Conversion and ROAS noise: what’s a real change and what’s just chance
A client writes: “ROAS halved yesterday, what happened?” Most of the time, nothing. With 15–20 conversions a day and a different order value on each, ROAS swings several-fold on its own. Upload an export from your Google Ads account and I’ll show you, on your own numbers, which swings are noise and which are worth digging into.
How to export a CSV from Google Ads
- Google Ads → Reports → Report editor → new Table.
- Rows: Day. Columns: Cost, Conversions, Conv. value (or Conv. value / cost).
- Date range: ideally 6–12 months, at least 90 days.
- Download → CSV. Drop the file into the form above.
If the report is split by campaign, that’s fine — the tool sums rows for the same day. One requirement: the whole export must come from one account in one currency. The tool calculates cost per conversion (CPL) itself from cost and conversions.
What the tool shows
For each level — day, week, month — the tool shows the average, minimum, maximum, the gap between the best and worst period, and the typical deviation from the average. The longer the period, the less random swing: daily values are spread wide, monthly ones sit in a tight cluster.
Under each table there’s a one-sentence conclusion: from which level you can draw conclusions. For those who want to go deeper, a collapsed block shows how long to wait after a change to tell a real effect from noise.
How to read the result
- Max/min — how many times bigger the best period is than the worst. If daily data shows “5×”, a single “bad day” means nothing.
- Typical deviation — how far a period usually sits from the average.
- The conclusion under each table tells you whether to judge by day, week or month.
- Spread on one scale — at a glance: daily dots are spread wide, weekly narrower, monthly in a cluster.
The real spread in live accounts is even larger: there’s weekly seasonality and conversion lag.
FAQ
Why does daily ROAS jump around when I haven’t changed anything?
ROAS depends on two random things: how many conversions came in and what their order value was. At 10–20 conversions a day, one large order can double ROAS and its absence can cut it in half. That’s normal small-sample behaviour, not a signal.
How many conversions do I need for the analysis to make sense?
Rule of thumb: the random spread in conversion count is roughly ±2/√N. With 25 conversions per period that’s ±40%, with 100 it’s ±20%, with 400 it’s ±10%. ROAS is noisier still because order value varies too. For the exact figure for your account, see the “Advanced: how long to wait after a change” block.
Can I analyse by day if I get hundreds of conversions a day?
Yes, at high volume daily noise is much smaller. But weekly seasonality remains: Monday and Saturday behave differently. Compare a day with the same weekday, not with yesterday.
Why are the last few days in Google Ads always worse?
Google Ads records a conversion on the click date, not the purchase date. Someone who clicked today may buy next week. So the last 7–14 days are almost always under-reported. You can exclude them with the “Exclude the last” setting.
Where does my file go?
Nowhere. The calculation runs in your browser, the file isn’t uploaded to any server, and I never see it.
Other tools
Let's discuss
your project
Leave a request — we'll assess your situation, discuss your goals and find the best solution for your business and budget.