Skip to main content

Engagement Frequency and Intensity

Engagement is not one natural number. Frequency asks how often behavior occurs, intensity asks how much behavior occurs per actor or session, breadth asks how many capabilities are used, and recency asks how long since the last qualifying action. A composite score can conceal those distinctions and silently reward instrumentation volume.

Prerequisites​

Complete activation and feature adoption. Continue using the frozen event population and successful-use semantics.

Learning goals​

  • Match frequency, intensity, breadth, and recency to a decision.
  • Report distributions rather than only a mean or total.
  • Test threshold sensitivity and denominator changes.
  • Recognize event-volume, sessionization, and background-event bias.

Choose a behavioral measure​

For this fixture, frequency is the count of valid page_view and feature_used events per eligible actor in one UTC day. It is not time spent: the fixture has no foreground-duration field. It is not feature breadth: only one valid feature value appears. Counting unsupported proxies would manufacture precision.

The median describes the middle actor in this four-actor teaching population. The threshold counts at two and three events expose sensitivity. In production, also show the full distribution, zero-eligible actors when the decision requires them, bot/internal-traffic policy, and quantiles robust to heavy users.

Metric contract​

  • Population: actors with at least one valid, consented page_view or feature_used during 2026-10-01 UTC.
  • Unit: resolved actor × UTC calendar day.
  • Observation window: one half-open UTC day at the 2026-10-04 restated close.
  • Denominator: four eligible active actors; the threshold share uses this same population.
  • Missingness: malformed feature events are excluded; absence of foreground time prevents a duration metric; collection outages would require a coverage flag.
  • Identity: frozen 2026-10-02 mapping; anonymous actors remain separate unless an approved deterministic link exists.

Worked example​

import json
from collections import Counter
from datetime import datetime
from pathlib import Path
from statistics import median

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"]
kinds = {"string": str, "boolean": bool}
return all(key in row["properties"] and isinstance(row["properties"][key], kinds[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"])
counts = Counter(actor(row) for row in deduped.values() if row["analytics_consent"] and row["purpose"] == "product_measurement" and valid(row) and row["event_name"] in {"page_view", "feature_used"})
values = sorted(counts.values())
at_two = sum(value >= 2 for value in values)
at_three = sum(value >= 3 for value in values)
print(f"actors={len(values)} event_counts={values} median_frequency={median(values):.1f}")
print(f"actors_at_2_events={at_two} actors_at_3_events={at_three} threshold_sensitivity=True")

Expected output:

actors=4 event_counts=[1, 2, 2, 2] median_frequency=2.0
actors_at_2_events=3 actors_at_3_events=0 threshold_sensitivity=True

Limitations and common mistakes​

  • Treating more events as better when repeated actions may signal friction.
  • Comparing counts after an instrumentation release that emits more events per action.
  • Averaging per-day ratios instead of recomputing from declared numerators and denominators.
  • Using session duration when background tabs or apps distort exposure time.
  • Selecting an “engaged” threshold because it maximizes association with a later outcome.

Practice​

Define one frequency metric and one breadth metric for a reporting product. State how internal traffic, retries, background events, zero activity, and a telemetry outage affect each. Compare thresholds of one, two, and three qualifying actions.

Next steps​

Behavioral segmentation turns declared behavioral rules into reproducible groups without turning descriptions into personas or causes.

Primary sources​

Checked 2026-10-19: Google Analytics engagement definitions, Google Analytics session definitions, and Google's HEART framework paper.