Explorations: funnels, paths, cohorts, and better questions

Match the technique to the question

Free form is a flexible table or chart for dimensions, metrics, filters, and segments. Funnel exploration shows completion and abandonment across defined steps. Path exploration shows common next or previous nodes in a journey. Segment overlap compares shared membership, while cohort exploration follows groups that share a starting condition over time.

Do not start with a template because it looks impressive. Start with the question. “Where do users abandon checkout?” suggests a funnel. “What do users do after visiting help content?” suggests a path. “Do two behavior groups overlap?” suggests segment overlap.

• Free form: compare categories and values.
• Funnel: measure a defined sequence.
• Path: inspect common transitions.
• Cohort: compare behavior after a shared start.
• Segment overlap: examine intersections between groups.

Build a funnel with explicit rules

Define steps that reflect the real product journey, give each step a clear condition, and decide whether steps must occur indirectly or directly after one another. Set the time frame or elapsed-time constraints where the analysis requires them. An open funnel can allow users to enter at a later step; a closed funnel requires entry at the beginning.

Apply the same eligible population and date logic throughout. Then inspect both the completion rate and the absolute number at each step. A huge percentage drop on a tiny step may matter less than a modest leak affecting thousands of high-intent users.

A path is not a person’s whole story

Path reports aggregate transitions into common nodes. They simplify repeated events and depend on the starting point, node type, and included data. A path can show that many users move from a product page to cart, but it cannot tell you why any individual made that move.

Look for loops, unexpected exits, and measurement artifacts. Repeated page_view events may reflect single-page application handling rather than meaningful navigation. Pair behavioral evidence with usability research, session-level product knowledge, or experiments before making a causal claim.

Build a free-form exploration from the grain upward

Before dragging fields, decide what one row should represent. A row per landing page, campaign, device, item, or event leads to a different layout. Place the dimension that defines that grain in rows, use columns for a comparison only when needed, and choose metrics that are valid for that population.

Add segments and filters for different reasons. A segment defines the population entering the analysis; a filter restricts the values shown in the visualization. Confusing the two can change your answer. If you want to compare mobile and desktop within purchasers, define the purchaser population first, then split by device.

Start with a small number of rows and validate the totals. Once the base table makes sense, add breakdowns, date comparisons, or visual types. Complexity should be earned by a question that remains unanswered after the simpler view.

Understand funnel rules deeply enough to defend the result

The same events can produce different funnel results when entry, order, timing, or population rules change. A closed funnel asks whether users entered at the first step and followed the required sequence. An open funnel can count users who entered later, which is useful when later actions can occur independently.

Directly followed steps require the next step to happen immediately after the previous one under the exploration’s rules; indirectly followed steps allow intervening actions. Use direct steps only when intermediate behavior should invalidate the journey. Many real checkouts include legitimate page views or interactions between business steps.

Elapsed-time constraints answer a different question from the report date range. The date range selects the analysis period; a step time limit defines how long users can take between actions. Record both. Otherwise a “same-day checkout funnel” can be misread as every checkout that happened on that date.

Use path and cohort techniques for questions funnels cannot answer

Paths are exploratory rather than prescriptive. Use them when you do not already know the sequence you want to enforce. Start at an important event or page and inspect common next nodes, then switch direction to investigate what happened before an outcome. Simplify node choices where repeated technical events obscure meaningful behavior.

Cohorts follow users who share a starting condition over subsequent periods. They can reveal whether a campaign or onboarding change is associated with different return behavior. Keep cohort definitions, acquisition dates, and elapsed periods clear; a cohort with fewer mature days should not be compared naively with one that has had months to return.

Neither technique proves motivation. A loop in a path can be useful comparison behavior or confusing navigation. Stronger return in a cohort can reflect audience quality rather than onboarding design. Pair the pattern with qualitative research or a controlled test before assigning a cause.

Document an exploration so someone else can reproduce it

A chart title is not enough documentation. Record the exploration technique, date range, segments, filters, dimensions, metrics, step rules, and row grain. For funnels, include open or closed status, step conditions, direct or indirect rules, and elapsed-time settings.

Save a short interpretation note beside the output. State the observation, the largest contributing segment, the denominator, and any sampling, thresholding, or data-quality limitations that apply. This prevents a shared exploration from becoming a screenshot stripped of its assumptions.

When an exploration becomes a recurring stakeholder report, consider whether the definition belongs in a governed dashboard or reusable BigQuery workflow. Repeated manual rebuilding increases the chance that two people use slightly different rules while thinking they are looking at the same metric.

Worked example: analyze a SaaS onboarding funnel

A software team wants to understand activation from sign_up to create_project to invite_team_member. The funnel is closed because the analysis specifically requires the full onboarding sequence, but steps are indirect because users may legitimately read documentation between actions. A seven-day elapsed-time rule reflects the activation definition.

The first validated view shows absolute users and completion at each step. Only then does the analyst break it down by signup method or device. If invitation is the largest drop, alternative explanations include unclear collaboration value, permission friction, empty projects, or users who signed up for solo use.

The next evidence might include a path after project creation, usability research on the invitation step, and an experiment with a clearer collaborator prompt. The funnel locates the problem; it does not select the solution by itself.

Exploration interpretation checks before sharing

Explorations offer enough controls that two charts with the same title can represent different populations. Before sharing, reproduce the analysis from a clean state and compare the output. If you cannot describe the settings in plain language, the result is not ready for a stakeholder.

Pay particular attention to sequence timing, segment scope, and the difference between a filtered population and a filtered display. Record any sampling, thresholding, or incomplete-data behavior that affects the view.

• Analysis question and row grain are explicit.
• Date range and elapsed-time rules are separate and documented.
• Open or closed funnel logic matches the journey.
• Direct or indirect steps reflect legitimate intervening behavior.
• Segments and filters have not changed the population accidentally.
• Absolute counts accompany every important rate.
• Alternative explanations and follow-up evidence are included.