Funnel Analysis
A funnel is an ordered measurement rule over events, not a picture of an inevitable customer journey. Before counting, specify who may enter, whether entry must begin at step one, which repeated occurrence wins, how long a sequence may take, and which clock closes the data. Different defensible rules can produce different rates.
Prerequisites
Complete identity resolution and privacy boundaries. This lesson reuses its frozen identity snapshot, UTC timezone, consent boundary, event validation, duplicate rule, and late-arrival policy.
Learning goals
- Distinguish open from closed funnels and direct from indirect step ordering.
- Define eligibility, step order, elapsed-time windows, and repeated attempts before computing.
- Keep the funnel denominator attached to the entry population.
- Separate late-arrival restatement from event-time ordering.
Specify the sequence first
For the signup question, use a closed actor-level funnel: signup_started followed by account_created, with intervening events allowed. The first eligible start per actor is the attempt. Completion must occur at or after that start and no later than 15 minutes afterward, using occurred_at. Reception time determines whether a row is available at a close; it does not rewrite the behavioral order.
The interval is inclusive at 15 minutes. A completion one second later fails the rule. A completion before its chosen start is not repaired by sorting. Actors without a valid, consented start never enter the denominator. An account creation can therefore exist without being a conversion in this funnel.
Metric contract
- Population: actors with a valid, consented
signup_startedduring 2026-10-01 UTC. - Unit: resolved actor × first eligible signup attempt.
- Observation window: entry in
[2026-10-01T00:00:00Z, 2026-10-02T00:00:00Z); completion from entry through 15 minutes inclusive; restated through 2026-10-04. - Denominator: distinct eligible actors entering step one; never all visitors or all accounts.
- Missingness: malformed, denied, or not-yet-received events are excluded and reported; no completion means incomplete, not proof of abandonment intent.
- Identity: observed
user_id, then the approved 2026-10-02 identity snapshot, otherwiseanonymous_id; shared-device uncertainty remains.
Worked example
import json
from datetime import datetime, timedelta
from pathlib import Path
root = Path("tests/fixtures/product-analytics")
rows = [json.loads(line) for line in (root / "events-v1.jsonl").read_text().splitlines()]
plan = json.loads((root / "tracking-plan-v1.json").read_text())
parse = lambda value: datetime.fromisoformat(value.replace("Z", "+00:00"))
deduped = {}
for row in sorted(rows, key=lambda item: parse(item["received_at"])):
deduped.setdefault(row["event_id"], row)
def valid(row):
required = plan["events"][row["event_name"]]["required_properties"]
types = {"string": str, "boolean": bool}
return all(key in row["properties"] and isinstance(row["properties"][key], types[kind]) for key, kind in required.items())
def actor(row):
return row["user_id"] or plan["identity_snapshot"]["links"].get(row["anonymous_id"], row["anonymous_id"])
eligible = [row for row in deduped.values() if row["analytics_consent"] and row["purpose"] == "product_measurement" and valid(row)]
starts = {actor(row): parse(row["occurred_at"]) for row in eligible if row["event_name"] == "signup_started"}
created = {actor(row): parse(row["occurred_at"]) for row in eligible if row["event_name"] == "account_created"}
converted = {key for key, start in starts.items() if key in created and start <= created[key] <= start + timedelta(minutes=15)}
first_close = datetime.fromisoformat("2026-10-03T00:00:00+00:00")
first_available = {actor(row) for row in eligible if row["event_name"] == "account_created" and parse(row["received_at"]) <= first_close}
print(f"starters={len(starts)} completed={len(created)} within_15m={len(converted)} conversion={len(converted)/len(starts):.3f}")
print(f"first_close={len(first_available)}/{len(starts)} restated={len(converted)}/{len(starts)} ordering=occurred_at")
Expected output:
starters=2 completed=2 within_15m=2 conversion=1.000
first_close=1/2 restated=2/2 ordering=occurred_at
The restatement changes availability, not the historical timestamps. On this tiny fixture, the rate is a worked arithmetic result—not evidence that the funnel will generalize.
Limitations and common mistakes
- Switching between open and closed funnels without versioning the metric.
- Letting a later step precede an earlier one after a broad join.
- Treating “drop-off” as a reason: missing events, cross-device identity, deliberate delay, and instrumentation failures can look identical.
- Comparing cohorts with unequal follow-up or an unannounced late-arrival restatement.
- Reading a funnel difference as a causal effect of a product feature.
Practice
Recompute the funnel with a five-minute inclusive window, then with an open funnel. Predict which actors change before writing code. Add a completion exactly at 15 minutes and one at 15 minutes plus one second to test the boundary.
Next steps
Cohort analysis freezes an inclusion event and compares behavior across aligned periods.
Primary sources
Checked 2026-10-19: Google Analytics funnel exploration, OpenTelemetry event timestamp conventions, and W3C privacy principles on data minimization and purpose limitation.