Measures and attributes in Customer Insights — Data

A unified profile is a snapshot of who someone is. Measures and computed attributes in Customer Insights — Data layer derived signals on top — RFM scores, lifetime value, days since last purchase, engagement scores. These derived signals are what marketing, sales, and service actually use to drive decisions.

The distinction.

• Measure — a single aggregate value across all customers (or a segment): "Total revenue last quarter."
• Computed attribute — a per-profile value: "This customer's total purchases."

Both layer derivation on the underlying data; different scope.

Common measure examples.

• Total customers — count.
• Average lifetime value.
• Total revenue — sum.
• Active customer rate — percentage active in last 30 days.
• Churn rate — percentage who unsubscribed.

Measures appear on dashboards and reports; aggregate views of the customer base.

Common attribute examples.

• Total purchases — per customer.
• Days since last purchase — recency.
• Purchase frequency — over past year.
• Average order value.
• Lifetime value (LTV) — total spend.
• Engagement score — composite metric.
• Churn risk — ML-predicted.
• Preferred channel — most-used channel.

Attributes available on each profile; usable in segments, journeys, personalization.

Defining a measure. In the maker portal:

1. Add new measure.
2. Define the calculation logic (sum, count, average, formula).
3. Optionally segment-scope the measure.
4. Save.

The measure computes; result visible on dashboards.

Defining a computed attribute.

1. Add new attribute.
2. Choose the source table.
3. Define per-profile calculation.
4. Save.

The attribute computes per profile; available everywhere profiles are used.

Aggregations.

• Sum, Count, Average, Min, Max — standard.
• Distinct count — unique items.
• Conditional aggregations — sum where condition.
• Custom formulas — Power Fx-like expressions.

The formula language enables business-meaningful calculations.

Dependencies.

• Measures depend on tables.
• Attributes depend on tables.
• Computed-attribute-on-computed-attribute possible.
• Changes ripple through dependency graph.

Understanding dependencies prevents breaking changes.

Refresh cadence.

• Continuous — incremental as data changes.
• Scheduled — periodic.
• On-demand.

For real-time use cases (journey triggering), continuous matters; for periodic reporting, scheduled fine.

Predictive attributes. ML-derived:

• Customer lifetime value prediction.
• Churn risk.
• Product affinity.
• Next likely action.

Built on Microsoft's models; data quality and history determine accuracy.

Custom ML models. Beyond built-ins:

• Train your own model in Azure ML.
• Score profiles.
• Score becomes an attribute.

This is the integration path for unique business prediction needs.

Using attributes in segments. Computed attributes are first-class segment criteria:

• "LTV > $1000" — high-value segment.
• "Days since last purchase > 90" — at-risk segment.
• "Churn risk > 0.7" — proactive intervention.

The richer the attribute set, the more segmentation possibilities.

Using attributes in journeys.

• Trigger journey when attribute crosses threshold.
• Branch journey paths based on attribute.
• Personalise content with attribute.

Attributes drive both segmentation and personalisation.

RFM analysis. Classic customer scoring:

• Recency — days since last purchase.
• Frequency — purchase count in period.
• Monetary — total spend.

Each scored 1-5; combined into RFM score. Each component is a computed attribute; combined score derived from them. CI-Data supports natively.

Cohort analysis.

• Group customers by attribute (signup month, first purchase channel).
• Track cohorts over time.
• Compare cohort behaviours.

Attributes enable cohort definitions.

Performance considerations.

• Complex attribute calculations slow refresh.
• Cross-table aggregations expensive.
• Dependency depth multiplies cost.

For high-volume Customer Insights deployments, performance tuning matters.

Common pitfalls.

• Attribute proliferation. Hundreds of attributes; nobody knows what's used.
• Stale attributes. Calculated once, never refreshed; misleading.
• Circular dependencies. Attribute A depends on B depends on A; refresh fails.
• Source data wrong. Garbage in, garbage out.
• No documentation. Future maintainers don't know what each attribute means.

Best practices.

• Name attributes clearly — LifetimeValue, DaysSinceLastPurchase.
• Document the calculation — what data, what formula.
• Owner per attribute — who maintains.
• Periodic audit — retire unused.
• Test changes — recalc subset before full refresh.

Strategic positioning. Measures and attributes are how customer data becomes operational. Without them, the unified profile is descriptive; with them, it's predictive and actionable. Mature CDP deployments have rich attribute libraries that consuming teams use confidently. The investment is in:

• Defining attributes thoughtfully.
• Maintaining data quality upstream.
• Refresh discipline.
• Documentation and governance.

The teams that get this right have a competitive customer data foundation; the teams that don't have an expensive CDP that no one uses.