The 3-Layer Protocol for International Funnel Analysis

Your country-level conversion rate is the most dangerous metric in your international dashboard.

Not because it’s wrong. Because it’s precisely wrong — an exact number with almost no relationship to reality.

One brand almost shut down a market doing six figures a month in net profit. Blended CR had drifted from 3.2% to 0.9% over a quarter. The board voted to exit. Before the kill order went through, we ran this protocol and found 40% of that market’s traffic was phantom “direct” — sessions that GA4 couldn’t attribute because the referring app (Maps, Instagram, LINE, a QR code) opened an external browser without passing referrer headers. We changed nothing in the funnel. The “CR problem” disappeared. The market is still running.

That’s not an edge case. I’ve watched three expansion teams nearly exit healthy markets in the last year. Every time, the “underperformer” had a traffic source mix problem, not a market problem.

Here’s the protocol I now run on every international funnel before anyone touches the exit button.


The 3-Layer Protocol
The 3-Layer Protocol: Source Normalization → D:G Diagnostic → Constraint Segmentation

Layer 1: Source Normalization

Never compare blended country CRs. A country with 80% high-intent Google search traffic will always outperform one with 80% app-referral traffic — even if the product experience is identical.

Pull your GA4 data by source/medium for each country. Compare Google organic CR across countries. Not blended. Google-to-Google.

When the gap between countries collapses once you normalize by source, your “country problem” is actually a “source mix problem.” You don’t need to fix the market. You need to fix what’s sending it traffic.


Layer 2: The D:G Diagnostic

D:G is Direct-to-Google ratio — (direct)/(none) sessions divided by google/organic sessions, per country.

When direct massively outweighs Google in a specific country, something is injecting phantom direct sessions. The three biggest culprits: Google Maps “Reserve” buttons, social app bio links, and QR codes without UTM params. All open a browser without referrer headers. GA4 has no way to tell them apart from a real bookmark visit.

I’ve seen this inflate traffic by 25-30% across entire platforms. Users landing for 20 seconds and bouncing. Not prospects. Accidental visitors dragging your blended CR down.

The Thresholds

D:G > 3x AND direct engagement < 30% = Phantom direct. Not real users.

D:G < 1.5x with engagement > 50% = Healthy.

D:G between 1.5x and 3x = Investigate.

Example: Country A has 10k direct sessions, 2k Google sessions, 1% engagement. D:G is 5x — almost all noise. Country B has 500 direct, 800 Google, 63% engagement. D:G is 0.6x — real users.

The Stripe/Direct Proxy

Second lens. Stripe payment redirects back from checkout.stripe.com. Calculate Stripe referral sessions divided by direct sessions per country.

Stripe/Direct > 10% = Real returning users. < 2% = Noise.

When D:G and Stripe/Direct both point the same direction, you have your answer.


D:G Diagnostic Thresholds
D:G Diagnostic: Three zones for identifying phantom direct contamination

Layer 3: Constraint Segmentation

Layers 1 and 2 strip the noise. Layer 3 finds the real constraints.

Look at step-level CR by country by device. Build the matrix. The red clusters are where your money leaks. The constraint is almost never “this country doesn’t convert.” It’s “mobile users in this country drop off at this specific step for this specific reason.”

One pattern I keep seeing: mobile UX scores higher on heuristic audits than desktop, yet desktop converts 2x better. The gap is intent quality, not UX quality. Desktop users arrive via search. Mobile arrives via everything, including phantom direct. Strip the source mix first, then diagnose the step.


The A/B Paradox: Losing at 3 Steps, Winning Overall

A variant added transparency — pricing, cancellation policies, commitment details — directly on the page. It lost at 3 of 5 funnel steps. Top-of-funnel dropped 10-15%. The team wanted to kill it.

But form completion jumped nearly 60%. Users who saw the real cost upfront and still started the form were pre-qualified. They completed at almost double the rate.

Net result: over 20% more payments. Statistically significant.

Higher drop-off at the TOP + higher conversion at the BOTTOM = quality filter, not leak. The users lost at the top had near-zero conversion probability. They were going to abandon at the form, or at payment, or after seeing the final price. The transparency just moved their exit point earlier in the funnel.


The A/B Paradox
The A/B Paradox: Fewer enter, more complete, more payments

What Your CFO Needs to See

If your board deck shows blended CR by country without source normalization, you are making million-dollar decisions on polluted data.

Three numbers per market. Non-negotiable.

1. Real Addressable CR. Conversion rate of high-intent traffic only. Strip phantom direct, bots, misattributed sessions. The gap between reported CR and real CR is usually 30-50%.

2. Addressable Market Size. Total sessions minus noise. A market showing 800 sessions might have 300 real ones. “0% CR” on 300 sessions isn’t a market failure — it’s an insufficient sample. One brand almost exited a market on this exact misread.

3. Source-Normalized Revenue Potential. Real addressable users x real CR x AOV. This is the only number worth making investment decisions on.

Stop presenting blended CR by country. Start presenting: “Market X’s real addressable CR from intent traffic is approximately 4%, consistent with healthy markets. The blended 0.5% is distorted by phantom direct. The opportunity is growing intent-traffic volume.”


Implementation Checklist

Run this in one week. No consultants needed.

  • Day 1: Add UTM params to every touchpoint that opens an external browser — Maps links, QR codes, social bios, messaging share URLs, push notifications. Format: ?utm_source={source}&utm_medium=referral&utm_campaign={campaign}. This tags phantom direct at the source.
  • Day 2: Pull D:G ratio per country. Flag anything above 3x with sub-30% engagement. Pull Stripe/Direct ratio. Flag anything below 2%.
  • Day 3: Calculate source-normalized CR per country. Google organic CR, apples-to-apples. Document the gap between blended and real.
  • Day 4: Build the step-level constraint map by country x device. Find the red clusters.
  • Day 5: Rewrite your board metrics using source-normalized numbers. Reclassify any markets currently marked for exit.
5-Day Implementation
Run the 3-Layer Protocol in one week

Then monitor D:G weekly. Noise is permanent. Exiting good markets by mistake is optional.

DIAGNOSTIC

Free Margin Diagnostic

Green, Yellow, or Red verdict on your next market.

  • Verdict per market in 5 minutes
  • True CAC + payback modeled
  • No call required

CONSULTING

Book a Verdict Call

30-min call. One named market. Full P&L.

  • Live unit-economics review
  • Green, Yellow, or Red verdict
  • Fee fixed, not per market

Brief from the Field

One email a week. One market-entry case. One framework. No filler.

The form you have selected does not exist.