# Segment overlap

Put two saved segments side by side and see how many people are in both. A Venn diagram, three counts and a similarity score — the answer to whether two audiences you report on separately are really two groups or one group under two names.

## What this screen is for

Say you have two segments: "people who came from Google" and "people who purchased". Each has a number, and you can see both numbers elsewhere. What no other report tells you is **how many people are in both**.

Segment overlap, at `/overlap`, computes exactly that. The usual uses:

- **Checking whether channels overlap** — if two paid channels' user sets overlap heavily, they are touching the same person twice and their budgets cannot be judged independently.
- **Validating an audience before activating it** — before pushing an audience to a destination, see how much of it is the audience you are already targeting.
- **Separating two definitions you assumed were different** — two segments at near one hundred percent similarity are one segment with two names.

It sits in the sidebar under the Activate group, beside Audiences.

## Prerequisite: two saved segments

This screen defines nothing of its own; it compares segments you have already saved. With fewer than two, a note appears at the top telling you to create at least two segments first.

Segments are built at `/segments` — listed in the sidebar under the Coming soon group. Each has a name, a scope (user, session or event) and one or more conditions joined with AND. Conditions are written against the usual dimensions:

| Field | What it filters on |
| --- | --- |
| `ft_channel` / `ft_source` / `ft_medium` / `ft_campaign` | the user's first touch — channel group, source, medium, campaign |
| `lt_channel` | the last touch's channel group |
| `geo_country` | country |
| `ua_device_class` | device class |
| `event_name` | event name |
| `page_path` | page path |

The operators are equals, not equals, contains, and the two numeric comparisons `>` and `<`.

## Run the comparison

1. Open Segment overlap in the sidebar.
2. Pick Segment A in the first list and Segment B in the second.
3. Select Compute overlap.

> The overlap of two or three user segments, as a Venn diagram. — [analytics.adpix.io/en/overlap](https://analytics.adpix.io/en/overlap)

The result is a Venn diagram plus four lines of numbers:

| Line | Meaning |
| --- | --- |
| Segment A users | users meeting the first segment's conditions inside the selected range |
| Segment B users | the same, for the second segment |
| In both | users who are in both segments |
| Jaccard similarity | the intersection divided by the union, as a percentage |

The numbers drawn inside the Venn diagram are **exclusive**: each circle shows its own private share and the middle number is the intersection. So the figure on the left of the diagram is not the same as "Segment A users" — that one is the whole segment, this one is the whole minus the intersection.

### What the Jaccard similarity adds

"In both" is sensitive to segment size: a thousand shared users between two thousand-user segments means something entirely different from a thousand shared users between two hundred-thousand-user segments. Jaccard normalises that — the intersection over the union of the two. Zero percent means nothing is shared; one hundred percent means the two are the same set. When comparing several pairs of unequal size, this is the comparable number, not the raw intersection.

## The grain, and what may surprise you

Two things to know, or you will compare these numbers against the wrong report.

**The unit is the user.** Not the session, not the event. For each user, AdPix decides whether they have at least one event inside the selected range matching the segment's conditions; then it counts users.

**The segment's scope is ignored here.** A segment saved with session or event scope is still evaluated at user grain on this page. That is deliberate — overlap only means something between sets of people — but it does mean the same segment can produce a different number when used elsewhere in Explore.

> **The date range matters**
>
> This report follows the global date picker at the top of the page and only sees events inside it. So "Segment A users" means "users who met segment A's conditions in this range", not everyone who has ever met them. Change the range and all four numbers change. The range bounds are evaluated in the property's timezone.

Smaller points:

- Users are counted by their global user ID, and anonymous events from before they identified are attached to the same person — so someone who browsed anonymously and then signed in counts once, not twice.
- Test-mode events appear in none of the four numbers.
- The console compares two segments. The service behind it accepts up to three and returns the pairwise and three-way intersections as well; send more and only the first three are used.
- If a segment carries no valid condition, every user in the range counts as a member — a number that looks like a bug at first glance but is simply an empty segment.

## Frequently asked questions

### Why does the page tell me to create two segments first?

Because this screen defines nothing of its own; it only compares segments already saved on the property. Until at least two saved segments exist, the pickers are empty. Segments are built at `/segments`.

### What does the Jaccard similarity actually tell me?

The intersection divided by the union — users in both, divided by the two segments combined minus the users in both. Zero percent means no shared users; one hundred percent means the two segments are exactly the same set. Unlike the raw "In both" count it is not sensitive to segment size, which makes it the number to compare across pairs of different sizes.

### Does my segment's scope (user, session or event) change anything here?

No. This report always evaluates at user grain — a user counts as a member if they have at least one event in the selected range matching the segment's conditions. The scope you chose when saving the segment matters elsewhere, not here.

### Why does a segment's count differ from what I see in other reports?

Two usual reasons. The date range — this page follows the global date picker and only sees events inside it. And the grain — the unit here is the user, not the session and not the event.

## Related

- [Segments](https://docs.adpix.io/en/analytics/explore/segments/)
- [Explorations](https://docs.adpix.io/en/analytics/explore/explorations/)
- [Audiences and destinations](https://docs.adpix.io/en/analytics/admin/audiences-and-destinations/)
- [Date ranges and comparison](https://docs.adpix.io/en/analytics/reports/date-ranges-and-comparison/)

---

[Docs](https://docs.adpix.io/en/analytics/explore/segment-overlap/) · AdPix