You're Optimising for the Average Customer. The Average Customer Doesn't Exist.
Your Google Ads conversion rate is 3.4%. Not bad. Then one month it drops to 2.9%. The obvious move: pause something, rewrite the ads, complain to your agency.
Here's what actually happened in a real SaaS account. Desktop conversions had improved 30% after a site redesign. But a speed bug on Android mobile crushed mobile conversion to barely 1%. The blended number went down. The reality underneath went in two completely different directions.
The business nearly pulled budget from a campaign that was working better than ever. Because they were looking at the average. And the average customer doesn't exist.
This isn't a Google Ads problem. It's a thinking problem. And it affects every metric you look at, every report you read, and every decision you make about where to spend your marketing budget.
The Statistical Trap Hiding in Every Marketing Report
There's a name for what happened to that SaaS company. It's called Simpson's Paradox: when a trend that appears in aggregated data reverses or disappears when you break the data into meaningful groups.
Sam Tomlinson, who advises some of the most sophisticated digital advertisers in the world, calls this one of the most dangerous paradoxes in marketing. And it shows up constantly.Here's another example from programmatic advertising research. A campaign showed that female audiences had 50% lower click-through rates than males overall. The obvious conclusion: shift budget toward male audiences. But when AdExchanger broke the data by age group, females aged 18-24 actually outperformed males in the same bracket (1.88% CTR vs 1.85%). The aggregate gap was caused by age distribution differences between genders, not by gender preference.
The business would have cut spend on its highest-performing audience segment. Because the blended number pointed them in the wrong direction.
This isn't rare. It's the default. Every time you look at a single number that blends multiple populations, you're at risk of making a decision that's the exact opposite of what the data actually supports.
| What the blended metric says | What the segmented data reveals |
|---|---|
| Conversion rate dropped from 3.4% to 2.9% | Desktop CVR up 30%. Android bug crushed mobile CVR to 1%. |
| Female audiences convert 50% worse | Females 18-24 outperform males 18-24. Age distribution created the gap. |
| Overall ROAS is 4.2x across all products | High-margin products returning 8x. Low-margin products returning 1.8x. Budget flowing to the wrong ones. |
| Average cost per lead is $85 | Brand search CPL is $12. Non-brand CPL is $140. Brand search is inflating the "performance." |
| Campaign CPA improved 15% this month | Remarketing drove the improvement. Prospecting CPA actually worsened 20%. |
That last row is particularly common. We've written before about why your marketing dashboard can lie to you. Simpson's Paradox is the mechanism behind many of those lies.
Smart Bidding Makes This Worse, Not Better
Here's the part that should worry you.
78% of Google Ads spend now runs through automated Smart Bidding strategies, up from 68% just two years ago. The algorithm optimises toward a portfolio-level target. That's a blended target by definition.When you tell Google "get me conversions at $80 CPA across this campaign," the system will happily hit that number. But it might do it by spending heavily on cheap branded searches at $15 each while letting non-brand prospecting blow out to $160. The average is $80. The average is also meaningless.
Sam Tomlinson's audit framework puts it bluntly: "Good blended numbers often mask what's driving the underlying performance." He sees accounts regularly where performance looks healthy on the surface because remarketing, brand search, or existing customer purchases prop up the aggregate. Strip those out and the prospecting engine (the part that actually grows the business) is underwater.The AI isn't lying to you. It's doing exactly what you asked. The problem is that the average target you gave it doesn't represent any real segment of your business.
This connects directly to how Smart Bidding can hurt more than it helps when the targets aren't set with segmented understanding.
Your "Average Customer" Is a Fiction
The same problem applies to how businesses think about their customers.
Byron Sharp and the Ehrenberg-Bass Institute have spent 40 years studying how brands actually grow, across 130+ brands in 13+ product categories. One of their most important findings: your customer base is not what you think it is.
Most business owners picture their "typical" customer as someone who buys regularly, knows the brand well, and comes back often. That person exists. But they're a small minority.
The actual breakdown, validated by Bain & Company across 4,000 brands:
| Customer segment | % of customer base | % of current revenue | % of revenue next year |
|---|---|---|---|
| Heavy buyers (top 20%) | ~20% | ~60% | ~45% |
| Medium buyers | ~30% | ~30% | ~30% |
| Light and non-buyers | ~50% | ~10% | ~25% |
Notice what happens over time. Your heavy buyers contribute 60% of revenue this year, but only about 45% next year. That's Sharp's Law of Buyer Moderation: heavy buyers naturally buy less over time, and light buyers naturally buy more. The customer base is fluid, not fixed.
The old "80/20 rule" (80% of revenue from 20% of customers) is a myth. The real ratio is closer to 60/20. And that 60% shrinks every year.
What does this mean practically? When you optimise your marketing for the "average customer" (or worse, for your best customers), you're aiming at a target that barely exists and is actively shrinking. The loyalty trap is real: growth comes overwhelmingly from reaching more light buyers and non-buyers, not from squeezing more from your regulars.
The Real Cost of Averaging
Let's make this concrete for a typical Australian SME.
Imagine a plumber running $3,000/month in Google Ads. The monthly report says: 40 leads, $75 CPL, 4.2% conversion rate. Looks fine.
But segment it:
- Emergency plumbing keywords: 15 leads at $45 CPL, 62% close rate, average job $450
- Renovation/bathroom keywords: 10 leads at $95 CPL, 15% close rate, average job $8,500
- General plumbing keywords: 15 leads at $85 CPL, 28% close rate, average job $280
The right strategy might be to double down on emergency and renovation while cutting general. But the blended report makes all three look roughly the same.
This is what knowing what's working without knowing why looks like in practice. The aggregate number can be accurate and completely useless at the same time.
The Paradox of Adding More
There's a related trap Sam Tomlinson documents called Braess's Paradox: adding capacity to a system can make overall performance worse. He cites a DTC apparel brand that expanded from two advertising channels to five. More channels, more reach. Obvious win.
Instead, CPM rose 15% and customer acquisition cost rose 24%.
Why? The new channels attracted overlapping audiences. The same people saw the same brand across five platforms instead of two. Frequency went up but reach didn't. The "more is better" instinct, which feels right when you're looking at aggregate channel counts, produced the opposite result.
Avinash Kaushik's response to this kind of problem is structural. He argues that businesses need to separate their metrics into three distinct layers:
- KPIs (no more than 4): Directly impact the bottom line. Revenue, profit, market share, customer acquisition.
- Diagnostic metrics (4-6 per KPI): Explain why KPIs move. Conversion rate by segment, CPL by intent type, close rate by lead source.
- Influencing variables (5-25 per diagnostic): Explain why diagnostics move. Page speed by device, ad relevance score, keyword match type distribution.
What to Actually Do About This
Stop looking at blended numbers without immediately asking: "What segments make up this average?"
For Google Ads specifically:Break every metric by at least these dimensions before making decisions:
- Brand vs non-brand search. Your branded traffic is demand you already created. Non-brand is new demand capture. Blending them makes everything look better than it is.
- Device type. Mobile and desktop users behave fundamentally differently. A "3% conversion rate" might be 5% desktop and 1.5% mobile.
- Campaign type. Search, Performance Max, Display, and YouTube serve different purposes. Averaging their ROAS together is like averaging a sprinter's time with a marathon runner's.
- New vs returning visitors. A campaign that looks efficient might just be remarketing to people who were already going to buy.
- Segment leads by source AND intent type, not just source.
- Track close rate and job value per segment, not just per month.
- Watch for the Buyer Moderation effect: if your revenue is growing but your customer count isn't, you're extracting more from existing buyers instead of growing. That trend reverses.
- Put no more than 4 KPIs on the first page. Everything else goes in an appendix.
- Never present a blended metric without the segments underneath it.
- When a number changes, ask "which segment moved?" before asking "what do we do?"
What This Means for Your Business
Every marketing decision you make is based on a model of your customer. If that model is an average, it represents nobody. You'll optimise for a person who doesn't exist, measure success against a benchmark that's meaningless, and make budget decisions that actively work against your growth.
The fix isn't more data. It's disaggregated data. Break the average. Look at what's underneath. Then decide.
Your best campaign might be the one your blended report says is underperforming. Your worst might be the one propping up the average. You won't know until you stop trusting the fiction of the average customer.
Further Reading
- Sam Tomlinson: Marketing Paradoxes - Simpson's Paradox, Braess's Paradox, and Goodhart's Law applied to digital marketing, with specific case studies
- Byron Sharp: How Brands Grow - The Ehrenberg-Bass Institute's research on light buyers, penetration, and the Law of Buyer Moderation
- Avinash Kaushik: KPIs, Metrics, and Influencing Variables - The three-layer metric hierarchy that stops data puking
- AdExchanger: Simpson's Paradox in Programmatic Buying - How aggregate audience data can point you toward the wrong demographic
Dream Outcome is an Australian digital marketing agency helping SMEs grow through Google Ads, Facebook Ads, and Email Marketing.