Skip to content
Academy
Marketing Academy · Field Work●Analytics & Attribution
All projects·From: Predictive Analytics
MiniHead-to-Head· 30 minutes

Spreadsheet Rule vs. GA4 Model: A Churn-Scoring Head-to-Head

Coinbase

Objective: Given the same 20-customer behavior export, score churn risk two ways, a simple rule-based spreadsheet model and GA4's built-in churn probability logic, and decide which one a lean team should operate now.

Coinbase's retention team won't get a data-science hire for two quarters. You have to decide whether a rule-based spreadsheet score is good enough to run retention campaigns on until GA4's predictive audiences fully activate.

Score both models against the same customers, compare both to actual outcomes, and make the call on which one ships this week.

Before you start

What you'll need

Free path (everything below is enough to finish)

Google Sheets
FreeBuild the rule-based score and compare both methods side by side

Free, and the scoring logic stays visible and auditable to a non-technical retention team

Google Analytics 4
FreeSource the free built-in churn-probability metric to compare against the rule-based score

No cost, no setup beyond an active GA4 property with sufficient conversion volume

The process

2 steps

Step 01 of 02

Building a rule-based churn score without a data science team

The lesson's Step 3 describes a rule-based propensity model: take past churned customers, write down what they had in common (days since last purchase, low email opens, no logins), and build a simple scoring rule from those signals.

Past churned Coinbase users share 3 traits: no login in 45+ days, 2+ failed payment attempts, and a support ticket in the last 30 days. How do you turn that into a score for the current 20-customer list?

Google Sheets— Import customer-behavior-export.csv and add a points column per rule.

Procedure

  1. Import customer-behavior-export.csv
  2. Add 1 point per matched trait: no login 45+ days, 2+ failed payments, recent support ticket
  3. Sum points per customer into a rule_score column (0-3)
  4. Flag anyone with rule_score >= 2 as high risk
Sample output
customer_id  no_login_45d  failed_pay  support_ticket  rule_score  flag
C-1042       1             1           0               2           HIGH
C-1058       0             0           1               1           low
C-1071       1             1           1               3           HIGH

Healthy

The rule-based score correctly separates most of the 20 customers into sensible risk tiers using only data already sitting in the CRM, no model training required.

Unhealthy

The rule flags almost everyone or almost no one as high risk, meaning the point thresholds need recalibrating against real churned-customer traits, not guessed.

What this means

A rule-based score is a legitimate propensity model, it's just hand-built instead of machine-learned, and it works when the underlying traits are genuinely predictive.

So what do I do about it?

SymptomActionEffort
The rule-based flag matches almost no one to 'high risk'Re-derive the point thresholds from a fresh sample of the last 50 actually-churned customers30 min
YouYou can do this yourself, no engineering access required.

Step 02 of 02

Using GA4's built-in predictive metrics as a free starting point

The lesson notes GA4 activates free predictive metrics, including churn probability, automatically once a property has at least 1,000 returning users who triggered the conversion event in the past 28 days.

GA4's churn-probability export for the same 20 customers flags 7 as high risk. The rule-based score flags 9. Comparing both to actual 30-day outcomes, which one should the team operate on this week?

Google Sheets— Add a ga4_flag column next to rule_score, then compare both against actual_churned.

Procedure

  1. Add GA4's churn-probability flag as a new column
  2. Add the actual 30-day outcome column
  3. Calculate how many true positives each method produces
  4. Compare against the effort each method requires to maintain
Sample output
Method            Flagged   True Positives   Setup Effort
Rule-based score  9         6                Built today, in Sheets
GA4 churn model   7         6                Already running, free, updates automatically

Healthy

Both methods catch a similar number of true churners, so the team picks GA4's version since it updates automatically with zero manual maintenance.

Unhealthy

The team keeps maintaining the rule-based spreadsheet manually every week even though GA4's free model performs equally well with no upkeep.

What this means

When two methods perform comparably, the tiebreaker is which one keeps working without someone manually re-running it every week.

So what do I do about it?

SymptomActionEffort
Both methods catch a similar share of real churnersRetire the manual rule-based sheet and route retention triggers off GA4's predictive audience instead30 min
YouYou can do this yourself, no engineering access required.

Final deliverable

A head-to-head scorecard comparing the rule-based score and GA4's churn probability against actual outcomes, with a recommendation on which to operate this quarter.

See a reference example
Sample output
Chewy, Churn-Scoring Head-to-Head (excerpt)

Rule-based score: 8 flagged, 6 true positives, manual weekly upkeep
GA4 churn model: 6 flagged, 6 true positives, automatic
RECOMMENDATION: Adopt GA4's model. Equal accuracy, zero manual maintenance.

Success criteria

You're done when you can:

  • Builds a working rule-based score from named customer traits
  • Correctly compares both methods' true-positive counts against actual outcomes
  • Recommendation weighs maintenance effort, not just raw accuracy