Funnels, Cohorts and Dashboards for Decision Decisions
A step-by-step guide for PMs to turn a shipping choice into a testable hypothesis and answer it with funnels, cohorts and dashboards.

This guide is for working product managers who must decide what to build or stop. You will leave with a repeatable path: start from a shipping choice, state a tight hypothesis, and use a funnel, a cohort, and a dashboard in order to reach a decision you can stand by.
In brief
Dashboards are noisy. Funnels miss cohort shifts. Teams argue when there is no clear claim. Start from the decision, write one hypothesis, and map it to the right view.
- Decision first → hypothesis on behavior, metric, timeframe
- Map: funnel for where, cohort for permanence, dashboard for rollout
- Instrument only what is needed and define cohorts up front
- Follow one interrogation path and preregister the cadence
Who this is for and what you get
This is for PMs who need analytics to make a shipping choice. By the end, you will have a filled Hypothesis Card, a minimal instrument plan, and an ordered path through funnel → cohort → dashboard that ends in a clear go or no-go.
Step 1 · Name the decision
State the shipping choice in plain words. Example: Ship a shorter onboarding flow next sprint, or keep the current one. Tie it to a single user or business outcome. Write the exact choice you will make if the data supports it.
- What will you ship or stop?
- Which users are in scope?
- What outcome matters for this choice?
Step 2 · Translate to a testable hypothesis
Turn the choice into one or two tight hypotheses. Each must include a behavior, a metric, and a timeframe. Keep it narrow so it can be true or false. A Hypothesis Card keeps you honest and visible to the team.
Tip
Use a Hypothesis Card once, at the start, to lock the metric and cohort and to stop scope creep.
Step 3 · Map the hypothesis to the right view
Each hypothesis maps to one or more analytic views. Use the view that answers your claim with the least noise.
| View | Use when | Best for | Watch out for |
|---|---|---|---|
| Funnel | Tracing a sequence of events | Finding where users drop | Ignores who the user is and when they joined |
| Cohort | Comparing groups by start week, device, or variant | Seeing permanence of change over time | Masks step-level leaks; needs clear cohort rules |
| Dashboard | Monitoring a stable metric during rollout | Team visibility and trending | Noise and seasonality; can mislead without cohorts |
Step 4 · Design minimal instrumentation and cohort rules
Instrument only what your hypothesis needs. Define the cohort before you pull numbers. The cohort is the unit you will compare over the timeframe in the hypothesis.
- Events: name, when it fires, and required properties
- Users: inclusion rules (e.g., new mobile signups), exclusions
- Timeframe: window for conversion or retention
- Metric: exact formula and unit
Step 5 · Build the interrogation path
Run a sliced funnel for the cohort to see where the drop is. This isolates the step you will change.
Compare the target cohort over time (e.g., signup week) to see if behavior improves and stays improved.
Track the single metric from the hypothesis during ramp. Keep the cohort filter on while rolling out.
Step 6 · Preregister analysis and cadence
Make it boring
Write down your cuts, the success rule, the cohort, and how often you will roll up results. Share before you query. Then stick to it unless you restart the analysis.
How to do it well
- ■Name the decision and the exact ship or stop choice.
- ■Lock one metric, one cohort, one timeframe on the Hypothesis Card.
- ■Keep the hypothesis narrow. Avoid multi-metric success rules.
- ■Define events and properties before you implement changes.
- ■Use funnel → cohort → dashboard, in that order.
- ■Preregister cuts, decision rule, and rollup cadence.
- ■Check sample size and expected variance before you call the decision binary.
Common mistakes
- Mixing exploratory and decision analysis in one run
- Using overall averages instead of cohort comparisons
- Letting vanity metrics drive the call
- Changing definitions mid-analysis
- Ignoring sample size and noise for a yes/no ship decision
Worked example · Shorter onboarding and 7‑day retention
Decision: If a shorter onboarding increases 7‑day retention for new mobile signups, ship it to 100%. If not, keep the current flow and revisit the drop step.
Hypothesis: Reducing onboarding steps increases 7‑day retention by 15% for new mobile signups.
Events and properties
- app_install (platform)
- signup_start (platform, source)
- signup_complete (platform)
- onboarding_step_view (step_id, variant)
- first_session (platform, ts)
- active_session (platform, ts)
Cohort definition
- Users: first-time mobile signups (platform = mobile) in rollout period
- Include: users with signup_start followed by either variant
- Exclude: returning users; non-mobile platforms
- Index by: signup_week
Funnel steps (sliced by variant)
- signup_start →
- onboarding_step_view (step 1) →
- onboarding_step_view (final step) →
- signup_complete →
- first_session
Slice by variant = control vs shorter. This locates the largest delta by step.
Timeframe and metric
- Window: 7 days after signup_complete
- Metric: 7‑day retention = share of users with any active_session on day 7 ±1 day window
- Success rule: ≥ 15% relative lift for the target cohort
Cohort trend
Compare 7‑day retention by signup_week for control vs shorter onboarding. The effect should show up in the first affected week and persist in subsequent weeks as rollout expands.
Dashboard needle for rollout
- Chart: weekly 7‑day retention, filtered to new mobile signups
- Annotations: rollout milestones (start, 25%, 50%, 100%)
- Breakout: control vs variant until 100% ship, then overall with cohort filter
Interrogation path: First, use the funnel to see if the shorter flow reduces drop at a specific step. Next, use cohort trends by signup_week to confirm the lift is durable. Last, monitor the dashboard during ramp to ensure the lift holds at scale.
Decision rule: If the cohort analysis shows at least a 15% relative lift with stable week-over-week performance during ramp, ship to 100%. If not, do not ship and focus on the highest-leak step from the funnel.
When to bring in the Hypothesis Card
Use it once, up front. Write the behavior, metric, timeframe, and the ship/no-ship rule. Share it with engineering and design so instrumentation and scope stay aligned. Refer back to it when questions about cuts or extra metrics come up.
Before you call the decision
- ■Did you analyze the right cohort only?
- ■Is the effect visible in cohorts across at least two adjacent index periods?
- ■Is your sample size enough to detect the expected change?
- ■Are definitions unchanged from preregistration?
- ■Does the dashboard reflect the same metric and cohort as the analysis?
Put it to work this week
- Write one shipping decision in one sentence.
- Fill a Hypothesis Card with behavior, metric, timeframe, and decision rule.
- List the minimal events and properties you need.
- Run the funnel slice, then the cohort trend, then pin a dashboard chart.
- Share your preregistered plan and cadence with the team.


