How much data before you decide

The question I get most is some version of “how many clicks before I pause this?”

You want enough data to be confident, but not so much that you’ve paid for a pile of clicks that were never going to sell. Waiting for real statistical certainty would take hundreds of clicks per decision, and you’d run out of budget first.

The answer also depends on what you’re about to do. Pausing a keyword or negating a search term is hard to undo, because once it stops getting traffic it stops collecting data. That deserves more evidence. Changing a bid by 10% is small and easy to reverse, so it needs less.

For bid changes, count orders

Bid changes use ACOS, so the data you need is orders. I want at least two orders before I move a bid, and three or four makes the ACOS a lot more reliable.

For pausing and negating, count clicks

Pausing and negating are for things spending money without orders, so you count clicks.

A simple way to set the number is to start from the lowest conversion rate you’d accept and double it. At a 10% conversion rate, it takes about 10 clicks to get an order on average. Double that and you get 20. So a search term, an exact keyword or a single product target with 20 clicks and no orders has had a fair shot. (For a version based on your actual conversion rate and how sure you want to be, use the calculator below. At 10% and 90% confidence it says 22.)

Broad targets need more. A phrase or broad keyword, an auto campaign target group or a category target spreads its clicks across lots of different searches, so 20 clicks might be one click each on 20 searches. I’d want 40 or more before pausing one of those, and often the better move is to negate the searches that aren’t working and keep the target.

Date ranges

When you don’t have enough data, use a longer range. If the last 7 days has two clicks, look at 30. If 30 isn’t enough, look at 60 or 90. For pausing and negating, look at 90 days or longer, so a target in a temporary slump doesn’t get cut.

When you do have plenty of data, use a shorter range, because recent data is more relevant. If a keyword had 20 orders last week, what it did a month ago doesn’t matter much. That means different ranges for different keywords, which is more work, and it’s also where a lot of the performance comes from.

Leave today and yesterday out of bid decisions. Amazon credits an order to the click for up to 7 days after it (14 for vendors and KDP authors) and reports it on the day of the click, so recent days always look worse than they’ll end up.