How segments are built from goal sets

Tailkey generates segments automatically from its attribution scoring. Each segment is a group of visitors identified by behavior and linked to device IDs for targeting. This page describes how segments are created and how to use them.

Segment types, and when each becomes available

Availability depends on whether a goal set is defined
Raw events
GA4 or Tailkey tracker
General segments
available the moment data lands
You pick a Goal Set
the only manual step
Attribution engine runs
journeys → model assembly → score
Attribution segments
one set per goal

General segments are computed from raw behavior and are available before a goal set is defined. Attribution segments are computed from the attribution model's output and become available after a goal set is defined and the pipeline has run.

Segment types

Tailkey computes two types: general segments, built from raw behavioral data, and attribution segments, built from the attribution model's output. This distinction determines when each type becomes available and what question it can answer. The comparison below shows both:

Available immediately
General segments
Computed directly from raw event data, including source, device, behavior, and recency. No goal set is required. These segments are available as soon as data connects.
Built from
Raw GA4 or Tailkey tracker events
Depends on
None; runs on a schedule
Answers
Who these visitors are, and whether they are still engaged
Examples
Acquisition source, technology, behaviour, recency and frequency, churn risk
Needs a goal set
Attribution segments
Computed from the attribution model's output: the journeys it reconstructed, the model assembly behind it, and how strongly it trusts each path. These are available after a goal set has been defined and the pipeline has completed a run for it.
Built from
Journeys, blended model scores, journey strength
Depends on
A goal set, and one completed pipeline run for it
Answers
Who was close to this goal, and what nearly caused conversion
Examples
High-impact engagers, drop-off retargeting, near-miss, graduation

Because attribution segments are computed once per goal, the same visitor can appear in a segment for one goal and not for another. This is expected: proximity to one goal does not imply proximity to another, and the model reports each goal separately.

The goal set is not a filter applied at the end of the pipeline. It defines what the entire model measures.

One site, three goals, three outcomes

A goal set defines the event that counts as a conversion: a purchase, a signup, or a demo request. Tailkey runs the attribution pipeline once for each goal set, producing separate journeys, scores, and segments. The following example uses the same site with three different goal sets:

Goal Set
Pipeline last ran
Aug 12, 2026 · 14:02 UTC
Converting journeys
12,480
Goal event
/checkout/complete
The event that defines success for this goal set. All downstream scoring is measured against it.
Most pivotal touchpoint
block:pricing-table
The touchpoint with the highest removal effect: journeys without it convert less often.
Retargeting deadline
16 days
Calculated from this goal's return-probability curve.
A journey towards this goal
google / organic
/pricing
block:pricing-table
/checkout/complete

For the purchase goal, the pricing table has the highest removal effect: journeys that skip it convert less often. Segments built from this goal are centered on that page.

When the run finishes, the segments below are available. They are generated by the model, not defined manually.

Segment catalog

A completed pipeline run produces the following segments, grouped by the decision they support: five are audiences you can target directly, two indicate how much to trust those audiences, and one controls timing. All figures update when you change the goal set.

Showing figures for
Audiences you can act on
Each of these segments can be used in a campaign immediately. The model provides the reason each segment exists, which can be used directly in messaging.
High-impact touchpoint engagers
Visitors who engaged with pages the model has identified as high-impact, ranked by how many they reached and their scores.
Touchpoints reached
Devices
4 or more
96
Deeply engaged with proven pages
2–3
412
Meaningful engagement, not yet convinced
Just one
1,204
Single high-impact touch
Use for expansion campaigns. These visitors have already engaged with high-performing pages, regardless of conversion status.
Attribution-aware drop-off
Visitors who did not convert but passed through a touchpoint the model identifies as pivotal for this goal.
Where they stalled
Devices
block:pricing-table
212
Highest removal effect for this goal
/demo-request
96
Second most pivotal exit point
Retarget with content addressing the specific page where the visitor stalled.
Time-bound high-intent drop-off
The same population as above, segmented by whether retargeting spend is likely to be effective.
Urgency tier
Devices
Retarget now
412
Inside the window, so act before it closes
Too early
1,204
Still a normal browsing gap; don't spend yet
Likely lost
3,880
Past the point retargeting historically recovers
This 16-day deadline is calculated from historical return-rate data for this goal, not a fixed default.
Habitual near-miss
Visitors who stalled at the same pivotal page across multiple visits.
Times stalled
Devices
4 or more
19
A specific, recurring blocker
3
64
Repeated hesitation
2
288
Second pass, worth a nudge
Repeated hesitation typically indicates a specific objection. Address it directly with a discount, FAQ content, or live chat.
Multi-goal graduation
Visitors who completed one goal and show intent toward a second goal.
Goal pair
Devices
Signup → Purchase
288
Score 0.71 toward the second goal
These are existing customers. Owned channels are more effective than paid retargeting for this segment.
Diagnostic segments
These segments are diagnostic, not audiences. They indicate which of the segments above are based on high-confidence signal, and which require manual review.
Journey strength tiering
The model's confidence level for each reconstructed journey.
Tier
Devices
High strength
3,204
Build audiences from these with confidence
Moderate
5,110
Usable, treat conclusions as directional
Low strength
1,900
Noisy and exploratory, so exclude from audiences
Apply this filter to any other segment on this page; it is not a standalone audience.
Model disagreement
Steps where the model assembly disagrees internally about who deserves the credit.
Step
Journeys
/features
88
Highest spread across the model assembly
block:hero-cta
54
Worth a manual look before you trust it
Review these journeys manually before using them in a campaign.
Timing, not targeting
A modifier applied to any audience above. It determines follow-up timing, not segment membership.
Consideration-cycle length
The time converted visitors took from first touch to goal completion.
Decider tier
Avg time to convert
Fast
8.4 hrs
1,890 devices, so follow up quickly
Moderate
78.2 hrs
3,204 devices on a standard cadence
Slow
412.6 hrs
640 devices, so give them room
Use this to set follow-up cadence. Fast deciders respond to immediate contact; slow deciders require more time.
Exporting segments
Every segment exports as a device ID list

Segments are not limited to reporting. Any tier or row above can be exported from the dashboard as a list of device IDs, which can be uploaded as a custom audience to Google Ads, Meta, or an email platform. Select the tiers to export; the resulting campaign targets exactly the segment identified by the model.

✓
Retarget now
412
✓
Habitual near-miss, 3+
83
Too early
1,204
495 device IDs selected
Export all

General and attribution segments are most useful together: one identifies who is at risk, the other explains why it matters.

Combining churn risk with attribution

Churn risk is a general segment. It tracks engagement decay across the site and identifies visitors who are becoming less active, without requiring a goal set. On its own, it identifies who to prioritize but not why. Combined with an attribution segment, it produces a shortlist with a stated reason:

Goal-agnostic
Churn risk
Recency, frequency trend, engagement-time trend. Scored daily, no goal required.
Per goal
Attribution drop-off
The touchpoint the model identifies as pivotal, and the model's confidence in that journey.
Together
A ranked shortlist
Ranked by churn risk, then by journey confidence, with the page to address for each.
Churn risk
Where they stalled
Trust
Goal
High
block:pricing-table
0.81
Purchase
High
/demo-request
0.74
Signup
Medium
block:plan-comparison
0.68
Signup

Each row combines churn risk with attribution data. For example, a visitor with high churn risk who stalled at the pricing table, with high journey confidence, for the purchase goal, indicates both an at-risk visitor and the reason for the risk. Export the row to get device IDs for targeting.

Define a goal set to generate segments

General segments are available as soon as your data connects. Define a goal set to generate attribution segments after the first pipeline run. Segments are scored against your visitors and can be exported for use in campaigns.

The model behind it

How Tailkey's attribution modeling works

See how touchpoints are scored before they become segments.

Read the attribution guide →