A sudden unexplained drop in conversions is terrifying. You’re paying for clicks but you’re not getting leads or sales. It’s bad when you’re spending your boss’s or client’s money. It’s even worse when it’s your business on the chopping block.
Conversion rate is a two-sided problem. It’s a function of the traffic to your site and the website itself.
This essay focusses on a process for finding the root cause of the traffic half of the problem. I'm starting with the traffic half because it might be the problem even if you've changed the site.
I’ll be up front with you. The process takes time - especially the first time you do it. It might not give you one definitive answer. You might have to test a few theories.
It's worth doing because the techniques you master today will help you optimise for a better conversion rate in the future. The finest way to avoid a drop in conversions is to have a conversion rate that keeps going up.
We humans are hardwired to react without thinking. It’s what kept our ancestors alive. They jumped every time they saw something that looked like a sabre-toothed tiger in the bushes. It didn’t matter if they ran when it was Bambi in the shadows. All that mattered was that they got away from what their subconscious thought was danger.
Your ninja flight-or-fight reflexes don't do you any favours when it comes to fixing a drop in conversions.
One of three things will happen if you make panic changes to your Google Ads campaigns:
It won't fix the problem. You’ll have raised bids, adjusted budgets or changed bidding strategy and you’ll still be paying for clicks that don’t convert.
Things will get worse. You’ll spend more money on even fewer conversions. If it continues you’ll end up living in a cardboard box under a bridge. In the rain.
The problem will go away. That’s great but you won’t know if your changes caused the improvement or if it was coincidence. You’ll still feel that sick worry the next time conversions fall off a cliff.
Instead of reacting, the solution is to develop a testable hypothesis before you make any changes.
A testable hypothesis is a fancy term for "I think conversions dropped because more of our searches came out of business hours and our service is mainly B2B. So I’m going to change the ad schedule to show ads during working hours only and see if that improves things".
The hypothesis is what you think the reason might be. You’re not married to it, but it’s your best guess given the information you currently have: Conversions dropped because more of our ads showed after business hours.
It’s testable because you can change something - in this case the ad schedule - to see if it brings conversions up.
Let's see if we can get a testable hypothesis from the total number of clicks and conversions for a campaign in trouble.
Last month we got 25 187 clicks and 3 778 conversions. This month we got 24 653 clicks and 986 conversions. (I’m comparing months in this example but you can apply this technique to any period.)
Can we get a testable hypothesis from that?
Nope. The totals give us nothing to work with.
Diagnosing a drop in conversions by looking at the campaign totals is like trying to tell if you're sitting on a chair or a horse by counting the legs.
We need to dive deeper and examine the distribution of those clicks.
The clicks are distributed across many dimensions. They're distributed across the keywords in your campaign. They're distributed across the ads. They're distributed by device, network, demographic, location, time and intent.
The distribution of the 24 653 clicks we got this month is different to the distribution of the 25 187 clicks we got last month.
The people who clicked this month are different to the people who clicked last month. They’re older, younger, richer, poorer. They live in different areas. They clicked at different times. They started their searches with different search terms. They saw a different set of adverts. They searched on different devices.
The reason for a drop in conversions is often found in a change in the distribution of the clicks.
I’ll show you how to read the distribution of clicks across your keywords today. I'll write about the distribution of clicks across the other dimensions in future.
Pop your email in the box below and I’ll let you know when I publish the next essay.
We’ll start by considering only keywords that converted last month. Here’s how to see them in the Google Ads UI.
Now you’re comparing keywords that converted before with the same keywords in the period you’re worried about.
Now take a look at each keyword. Ignore it if the performance - impressions, clicks, cost and conversions - was roughly the same as last month. Otherwise we need to look at what changed so we can find out why it changed.
Let's assume that a keyword got fewer conversions this month than it did last month. There are two likely reasons:
The conversion rate was roughly the same as before but it had fewer clicks.
The conversion rate dropped but it had about the same number of clicks.
The process of getting to some testable hypotheses is the same for either case.
If we assume that this keyword had fewer clicks this month but the conversion rate was more or less the same, it could be for one of two reasons:
The keyword had fewer impressions but the clickthrough rate (CTR) was more or less the same.
The keyword had the about the same number of impressions but the clickthrough rate is lower.
Which is it? Has the keyword had fewer impressions or is it a CTR problem?
The difference is important. You can’t fix an impressions problem with the same tactics you’d use to fix a CTR problem.
But, it's still too early to think about tactics. We need to get to the root cause and we're not there yet.
If it’s an impressions problem - impressions are down but the CTR is more or less the same - it might be because:
If it’s a CTR problem there are a different set of possible causes:
Some of these possible causes can be ruled out by looking at the stats in your account.
You could compare the number of searches this month with the same month last year. If there was a big drop in searches that recovered afterwards it might suggest a seasonal dip.
You can see which ads in the ad group got the most impressions and clicks.
You can compare average position to see if your ads were shown in a worse position.
Ruling other possible causes out needs more effort. For instance, I don’t think you can see change in impression share by keyword without a little spreadsheet work.
Sherlock Holmes said
…when you have eliminated the impossible, whatever remains, however improbable, must be the truth…
It's the same here. The possible causes that aren't ruled out become the basis of your testable hypotheses.
Say for instance the data showed that we had a lower CTR rather than fewer impressions.
Let's pretend we ruled out a losing split test because we’ve only got one ad in the ad group. And we ruled out a relevance mismatch between the ad and the searches that triggered the ad by looking at the search terms for this keyword.
The data shows our ads appeared at position 3.4 this month but last month they were at position 1.1. This looks like a strong possibility for the reason for the lower CTR.
I say "looks like" because I don’t know if it is the root cause, but it's something we can test.
Here’s my testable hypothesis. "The ad for this previously high converting keyword has been shown lower on the page. I think this has led to a drop in CTR. The drop in CTR has lead to fewer clicks and thus fewer conversions.
I’ll test this by raising the bids for this keyword till the ad shows at an average of position 1.1. The CTR should increase. The increase in clicks should result in more conversions."
It’s a pretty trivial example but it illustrates the process.