Segment refresh cycles and membership behaviour

What it does

Real-time segments are evaluated on a schedule the platform manages for you. Since 17 February 2025, Customer Insights – Journeys varies that schedule automatically based on whether a segment is actually being used, and retires segments nobody has touched for a long time. There's nothing to configure — but you do need to know the cadence, because it sets the floor on how quickly a data change shows up in a journey.

Key facts

Automatic segment management is GA, not preview, and applies without any admin action.
A segment used in a live journey — or referenced by another segment or by an email — refreshes roughly every 30 minutes.
If that in-use segment contains at least one interaction (behavioural) filter, the cycle stretches to roughly 60 minutes.
A newly created segment that isn't used anywhere refreshes every 30 minutes for its first 24 hours, then drops to a 24-hour cycle.
A segment left unused for 120 days moves to an Expired state: it stops being evaluated, it stops counting toward segment limits, and it has to be republished before a journey can use it again.
Membership doesn't appear instantly after a segment goes live. The members list typically takes a few minutes to populate.
An exact member count and the full member list are only available once the segment is marked ready to use. Before that, the builder gives a best-effort estimate via Refresh and a partial list via View sample of included members.
Segment membership can be exported to CSV with Export to Excel on the segment's command bar, capped at 750,000 members.
The export contains five columns only — customer ID, first name, last name, email address, phone number. Custom fields, segment attributes and calculated measures are not included.
Export needs the Export Segment Members privilege plus read access to the segment; write access isn't required. It's on the Marketing Manager and Marketing Professional roles.

When to use / skip

This isn't a feature you switch on, it's a constraint you design around. The place it matters is any conversation containing the words "real time". A scheduled journey that enters people from a segment inherits the segment's refresh cadence, so the honest answer to "how fast does someone enter after they update their preference?" is tens of minutes, not seconds — and up to an hour once behavioural filters are involved.

If the business genuinely needs sub-minute reaction, the answer is a trigger-based journey keyed on the event itself, not a segment that refreshes faster. Segments are for defining populations; triggers are for reacting to things. Say that plainly the first time someone asks for a faster refresh, because there isn't a dial to turn.

Configuration decisions

Which audiences are expressed as segments and which as triggers, based on the latency the business will accept.
Whether behavioural filters are worth the extra 30 minutes of refresh latency on a given segment, or whether the same logic can sit as a condition inside the journey.
How segments are kept out of the 120-day expiry window — either by genuine use, or by a deliberate housekeeping pass that retires them properly.
Who holds the Export Segment Members privilege, given the export includes email and phone for up to three-quarters of a million people.
Whether the five-column export is enough for the client's offline analysis, or whether reporting needs to come from Dataverse or the analytics surface instead.
What counts as "ready to use" sign-off in your process, since that's the gate for an exact count.

Gotchas

Expired segments fail quietly at the point of reuse. A marketer opens a seasonal segment eleven months later, it looks fine, and it won't run until republished.
The refresh cadence applies to segment membership, not to consent. Consent is enforced at send time, so a segment count and an actual delivery count will differ, and clients will read that as a fault.
Segment counts shown during a demo are estimates unless the segment is ready to use. Quoting an estimate to a stakeholder and then delivering a different number is an avoidable argument.
Adding a single behavioural condition to an otherwise attribute-only segment halves its refresh frequency. Nobody will connect the two.
The export cap is 750,000 members and the Export to Excel button simply goes disabled above it — no message explaining why.
The relationship between a segment and the journeys using it isn't fully visible from the segment record. Historically the Related tab surfaces suppression usage rather than target usage, so don't rely on it to answer "what breaks if I change this segment".

Consultant notes

Put the refresh cadence in the solution design document in plain numbers. It prevents the "why isn't this real time" conversation from recurring at UAT.
Before go-live, review every segment older than a few months and either use it, rebuild it or delete it. Expiry is a housekeeping opportunity, not a problem.
Demo the difference between the builder estimate and the live count once, early, so the client stops treating the estimate as a number they can report on.
Restrict the export privilege deliberately. It's the easiest route by which a full customer contact list leaves the tenant as a CSV.
If a client asks for near-instant segment-based entry, redirect the requirement to a trigger before you spend a sprint trying to make segments faster.

Worth another look if Microsoft changes the refresh tiers or the 120-day expiry, or exposes cadence as a per-segment setting.