TechFlow's campaigns are running across Google, Meta, and emails. But at this month's board meeting, your CFO asked a tough question, "We're spending $15K/month across three channels — which one drives revenue growth? And how do you know?" To answer that, you need to know how to test what's working, how to read where conversions come from, and how to prove that your ads are generating revenue — not just coinciding with it. That's exactly what this unit is about.

Most unskilled marketers run tests like this: change ad headline Monday, see conversions rise Tuesday, declare winner. But valid A/B testing requires controlled conditions — showing variation A and variation B simultaneously to similar audiences, with enough volume to detect real differences from random noise.

TechFlow's conversions might rise on Tuesday because it's payday, not because your headline improved. Without control groups and statistical significance, you're reading patterns in randomness.

A proper test isolates one variable (headline), keeps everything else constant (targeting, budget, timing), runs both versions simultaneously to equivalent audiences, and waits for statistical confidence before declaring a winner.

But test structure alone doesn't tell you if differences are real or random luck. You need a sample size calculation. It determines how many conversions you need to confidently detect a real difference.

Here's TechFlow's current situation: your Meta campaigns generate 50 conversions per week at $52 CPA. You want to test a new headline to see if it improves performance. Before designing the test, you need to define three key parameters that determine how much data you'll need.

These three parameters determine your required sample size. Let's see how this works.

You've seen how to design statistically valid A/B tests that prove what works — testing one variable at a time (headline A vs B) with proper sample sizes. But there's a different challenge: TechFlow now runs campaigns across multiple channels (Google, Meta, and email). When someone converts, they've usually touched several channels before buying. Which channel deserves credit for the conversion?

This isn't an A/B testing question — it's attribution. Let's see why this matters for TechFlow's budget decisions.

TechFlow is spending $15K/month across Google, Meta, and email. But which channel drives revenue, and where should we invest more? You can't just look at last-click conversions — that ignores the journey prospects take before buying. Attribution models help you assign credit across multiple touchpoints so you can make smart budget decisions. But different models tell completely different stories about what's working.

Here's TechFlow's typical customer journey: reads an email on Monday → clicks Meta ad on Wednesday → searches "TechFlow" on Friday → converts. Which channel deserves credit?

Same journey, three different answers about which channel worked.

Most marketers trust their attribution model without questioning its logic. But models don't reveal causation; they assign credit based on rules you choose. Before we see how that plays out in TechFlow's data, let's make sure the fundamentals are clear

Attribution models are simply rules for distributing credit — each with its own blind spots. No single model tells the full story of what actually drives conversions. That's why it helps to compare them side by side. Let's use Claude to run that analysis on TechFlow's real conversion paths and see how the story changes depending on which model you apply.

Here's the conversion data we're working with. TechFlow has four distinct paths users take before converting:

Notice how Google Search appears in almost every path — but that doesn't necessarily mean it deserves all the credit. Let's ask Claude to run the attribution comparison and see what each model says.

These numbers show why the model you choose changes everything. Email (25 total credits) and Meta (20 total credits) show zero under last-click. Google Search (75 total credits) closes everything, but mostly converts the demand that those two channels created first. Before cutting a channel due to poor performance, always check which attribution model you're using. Zero last-click credit doesn't mean zero impact.

Manually calculating attribution across 60 conversions and 4 paths would take hours. Claude does this in seconds — processing all paths, applying two models simultaneously, and revealing how credit shifts across channels. But here's what attribution can't tell you: did a channel cause conversions, or just capture them?

Google Search getting 83 total credits across both models might mean it's essential — or it might mean people search TechFlow's brand name after seeing a Meta ad, and Google just captures the demand Meta created. That's why you need incrementality testing to prove which channels actually drive growth.

Incrementality testing answers the question, "If I pause this channel, do conversions drop, or do they just shift to other channels?" You discovered which channels create demand versus which channels capture existing demand (not incremental). Without incrementality data, you might double-invest in a channel that only captures demand created elsewhere, or cut a channel that's quietly driving most of your growth.

Let's practice designing an incrementality test to identify which channels drive growth for TechFlow.

An incrementality test shows how many conversions truly vanish when Search ads stop. If conversions drop, Search drives real demand — keep investing. If they stay similar, Search just captures existing demand. Reallocate budget to demand-creating channels like Meta or Email. Attribution tells you who got credit. Incrementality tells you who actually earned it.

You've learned that incrementality testing reveals true impact beyond attribution's correlation. TechFlow's Search might get 60 attributed conversions but drive only 50 incremental conversions, meaning Meta and Email are the true growth drivers.

You can now answer the CFO's question, "Which channel drives revenue growth?" Designing statistically valid A/B tests with proper sample sizes proves what works. Use Claude to compare attribution models and understand how different rules shift credit across channels. Run incrementality tests to measure true channel impact — separating channels that drive growth from those that capture existing demand. This is how you turn marketing into a system that proves what works.

In the next lesson, you'll discover how to detect and prevent ad fraud that drains budgets on fake clicks and bot traffic, identify suspicious patterns in campaign data, and implement verification tools that protect TechFlow's ad spend from fraudulent activity.