---
title: "Explorations: funnels, paths, segments and cohorts"
description: "When to use explorations Use an exploration when a standard report cannot answer a specific question — for example where do mobile users drop out of…"
url: https://optimizeall.com/learn/web-analytics-with-ga4/explorations
updated: 2026-10-05
---

Web Analytics with Google Analytics 4 · Reports and explorations · lesson 14 of 20 · 12 min

# Explorations: funnels, paths, segments and cohorts

## When to use explorations

Use an **exploration** when a standard report cannot answer a specific question — for example *where do mobile users drop out of checkout?* or *what do people do after viewing pricing?* Explorations are private by default (share when ready), and they rely on event-level data subject to your **data retention** setting.

## The exploration techniques

| Technique | Best for | Example question |
|---|---|---|
| Free form | Flexible pivot tables and charts | Key event rate by landing page and device |
| Funnel exploration | Step-by-step conversion and drop-off | Where do users abandon checkout? |
| Path exploration | What happens before or after an event | What do users do after viewing pricing? |
| Segment overlap | How audiences intersect | How many purchasers also used site search? |
| Cohort exploration | Retention of groups over time | Do users acquired in March return more than those from April? |
| User lifetime | Value over the whole relationship | Which first-user channels produce the most lifetime revenue? |
| User explorer | Individual event streams (use responsibly) | QA of a specific test user journey |

## Segments: the most powerful feature

A **segment** is a subset of data defined by conditions. GA4 offers three kinds:

- **User segments** — users who met a condition at any time (e.g. users who purchased).
- **Session segments** — sessions that met a condition (e.g. sessions landing on a blog post).
- **Event segments** — only the events meeting the condition.

Choosing the wrong scope changes the answer. "Users who viewed pricing" includes all their sessions; "sessions with a pricing view" includes only those sessions.

## Building a funnel exploration

```
Question: Where do mobile users drop off between product view and purchase?
Steps:
  1. view_item
  2. add_to_cart
  3. begin_checkout
  4. add_payment_info
  5. purchase
Settings:
  - Open vs closed funnel: closed (must enter at step 1)
  - Directly followed vs indirectly followed: indirectly
  - Breakdown: device category
  - Optional: time elapsed between steps
```

**Open funnels** let users enter at any step; **closed funnels** require entry at step 1. **Trended** funnel views show whether a step's conversion is improving over time — great for checking the effect of a checkout redesign.

## Path exploration for discovery

Start from an event (for example `view_pricing`) and see the next steps, or start from an ending point (for example `purchase`) and look backward. Use it to discover unexpected loops — such as users bouncing repeatedly between pricing and FAQ, which suggests unanswered questions.

## Cohorts and retention

A cohort exploration groups users by acquisition date (or first event) and shows their return behavior over subsequent weeks. For subscription apps, a course platform or a grocery delivery brand, cohort retention is often a leading indicator of revenue far better than monthly visitors.

## Worked example: a course platform's pricing puzzle

An online academy notices many pricing views but few purchases. Analysis:

1. **Path exploration** from `view_pricing`: a large share of next steps go to the FAQ page and then exit.
2. **Free form**: FAQ-page sessions have a lower key event rate on mobile.
3. **Segment overlap**: most purchasers never visited the FAQ — they came from webinar emails.
4. Hypothesis: pricing lacks clarity about refunds and certificates, which the FAQ answers poorly on mobile.
5. Action: add a concise refund and certificate section on the pricing page; test it (see the CRO course).

This is the essence of analysis: **observation → segmentation → hypothesis → action**, not screenshots of charts.

## Sampling and limits

Explorations may be **sampled** when queries exceed quota thresholds for standard properties; an indicator shows the sample size. To reduce sampling, shorten the date range, simplify the query, or move heavy analysis to the **BigQuery export**, which provides raw event-level data (standard properties have a daily export limit; check current terms). BigQuery is also the route for joining GA4 data with CRM or cost data.

## Hands-on: the same funnel in BigQuery

Explorations are sampled on large ranges and limited by retention. For a durable version of your main funnel, compute it from the export. This approximate closed funnel uses each user's first occurrence of each step in the period:

```sql
WITH firsts AS (
  SELECT
    user_pseudo_id,
    MIN(IF(event_name = 'view_item',      event_timestamp, NULL)) AS t_view,
    MIN(IF(event_name = 'add_to_cart',    event_timestamp, NULL)) AS t_cart,
    MIN(IF(event_name = 'begin_checkout', event_timestamp, NULL)) AS t_checkout,
    MIN(IF(event_name = 'purchase',       event_timestamp, NULL)) AS t_purchase,
    ANY_VALUE(device.category) AS device
  FROM `my-project.analytics_123456789.events_*`
  WHERE _TABLE_SUFFIX BETWEEN '20260901' AND '20260930'
    AND user_pseudo_id IS NOT NULL            -- consented users only
  GROUP BY user_pseudo_id
)
SELECT
  device,
  COUNTIF(t_view IS NOT NULL)                                            AS step1_view,
  COUNTIF(t_cart > t_view)                                               AS step2_cart,
  COUNTIF(t_checkout > t_cart AND t_cart > t_view)                       AS step3_checkout,
  COUNTIF(t_purchase > t_checkout AND t_checkout > t_cart AND t_cart > t_view) AS step4_purchase
FROM firsts
GROUP BY device
ORDER BY step1_view DESC;
```

It will not match the exploration exactly (different counting rules, no modeling, consent-denied rows excluded), but it is stable, unsampled and can be scheduled.

## Second worked example: a Gulf grocery app cohort check

A grocery delivery app in the UAE sees monthly active users rising and assumes retention is improving. A cohort exploration by first-open week shows the opposite: recent cohorts return less in weeks two to four than cohorts from three months earlier. Growth is coming from heavy acquisition spend, not stickier customers. The team shifts part of the budget to onboarding improvements and sets week-four retention as a KPI.

## Common mistakes

- Building explorations for questions a standard report already answers.
- Wrong segment scope (user vs session).
- Treating a closed funnel's numbers as comparable to an open funnel's.
- Drawing conclusions from heavily sampled data without noting it.
- Forgetting that explorations cannot look back beyond the retention window.

## Video lecture: Explorations: funnels, paths, segments and cohorts

Lecture coming soon · 15 chapters · about 8 minutes. Read the full transcript below.

1. Explorations
2. Why explorations
3. Techniques
4. Segment scope
5. Building a funnel
6. Paths and cohorts
7. Example 1: the pricing puzzle
8. Example 2: UAE grocery app
9. Watch me do it, part 1
10. Watch me do it, part 2
11. Sampling and limits
12. Common mistakes
13. The essence
14. Recap
15. Try this now

## Lecture transcript

### Explorations

A standard report tells you that mobile conversion is down. It doesn't tell you where mobile users drop out of checkout, what they do after viewing pricing, or whether customers acquired this month are coming back. For those questions, you need explorations. In this lecture you'll learn the exploration techniques and when to use each one, why segment scope changes the answer, how to build a funnel, how paths and cohorts reveal what reports hide, and how to move a key funnel into BigQuery so it's unsampled and permanent.

### Why explorations

Why explorations? Because real business questions are specific. Where do mobile users abandon checkout? Do people who use site search buy more? Are recent customers stickier? Standard reports are built for monitoring, not investigation. Explorations let you pick the technique, the segments and the breakdowns that fit your question. They're private by default, so you can work freely, then share when ready. Just remember they rely on event-level data, so they're limited by your retention setting.

### Techniques

Here are the techniques. Free form: flexible pivot tables and charts, like key event rate by landing page and device. Funnel exploration: step-by-step conversion and drop-off. Path exploration: what happens before or after an event. Segment overlap: how audiences intersect, like purchasers who also used search. Cohort exploration: retention of groups over time. User lifetime: value over the whole relationship. And user explorer: individual event streams, to be used responsibly, mainly for QA of test journeys.

### Segment scope

Now segments, the most powerful feature, and the easiest to get wrong. There are three scopes. User segments include users who met a condition at any time, and all their data. Session segments include only sessions that met the condition. Event segments include only the matching events. Here's an analogy: user scope is inviting everyone who has ever bought a ticket to a concert. Session scope is counting only the nights they actually attended. Users who viewed pricing includes all their sessions. Sessions with a pricing view includes only those sessions. Different answers.

### Building a funnel

Building a funnel. Start with the question: where do mobile users drop off between product view and purchase? Steps: view item, add to cart, begin checkout, add payment info, purchase. Then settings. Open or closed: a closed funnel requires entry at step one, an open funnel lets users enter at any step. Directly or indirectly followed: usually indirectly. Breakdown: device category. And optionally, time elapsed between steps. Trended funnel views show whether a step is improving over time, perfect for checking a checkout redesign.

### Paths and cohorts

Path exploration is for discovery. Start from an event, like view pricing, and see the next steps. Or start from an ending point, like purchase, and look backward. Use it to find unexpected loops, like users bouncing repeatedly between pricing and the FAQ, which suggests unanswered questions. Cohort exploration groups users by acquisition date or first event and shows how they return over the following weeks. For subscription apps, course platforms and grocery delivery, cohort retention is often a better leading indicator of revenue than monthly visitors.

### Example 1: the pricing puzzle

First example, a simple one. An online academy notices many pricing views but few purchases. A path exploration from view pricing shows a large share of next steps go to the FAQ and then exit. A free-form table shows FAQ-page sessions have a lower key event rate on mobile. A segment overlap shows most purchasers never visited the FAQ; they came from webinar emails. Hypothesis: pricing lacks clarity about refunds and certificates, which the FAQ answers poorly on mobile. Action: add a concise refund and certificate section to pricing, and test it.

### Example 2: UAE grocery app

Second example, a business case. A grocery delivery app in the UAE sees monthly active users rising and assumes retention is improving. A cohort exploration by first-open week shows the opposite. Recent cohorts return less in weeks two to four than cohorts from three months earlier. Growth is coming from heavy acquisition spend, not stickier customers. The team shifts part of the budget to onboarding improvements and makes week-four retention a KPI. Without the cohort view, they'd have celebrated the wrong thing.

### Watch me do it, part 1

Watch me build the funnel. In a new funnel exploration, I add five steps: view item, add to cart, begin checkout, add payment info and purchase. I set it to closed, so users must enter at step one, and indirectly followed. I add a breakdown by device category and a date range of the last twenty-eight days. The chart shows mobile losing far more users between checkout and payment. I switch to the trended view and see the drop started on a specific date. I note it as a hypothesis, not a conclusion, and check the release log.

### Watch me do it, part 2

Then I make it durable in BigQuery. The lesson's query finds each consented user's first timestamp for each step in September, plus their device. Then it counts users who reached each step after the previous one. It won't match the exploration exactly, because the counting rules differ, there's no modeling, and consent-denied rows are excluded. But it's unsampled, it isn't limited by retention, and I can schedule it. So my key funnel now has a permanent history, not just a view I rebuild every month.

### Sampling and limits

A word on sampling and limits. Explorations may be sampled when queries exceed quota thresholds on standard properties, and an indicator shows the sample size. To reduce sampling, shorten the date range or simplify the query. For heavy or long-range analysis, use the BigQuery export, which provides raw event-level data, and is also the route for joining G A four with C R M or cost data. Remember that standard properties have a daily export limit, which module seven covers.

### Common mistakes

Common mistakes. Building explorations for questions a standard report already answers. Choosing the wrong segment scope. Comparing a closed funnel's numbers with an open funnel's. Drawing conclusions from heavily sampled data without saying so. And forgetting that explorations can't look back beyond the retention window, which is why the BigQuery version matters for long-term funnels.

### The essence

Here's the essence of analysis, and it applies to every exploration you'll build. Observation, then segmentation, then hypothesis, then action. Not screenshots of charts. A good exploration ends with a sentence that starts, we think this is happening because, and a test or fix that will prove it right or wrong.

### Recap

Recap. Explorations answer specific questions that reports can't. Segment scope, user, session or event, changes the answer, so choose deliberately. Use funnels for drop-off, paths for discovery and cohorts for retention. Watch for sampling and the retention window. And move your key funnel into BigQuery so it's unsampled and permanent.

### Try this now

Try this now. Build a funnel exploration, or specify one in writing, for your main conversion journey. Break it down by device, find the biggest drop-off, and write one hypothesis for it in the form: we think this is happening because. If you have BigQuery, run the lesson's funnel query for the same period and compare the shapes.

## Key takeaways

- Explorations answer specific questions that standard reports cannot.
- Segment scope (user, session, event) changes the answer — choose deliberately.
- Use funnels for drop-off, paths for discovery, cohorts for retention.
- Watch for sampling and the retention window; use BigQuery for heavy analysis.

## Try it

Build a funnel exploration (or specify one in writing) for your main conversion journey, break it down by device, and write one hypothesis for the biggest drop-off.

- [Previous: Reading standard reports with confidence](https://optimizeall.com/learn/web-analytics-with-ga4/standard-reports)
- [Next: From data to insight: an analysis workflow](https://optimizeall.com/learn/web-analytics-with-ga4/from-data-to-insight)
- [All lessons of Web Analytics with Google Analytics 4](https://optimizeall.com/learn/web-analytics-with-ga4)