes. Fewer and the treatment is too blunt to be relevant. More and nobody owns each segment, campaigns default to whichever group is largest, and the rest become reporting furniture. The constraint is not analytical capacity, it is how many distinct treatments your team can actually run well.
Four tests. Would you genuinely treat this group differently from the others? Is it large enough to matter in your base? Can you identify its members reliably enough to action it? And will it still be true next month? If any answer is no, you have a label rather than a segment.
Recognition, access and service that matches their spend. Discounts are usually the wrong instrument, because they erode margin on members who would have paid full price. The common failure in the other direction is sending low-frequency members premium rewards they cannot realistically reach, which signals the programme is not for them.
Continuously, not on the planning calendar. A member who lapsed in week two of a quarter should not stay in the active segment until week twelve, because by then the window to re-engage them has closed. Segments built on behaviour can refresh themselves; segments built on a one-off campaign response decay quickly and should be avoided.
Where segmentation drives automated personalised offers, yes. Under GDPR and equivalent regimes, automated profiling requires its own lawful basis, separate from the consent to join the programme, along with the ability to explain the logic and let members opt out of profiling without forfeiting rewards they have already earned.
Neglecting the middle. Programmes tend to focus intensely on the top tier and on churn risks, while the large group of moderately engaged members in between gets whatever the default campaign is. That middle group is usually where the most growth potential sits, precisely because their behaviour is still movable.
![Diversity and Inclusion in the Workplace: Why It Drives Success [Activities + Survey Tips]](/Blog-cover-light/inclusion-and-diversity-in-the-workplace.avif?dpl=dpl_8fJikYxuCmYv3RQT1yaE4Git6nzR)
![How to Run an Effective Employee Commitment Survey [+ Sample Questions]](/Blog-cover-light/employee-commitment-survey.avif?dpl=dpl_8fJikYxuCmYv3RQT1yaE4Git6nzR)
