Customer Insights segmentation — a deep dive

A segment is a defined audience — "customers in California who bought twice in the last 90 days" or "leads from Industry X with a job title containing 'Director'." Customer Insights is built around segmentation; everything from journey orchestration to analytics depends on well-designed segments. The technical mechanics are straightforward; the discipline of segment management — definition, governance, performance, evolution — is the harder problem.

The two Customer Insights products.

• Customer Insights — Data (formerly CDP) — the data unification layer; pulls from multiple sources, resolves identity, builds a unified customer profile.
• Customer Insights — Journeys (formerly Marketing) — orchestration engine using segments to drive journeys (emails, SMS, push, in-app).

Segments exist in both, but with different scope and capability.

Segment types.

• Static — a defined snapshot at a point in time; doesn't update.
• Dynamic — query-based; refreshes against current data on a schedule (typically hourly or daily).
• Compound — segments combining other segments (segment A AND segment B, or segment A NOT in segment B).
• AI-derived (Data) — clustering or propensity models that produce segment-shaped output.

Most production segments are dynamic — they need to reflect current customer state, not snapshots.

Segment query design.

• Filter conditions on unified customer profile attributes — demographic, behavioural, transactional.
• Time windows — "in last 90 days", "in current quarter".
• Aggregations — "customers with >3 purchases".
• Joins through related tables — "customers whose company is in industry X".

The query builder is visual but understanding the underlying data model matters — choosing the right attributes is the difference between accurate and noisy segments.

Refresh patterns.

• Continuous refresh — segments update as underlying data changes (near real-time).
• Scheduled refresh — daily or hourly batch.
• Manual refresh — on-demand.

Continuous refresh is more expensive computationally; reserve for time-sensitive segments. Most marketing segments are fine with daily refresh.

Membership counts. Segments display current membership:

• Total profiles in segment.
• Trend over time.
• Demographic breakdown.
• Activity stats.

Watching segment sizes change is a quick health check — a segment shrinking week-over-week may indicate a real trend or a data quality issue upstream.

Using segments in journeys. A segment is a journey's audience:

• Trigger the journey when a profile enters the segment.
• Loop through the journey while the profile remains.
• Exit when no longer in segment, journey completes, or explicit unsubscribe.

Segments inform entry points; behavioural triggers within the journey then personalise the flow.

AI-derived segments.

• Lookalike segments — find profiles similar to a seed.
• Churn prediction segments — profiles likely to churn.
• Lifetime value segments — predicted high-value profiles.
• Product affinity segments — likely to buy product X.

Built on Microsoft's underlying ML models or custom models trained on your data. Setup requires reasonable historical data volume (months of behaviour).

Segment governance. A common antipattern: hundreds of segments accumulate without ownership. Mitigations:

• Naming convention — segment names indicate purpose, owner, refresh.
• Tagging — by department, by use case.
• Lifecycle management — segments that haven't driven action in 90 days marked for retirement.
• Owner accountability — each segment has a designated owner.

Without governance, segment proliferation overwhelms the platform.

Performance considerations. Segment refresh time depends on:

• Data volume — larger profile sets take longer.
• Filter complexity — joins and aggregations add cost.
• Refresh cadence — more frequent = more total compute.

For very large segments (millions of profiles) refreshing hourly, costs add up. Profile segment cost vs use value.

Suppression segments. A common pattern: "everyone except these people":

• Do-not-contact segment — explicit opt-outs, complaints, bounces.
• Excluded for compliance — recipients in restricted jurisdictions.
• Already in another journey — avoid double-touch.

Suppression layered before any send. Without it, GDPR/CAN-SPAM violations are easy.

Cross-segment analytics. Beyond using segments to send messages:

• Overlap analysis — which profiles are in multiple segments.
• Conversion analysis — segments driving the highest journey conversion.
• Trend analysis — segment growth over time.

These insights drive segment refinement.

Common pitfalls.

• One giant "all customers" segment. Default for lazy campaigns; no personalisation.
• Time-window drift. "Last 90 days" was right when defined; now it should be "last 30 days"; never updated.
• Stale criteria. Customer Insights model changed but segment query still references old attribute names; segment returns empty.
• No suppression hierarchy. Customers receive multiple competing journeys; experience fragmented.
• Segment used cross-system without check. Same segment definition exists in Dynamics, in a separate analytics tool, in Excel; they drift over time; reporting inconsistent.
• Performance ignored. Segments refreshing too frequently; costs compound.

Operational discipline. Segmentation is a craft. Build segments thoughtfully, test them against expected counts, observe their behaviour over time, refine. Treat segments as living artefacts with owners and lifecycle, not as one-time creations. The teams that get value from Customer Insights are the ones that treat segments as products — designed, maintained, retired with intent.