← Back to journal

Your Marketing Results Moved. That Doesn't Mean Anything Happened.

Dream Outcome · JournalFig. YOUR-M

Your Marketing Results Moved. That Doesn't Mean Anything Happened.

Your cost per lead spiked 38% last week. You fire off an email to your agency. Something must have changed. Can we pause that campaign? Should we rewrite the ads? Let's try different keywords.

Your agency makes the changes. The following week, CPL drops back down. Everyone congratulates themselves on the quick response.

Except nothing was wrong in the first place. The spike was random variation. And the "fix" you just applied reset your bidding algorithm's learning, introduced new variables nobody can isolate, and guaranteed that the next three weeks of data will be unreadable.

This happens in almost every small business ad account, every month. It's the most expensive habit in digital marketing, and almost nobody talks about it.

The Noise You're Swimming In

In 2021, Daniel Kahneman, Olivier Sibony, and Cass Sunstein published Noise: A Flaw in Human Judgment. The book makes a distinction most people miss: bias is a systematic error that pushes judgments in one direction. Noise is random scatter that pushes judgments in every direction. Both cause bad decisions, but noise is harder to see because it doesn't have a pattern.

Here's the uncomfortable part: Kahneman's research suggests noise causes at least as much damage as bias in organisational decision-making. And in marketing, it's everywhere.

A service business spending $3,000 per month on Google Ads might generate 30 leads one month, 22 the next, then 35. That's not a trend. That's not a strategy problem. That's what happens when you have small sample sizes in a live auction environment where competitors change bids, users change behaviour, and Google's own algorithm experiments with delivery.

Research into paid media fluctuation confirms this directly: a week with 24 enquiries can be followed by a week with 9, even when spend hasn't moved. Search demand changes. Competitors alter their bids. Platforms shift delivery. The auction is live and chaotic. Some variation isn't just expected. It's guaranteed.

The problem isn't the noise. The problem is what you do about it.

Why Your Brain Insists on Finding a Cause

There's a famous story from Kahneman's earlier work, Thinking, Fast and Slow. He was teaching Israeli Air Force flight instructors about feedback and performance. An instructor objected: "When I praise a cadet for a great manoeuvre, they always do worse next time. When I yell at them for a bad one, they improve."

Kahneman called this one of the most satisfying eureka moments of his career. The instructor wasn't wrong about the pattern. He was wrong about the cause. A cadet who performed exceptionally well was likely benefiting from some luck. Their next performance would naturally regress toward their average ability, regardless of praise or criticism. And a cadet who performed terribly would naturally improve, regardless of being yelled at.

The instructors concluded that punishment works and praise doesn't. They were completely wrong. They'd built an entire feedback philosophy around regression to the mean - the statistical tendency for extreme results to drift back toward the average.

This is exactly what happens with marketing data.

Your best month wasn't entirely because of that new ad copy you tested. Your worst month wasn't entirely because of that keyword you paused. Some of it was signal. A lot of it was noise. And the natural tendency of extreme results is to regress back toward the baseline.

Rory Sutherland, in Alchemy, identifies a deeper problem: defensive decision-making. In organisations, people choose the most defensible option, not the most effective one. Doing nothing when results dip feels irresponsible. Making a change - any change - feels like leadership. "We identified the problem and took action" is a much easier story to tell than "we looked at the data, determined it was random variation, and deliberately chose to hold steady."

The irony is brutal. The defensible action (making changes) often causes real damage. The effective action (doing nothing) looks like negligence.

The Compounding Cost of Reacting to Randomness

Here's where this gets expensive for real businesses.

Google's Smart Bidding algorithms need data to learn. As of September 2026, Google reframed its learning period guidance around conversion volume: calibration takes roughly 50 conversion events or 1-2 conversion cycles. During this learning period, performance is intentionally volatile as the algorithm tests different bid levels to find the optimal range.

Now consider a typical SME Google Ads account. A plumber spending $2,500 per month might get 20 leads. A B2B company spending $5,000 might get 12. At those volumes, reaching 50 conversions takes two to four months of uninterrupted data.

But the account rarely goes uninterrupted. The business owner sees a bad week, panics, and asks for changes. Each significant change to bid strategy, targeting, or budget can push the campaign back into a learning phase. The algorithm partially resets. The clock starts again.

The result is a campaign that never finishes learning. It's perpetually in a state of adjustment, perpetually underperforming, and perpetually generating the kind of volatile data that triggers more panic and more changes.

ScenarioMonthly conversionsChanges per monthTime to 50 conversionsAlgorithm state
Calm account180-1 minor~3 monthsFully calibrated, optimising
Reactive account183-4 significantNever reachedPerpetual learning phase
Panicked account18Weekly changesNever reachedRandom performance, no learning

This is the cycle: noise causes panic, panic causes changes, changes cause more noise, more noise causes more panic. We've written before about how training your algorithm on the wrong answer teaches it perfectly. But training it on no answer - because you keep resetting before it finishes learning - might be worse.

A 2026 analysis across 21,000+ Google Ads accounts by Optmyzr found that mid-market accounts ($10K-$50K/month) delivered 565.8% ROAS, compared to 387.8% for accounts under $10K. Part of that gap is budget. But a significant part is that larger accounts generate enough conversion volume for Smart Bidding to actually calibrate, while smaller accounts are stuck in statistical purgatory, reacting to data that hasn't yet told them anything meaningful.

Your Best Month Was Partly Luck. So Was Your Worst.

Byron Sharp's research at the Ehrenberg-Bass Institute confirms this regression pattern at the customer level. His Law of Buyer Moderation shows that heavy buyers naturally trend lighter over time, and light buyers naturally trend heavier - without any change in marketing strategy.

The data is specific: this year's top 20% of buyers generate about 60% of current sales. Next year, the same individuals generate about 45%. Not because they stopped liking the brand. Because extreme behaviour naturally regresses toward the mean.

The same phenomenon operates on your marketing results. A month where everything aligned - search demand was high, competitors backed off, your ad happened to show for high-intent queries - will naturally be followed by a more average month. That's not a problem to diagnose. That's statistics doing what statistics do.

This connects directly to something we've explored before: having all the data doesn't mean you know what to do with it. A metric moving isn't the same as something happening.

The Patience Dividend

Les Binet and Peter Field's IPA effectiveness research across nearly 1,000 campaigns provides the strongest evidence for why patience pays.

Their core finding: short-term marketing effects decay rapidly, while long-term brand effects compound. Short-term activation campaigns spike response but almost never build pricing power, loyalty, or margin. The long-term baseline lift from consistent presence builds over years, not weeks.

The implication for data interpretation is profound. If you're evaluating marketing performance on a weekly or even monthly basis, you're mostly looking at noise. The real effects of your marketing investment - the ones that actually determine whether you grow - operate on a timescale that weekly dashboards can't capture.

Time horizonWhat you're mostly seeingAppropriate action
DailyAlmost pure noiseMonitor for technical issues only
WeeklyMostly noise, some seasonal patternsNote trends, don't react
MonthlyMix of noise and signalIdentify patterns, consider changes
QuarterlyMostly signalMake strategic decisions
AnnuallyTrue performance trajectoryEvaluate overall strategy

Sam Tomlinson's audit framework makes this explicit: effective marketing evaluation starts with financials and business model understanding, not campaign-level metrics. If your CPL moved 25% in a week but your quarterly trend is flat, you don't have a marketing problem. You have a data interpretation problem.

As Avinash Kaushik argues for 2026: obsess over methodology, not metrics. For every KPI, ask how exactly is it measured? Two teams can report the same number and be living in different realities. The methodology is the integrity layer, and for small businesses, the most important methodological question is: do I have enough data to know if this number means anything?

How to Tell Noise from Signal

This isn't about ignoring your data. It's about knowing when your data has actually told you something.

Signal looks like this: Noise looks like this: Before diagnosing a performance change, check whether you're just measuring randomness with false precision. A campaign generating 15 conversions per month has a standard deviation so wide that a 30% swing in CPL is statistically meaningless.

What This Means for Your Business

The most valuable skill in marketing isn't knowing when to act. It's knowing when not to.

Three practical rules:

1. Set a minimum data threshold before making changes. For Google Ads, that means at least 30-50 conversions or two full conversion cycles. If your account generates 15 leads per month, you need a minimum of two months of stable data before any performance assessment is valid. 2. Use rolling averages, not snapshots. Replace weekly CPL tracking with a 90-day rolling average. When someone asks "how are the ads going?", answer with the trend line, not yesterday's number. The panic-driven approach to measurement is what creates most of the unnecessary changes. 3. Distinguish monitoring from evaluating. Check your account daily for technical issues (ads disapproved, tracking broken, budget exhausted). Evaluate performance monthly at earliest, quarterly for strategic decisions. These are different activities requiring different mindsets.

The business that holds steady through two weeks of noise while their competitor thrashes around resetting algorithms and rewriting ads doesn't just save time. They let the algorithm do its job. They preserve their data integrity. And they compound the small advantage of consistency into something meaningful over six, twelve, eighteen months.

Your data moved. Before you move with it, ask yourself: is this something that happened, or is this just what randomness looks like?

Further Reading


Dream Outcome is an Australian digital marketing agency helping SMEs grow through Google Ads, Facebook Ads, and Email Marketing.
Book a free call

Ready to grow profitably?

Get a free digital marketing plan tailored to your business. No lock-in, no long commitments.

Book my free call  →