When Is 18 Clicks Enough to Cut a Google Ads Keyword?

Monday morning. I'm pulling data and I see one of my Search keywords sitting at 18 clicks, 0 conversions.

Instinct says cut it.

I didn't.

This "cut vs. wait" deadlock — eight years of running Google Ads for myself and I still hit it about once a week. In the early years I'd just go on gut feel. If a keyword looked ugly, I killed it. Then I started keeping a retro journal, and discovered something that made me wince: a meaningful chunk of the keywords I'd cut on gut feel would have converted if I'd waited a few more days. Translation: my intuition was killing keywords that were actually correct.

That's when I turned it into a math problem.

The First Math: Reverse from Conversion Rate

I run all of my cross-border DTC stores myself — no agency, no contractor, just me from the Feed all the way to the Bid layer. In that mode I have no one to ask. The only judge that doesn't lie to me is the data itself.

Here's how I reason:

If a store's average transaction conversion rate is 2% (100 clicks → 2 orders), then in theory I need to accumulate 50 clicks before I can even see 1 conversion.

But 1 conversion is enormous variance. To actually trust a verdict on whether a keyword "works or doesn't," I need at least 15 conversions.

Reverse it: 50 × 15 = 750 clicks.

That 750 isn't a magic number. It drifts with your store's conversion rate:

• 1% CR → at least 1,500 clicks before you've banked 15 conversions
• 5% CR → 300 clicks is enough
• 0.5% CR → 3,000 clicks before you have any standing to make claims

After eight years of testing this, I'll say it plainly: 15 is the minimum where you won't regret your decision. Below 15, every "difference" you see in CPA, ROAS, and CR is dice noise — not the keyword telling you anything real.

This "≥N conversions" idea is exactly what's hardcoded into ADM (an AI agent that optimizes Google Ads automatically) inside the Anomaly_Alert rule — except ADM uses 5. When a campaign's mobile CPA looks abnormally high but the period's total conversions are below 5, ADM silently skips the alert, doesn't bother the user. Same logic as my 15: without enough sample, every judgment is noise. I set ADM's threshold at 5 instead of 15 because alerts are soft hints — 5 is enough to catch trend-level signal, and the user gets to decide whether to wait for more data.

The Second Math: The High-CPC, Low-CR Disaster

But there's a scenario where waiting for 15 conversions costs more than it's worth.

Real example from my own account:

CPC: $2. Order conversion rate: 0.5%.

To accumulate 1 order conversion: 200 clicks = $400.

To accumulate 15 order conversions: 3,000 clicks = $6,000.

That's for one keyword.

My account has 1,500+ keywords. If I waited for "enough data" on each one at this rate, I wouldn't have the budget for the test. And around 95% of keywords end up cut anyway — meaning I'd be burning 95% of my budget on terms destined to die.

I made this exact mistake early on. I locked in a strict testing rule, and a new product's Search Ad Group ran for 6 weeks with no verdict. When my boss (who, again, is also me) asked about account health, I felt sheepish.

So I changed my approach: when CR is too low and CPC is too high, stop waiting on order conversions.

The Third Math: Micro-Conversions Cut Test Cost by 75%

Switch the "target conversion" from order to something earlier in the funnel — add to cart, sign-up, key page view.

Same example, but with a micro-conversion layer:

Micro-conversion: sign-up, not order.

Sign-up rate: 2% (industry average — an order of magnitude higher than order CR).

50 clicks = 1 sign-up. To bank 15 sign-ups: 750 clicks = $1,500.

Test cost dropped from $6,000 to $1,500 — 75% off.

A more aggressive play: use sign-ups as an early stop-loss signal —

Keyword CPC $2, sign-up rate 2%.

Spent $100 (50 clicks) and not a single sign-up?

The audience this keyword is pulling and your product are clearly mismatched. Don't wait until $400 burns through to look at order data — at $100 it should already be on the watch list.

I wired this up as a Python script. Every dawn it pulls yesterday's data and emails me the keywords that hit "spend > $50 and zero micro-conversions." Ten minutes with my coffee, I scan the list and decide which to pause that day.

This "spend / micro-conversion" hard ratio is exactly what ADM's Zombie_Cleanup rule does — automatically flagging high-spend, zero-conversion zombie objects and proposing to pause them. I put the same rule into ADM because I'd rather not run the dawn scan myself anymore.

The Fourth Math: When You Don't Need to Wait for Data

Not every "wait" is worth it. There are several situations where I'll bypass the 15-conversion threshold and decide right then:

1. Traffic Quality Is Obviously Off

Bounce rate 90%+, average dwell < 5 seconds, mobile traffic all coming from a country you weren't even targeting.

You don't need to look at conversions — the traffic itself is already telling you the answer. Cut.

2. Search Terms Are Wildly Off-Intent

I set Exact Match for "buy noise cancelling headphones," but in the search-terms report 6 out of the top 10 are "free music download."

That's Google's close-variant matching going feral — a configuration problem, not a data problem. Add negatives, don't wait for 15 conversions.

(By the way, ADM's Variant_Leak rule does exactly this — it monitors close-variant drift on Exact Match keywords, and when a variant's CPA is far above the keyword's average or has zero conversions, it auto-suggests adding the variant as an Exact Match negative. This rule alone saves me about 2 hours of search-terms grunt work every week.)

3. High-AOV Products with Long Attribution

I have one DTC store selling industrial gear — $1,500–$3,000 unit price, average conversion cycle 14–21 days.

By the "15 conversions" standard, plus the attribution-window lag, I wouldn't be able to "draw a conclusion" until 8–10 weeks later. By then the moment is long gone.

For products like this I switch from conversions to full-funnel signals: traffic precision (geo / device / search-term semantic), first-visit depth, return rate. Order count is just after-the-fact validation, not the decision input.

The Decision Threshold Isn't a Number — It's a Framework

By now you've probably noticed: I haven't given you a "standard answer."

Because there isn't one.

15 is a default, not a command. The actual framework looks like this (numbers below assume roughly CPC $2 / CR 2% — adjust them to your own):

What this framework is really answering is one question: does the cost I'm paying to make this judgment match the upside the judgment can produce?

If a keyword can earn me at most $500 in profit per year, I shouldn't spend $400 to test it. If a keyword could potentially crack open a new market, $2,000 of test budget is worth every penny.

That's what "decision threshold" actually means from a programmer's perspective — not a magic number, but treating every decision as a math problem.

After I Wrote It Into a Rule

That's basically the whole post. One side note —

After eight years of running this math by hand, I got tired of pulling the dawn Python script every day. So I started writing the rules into ADM, one by one — "alert only if conversions ≥5," "flag the high-spend zero-conversion zombies," "auto-suggest negatives when an Exact Match close-variant drifts." ADM runs them in the background on a cadence that depends on the subscription tier — daily on the higher tiers, every 3 days on the entry tier. My day-to-day work is now mostly approving what it surfaces — it tells me a keyword hit zombie threshold, I click Approve; it says a campaign's mobile CPA is 2× desktop with conversions ≥5, I take a look and decide whether to adjust the Bid Modifier.

If your decision threshold isn't the same as mine — say your industry's average CR is naturally 5%+ and you don't need to wait for 15 conversions — let me know. I want to see how sellers in different verticals set this number, and your feedback might shape the next time I tune Anomaly_Alert's default.