Testing Google Ads Changes Without Killing Lead Flow

Someone told you to A/B test your Google Ads. So you built a campaign experiment, split it 50/50, waited three weeks, and the results panel still says no clear winner. Nothing is broken. Your account is simply too small for the tool you picked.

This post is about testing inside the ad account. If you want to test the page the ads land on, that is a different problem with different math — we cover it in A/B testing on low-traffic service-business sites.

Can a low-volume account run a valid Google Ads experiment?

Only if the campaign produces enough conversions to fill both arms, and most do not. Google's Experiments FAQs say to run an experiment for at least 4 to 6 weeks, to wait 1 to 2 conversion cycles, and that the first 7 days of data are discarded to account for ramp-up. Google's own advice for getting significant results is blunt: pick campaigns with high volumes and run them longer.

Do the arithmetic on your own account before you build anything. A campaign at 15 conversions a month, split 50/50, running six weeks with the first week thrown out, gives each arm roughly 11 conversions. Eleven versus thirteen is not a result. It is noise with a decimal point.

Google tells you this in the interface, quietly. In Monitor your experiments, a statistically significant result is marked with a blue asterisk. In a 15-conversion campaign that asterisk never shows up. Google even lists the reasons a result is not significant: the experiment has not run long enough, the campaign does not get enough traffic, or the split was too small.

How many conversions a month do you need before a test can conclude?

We use 50 conversions a month in the single campaign under test as the floor for even attempting a 50/50 experiment. That is not a Google number — it is the number that survives Google's own runtime rules. At 50/month, a six-week experiment with the first week discarded leaves each arm about 37 conversions, which is enough to see a large effect and still not enough to see a small one.

Note the word single. Account-wide conversions do not count. If you run four campaigns and 60 conversions a month, no individual campaign clears the floor.

Most single-location service businesses on a $2,000-$5,000 monthly ad budget live in the top two rows. That is not a failure. It just means the split-test tool is the wrong tool, and the agency that keeps showing you experiment screenshots is showing you theater.

What does a 50/50 split actually cost you while it runs?

It sends half your traffic to a version you are not confident in, for six weeks. Google's custom experiment setup docs recommend a 50% split because it gives 'the best comparison between the original and experiment campaigns' — and the same page notes you can schedule up to five experiments for a campaign but run only one at a time. There is no cheap way to hedge.

Two more clauses matter and almost nobody reads them. Google's Experiments FAQs state that experiments ended manually are never applied to the original campaign — only ones that run to their scheduled end date. And if you pick a split that is not 50/50, Google scales the reported numbers to make them comparable, so the raw counts you see are not the raw counts you got.

So the real cost of a test in a small account is six weeks of half-speed lead flow on a guess, ending in a grey no-clear-winner. If your business needs those leads to make payroll, that is a bad trade. This is exactly the kind of thing we argue about before touching an account in our paid ads program.

When is a disciplined before/after better than an experiment?

Below roughly 50 conversions a month, always. A before/after uses 100% of your traffic on one version at a time, so you accumulate data twice as fast as a 50/50 split — the tradeoff is that you cannot control for anything that changed in the world between the two windows. Your job is to make that tradeoff small on purpose.

• Change exactly one thing. Bids, budget, copy and match types in the same week means you learn nothing.
• Freeze everything else for the full window — including the landing page, the offer, and your call-handling.
• Use matched windows of equal length: 6 weeks before, 6 weeks after. Not 3 versus 8.
• Run a seasonality check: pull the same two windows from last year. If the year-over-year shape already moves, the change did not do it.
• Log the change date in writing, in a shared doc, with what you changed and why.
• Judge on cost per booked job, not cost per click or per form fill.

The seasonality check is the step everyone skips. An HVAC account that swaps ad copy on June 1 and sees cost per lead drop 30% did not win a copy test — it won summer. If your demand curve swings hard by season, read the trap in detail in Google Ads seasonality adjustments for seasonal trades before you claim any before/after win.

How long should a test run given your conversion lag?

Four to six weeks minimum, plus your conversion lag on top — Google says to run longer if you have a long conversion delay and to wait 1 to 2 conversion cycles. Conversion lag is the gap between the click and the conversion, and it differs wildly by trade.

Pull your own lag from the Google Ads time-lag report instead of guessing. Then add it to the six weeks. A law firm that reads results at week four is reading a window in which half the conversions from the last two weeks have not landed yet — and the arm that ran second always looks worse.

What ruins most ad tests before they start?

Changing more than one variable in the same window — the single most common way small accounts destroy their own data. Google's setup documentation warns that making changes to either the original campaign or the experiment while it runs may make it harder to interpret your results, and its FAQs say running several experiments at once is not recommended because they interfere with each other.

• Testing bids, budget and ad copy in the same week, then crediting the win to whichever one you liked.
• Editing the original campaign mid-experiment. Those edits do not carry into the trial arm, so the two arms quietly stop being comparable.
• Stopping the moment the numbers look good. Ended early means Google never applies the result, and you cashed in a coin flip.
• Running two experiments at once across the same account and search terms.
• Testing something too small to matter. A new headline will not fix a campaign bleeding money on the wrong match types.
• Judging on conversions when your conversion action is a page view, not a booked job.

That last one is the quiet killer. If your conversion tracking counts every thank-you page load, you can win a test and lose money. Get the measurement right before you get clever with the testing.

What should you test first in a service-business account?

Structure and targeting before creative — because they move cost per lead by large multiples, and copy moves it by percentage points. Big levers survive small samples. Small levers do not.

The honest verdict: for accounts under 50 conversions a month, negative keywords and geography beat every clever creative test you could run, and they need no split at all. Copy testing is real, but it belongs in ad variations — a cookie-split tool built for exactly this — which we get into in the responsive search ads guide.

None of this is a reason to stop testing. It is a reason to stop pretending. Anyone selling you statistically significant results out of a 15-conversion campaign is selling a story. If you want a second opinion on what your account is actually big enough to prove, get my free audit and we will tell you which of your last three tests proved nothing.