The essentials
- Volume traps: concluding on fewer than 100 conversions, reading a weekly rate on 12 events.
- Period traps: comparing a seasonal business with the previous month, judging a campaign before its conversion window closes.
- Calculation traps: averaging rates, adding up users, confusing correlation with cause.
- Measurement traps: reading attribution without knowing which model is used, ignoring the share of traffic lost to consent.
Marketing metrics are rarely wrong; it is the readings of them that are. A conversion rate that "falls 40%" over a week of 15 conversions, a campaign judged "loss-making" three days after launch, a channel called "useless" because the attribution model gives it nothing: the bad decisions come from these shortcuts. Here are the ten most frequent traps, each with a concrete example and the remedy. Most apply to any KPI, whatever the tool.
The volume traps
- Concluding on small numbers. A site going from 12 to 8 quote requests in a week shows "−33%"; statistically, that variation is noise. Below 100 conversions in the period, compare over a rolling 4 to 8 weeks and read the trend, not the variation.
- Reading rates on tiny bases. A 12% conversion rate on 25 visits from a campaign proves nothing; you need several hundred visits per variant to compare two ads or two pages. The testing method is in A/B testing in advertising.
| Monthly conversion volume | A reasonable reading granularity | A variation to treat as a signal |
|---|---|---|
| Under 30 | A rolling quarter | Above 40% over 3 months |
| 30 to 100 | Month, a 3-month average | Above 25% |
| 100 to 500 | Month, week with caution | Above 15% |
| More than 500 | Week | Above 10% |
The period traps
- Comparing a seasonal business with the previous month. A heating engineer sees enquiries fall 45% between January and April every year; comparing April with March concludes there is a problem that doesn't exist. Compare with the same period last year, and with the previous month only for stable businesses.
- Judging before the conversion window closes. A B2B campaign launched on the 1st is "loss-making" on the 10th; the meetings arrive between the 15th and the 45th. Set the reading window from the measured cycle (7 days in impulse e-commerce, 30 to 90 days in B2B) before concluding.
- Reading today's or yesterday's data. GA4 consolidates its data over 24 to 48 hours, Search Console runs 2 to 3 days behind, imported Ads conversions arrive with a lag. Reading the weekend's figures on Monday morning understates everything.
The calculation traps
- Averaging rates. Two campaigns, one at 5% conversion on 100 clicks, the other at 1% on 10,000 clicks: the "average" of 3% is wrong; the real rate is 1.04%. Always recalculate from the totals.
- Adding up users or reach. 3,000 users in January and 3,200 in February don't make 6,200 users over two months; some are the same people. The same goes for social reach across platforms. Only events (sessions, conversions, sales) add up.
- Confusing correlation with cause. Sales rise at the same time as your Instagram followers; it is often the season, or a parallel campaign, that explains both. Before attributing an effect, look for the period's other changes. The impact measurement method is detailed in measuring the impact of a communication.
The measurement traps
- Reading attribution without knowing the model. GA4 attributes by default using a data-driven model; Google Ads and Meta each count their own conversions, with different windows. Adding up the conversions the platforms report often gives 150 to 200% of the real total. Choose one source of truth (GA4 or the CRM) and read the platforms as relative indicators. See attribution in GA4.
- Forgetting the invisible share. In France, 25 to 45% of visitors refuse cookies; GA4 doesn't see them or models them. A "fall in traffic" after a change to the consent banner isn't a fall in business. Always reconcile the conversions measured with the enquiries and orders actually received.
| The wrong reading | What should have been looked at |
|---|---|
| "The bounce rate has exploded" | GA4 measures an engagement rate whose definition differs from the old bounce; check the session settings before concluding |
| "Direct traffic is rising, our brand awareness is growing" | Direct contains untagged links (emails, apps, QR codes); check the UTM naming |
| "Meta generates no sales according to GA4" | View-through conversions and cross-device journeys escape GA4; also read the before-and-after on brand searches |
| "The average position has fallen" | New queries in position 40 pulled the average down; look at the clicks and the positions on the target queries |
A reading routine that avoids these traps
Before commenting on a figure, ask four questions: how many events does it rest on? What is it compared with, and does that comparison make seasonal sense? Has the conversion window closed? Did the measurement change during the period (tracking, consent, attribution model)? If any answer is uncertain, the figure is noted but triggers no decision. The metrics with no decision value are listed in vanity KPIs, and the tool-specific traps in the GA4 traps to avoid.
How GreenRed helps
Rather than juggling several tools, GreenRed's Continuing training module brings these metrics together in a single dashboard, compares them over time and tells you which actions come first. You can try it free, with no card, from the Pricing.
Frequently asked questions
From how many conversions is a variation significant?
There is no single threshold, but in practice a variation of under 15% on fewer than 100 conversions can't be interpreted, and neither can a 30% variation on fewer than 30 conversions. Lengthen the period until you have accumulated several hundred events before comparing.
Why don't Google Ads and GA4 conversions match?
The two tools use different attribution models, conversion windows and counting methods. Google Ads claims a conversion if a click happened within 30 days, even if GA4 gives it to another source. A 20 to 40% gap is normal; beyond that, check the configuration.
Should you compare with the previous month or with last year?
With last year for any seasonal business, and with the previous month only for stable businesses or to track an action's immediate effect. Best of all is to show both comparisons, with the rolling 12-month trend to smooth the jolts.
How do you spot that a measurement change is distorting the figures?
Keep a log of changes: the consent banner, the tags, a redesign, a new attribution model, a change of tool. Any abrupt break in a metric coinciding with a log entry is a measurement break, not a change in the business. Without a log, those breaks are read as results.