Using Information for Customer Segmentation
Segmentation vs. targeting: a distinction worth keeping
Targeting (covered in the previous module) answers "who gets this specific campaign." Segmentation is broader and more durable: grouping the entire customer base into meaningful categories that many different campaigns, strategies, and reporting views draw from over time. A good segmentation scheme outlives any individual campaign — RFM tiers or lifecycle stages get referenced in dozens of campaign briefs across a year, recalculated on a schedule rather than built fresh each time.
Behavioral segmentation
Groups customers by what they actually do: purchase frequency, browsing patterns, feature usage, channel preference. Behavioral segments are powerful because they're grounded in observed action rather than inferred characteristics, but they require a reliable, reasonably complete event stream — a behavioral segmentation scheme is only as good as the tracking feeding it, and gaps in instrumentation silently misclassify customers.
Demographic segmentation
Groups by attributes about who the customer is: age, location, income band, company size (for B2B), industry vertical. Demographic segments are simple to build and explain but weaker predictors of actual behavior than they look — two customers in the same demographic bucket can behave completely differently. Demographic segmentation works best as a secondary layer on top of behavioral or value-based segments, not as the primary targeting axis on its own.
Value-based segmentation: RFM
RFM (Recency, Frequency, Monetary) is the classic value-based segmentation model: score every customer on how recently they purchased, how often they purchase, and how much they spend, then combine the three scores into a tier. RFM endures because it's simple to compute from transaction data alone, directly actionable (high-value-but-lapsing customers are an obvious win-back target), and doesn't require any behavioral tracking beyond the transaction record itself.
-- Basic RFM scoring: quintile each dimension, then combine
WITH rfm_base AS (
SELECT
customer_id,
CURRENT_DATE - MAX(order_date) AS recency_days,
COUNT(*) AS frequency,
SUM(order_amount) AS monetary
FROM fact_orders
GROUP BY customer_id
),
rfm_scored AS (
SELECT
customer_id,
NTILE(5) OVER (ORDER BY recency_days ASC) AS r_score, -- lower recency_days = better
NTILE(5) OVER (ORDER BY frequency DESC) AS f_score,
NTILE(5) OVER (ORDER BY monetary DESC) AS m_score
FROM rfm_base
)
SELECT customer_id,
r_score, f_score, m_score,
CASE
WHEN r_score >= 4 AND f_score >= 4 AND m_score >= 4 THEN 'champion'
WHEN r_score <= 2 AND f_score >= 4 AND m_score >= 4 THEN 'at_risk_high_value'
WHEN r_score <= 2 AND f_score <= 2 THEN 'lost'
ELSE 'standard'
END AS rfm_segment
FROM rfm_scored;Needs-based segmentation
Groups customers by the underlying problem or goal they're trying to solve, rather than their behavior or value. This is usually built from qualitative research — surveys, interviews, win/loss analysis — rather than pure transactional data, and it's the hardest segmentation type to operationalize automatically because it can't be fully derived from system-of-record data alone. Needs-based segments typically get mapped onto proxies in the data (a B2B customer segmented as "cost-conscious buyer" might be approximated by discount-sensitivity and deal-size patterns) so they can still drive automated targeting.
Segmentation methodologies compared
| Methodology | Built from | Strength | Weakness |
|---|---|---|---|
| Behavioral | Event/clickstream and transaction data | Grounded in actual observed action | Depends on tracking completeness |
| Demographic | Profile attributes (age, location, firmographics) | Simple, easy to explain | Weak standalone predictor of behavior |
| RFM (value-based) | Transaction history alone | Simple, directly actionable, no tracking needed | Blind to intent or needs, only value |
| Needs-based | Qualitative research + proxy data signals | Captures underlying motivation | Hard to automate, requires research investment |
The best schemes combine layers
Operationalizing segments into campaigns
A segmentation scheme only creates value once it's wired into recurring campaign logic — segments recalculated on a batch schedule (weekly RFM refresh is common), synced automatically to the ESP or ad platform as audiences, and referenced by name in campaign briefs so planning doesn't restart the analysis from scratch every time. The operational discipline (covered more in Optimizing Contact Strategy) is making sure segment membership changes get reflected downstream promptly enough to matter — a customer who just became a "champion" shouldn't wait a month to get treated like one.
What's next
Once segments exist, the natural next question is how to prove which treatment actually moves each segment — which is where rigorous response testing comes in.
Next: Direct Marketing Response Testing →
I build these systems professionally.
Whether it's a RAG pipeline, analytics migration, or AI workflow — let's talk.