Skip to Content

The author

Steve Baker

Chief Analyst

As a PPC analyst, I often feel myself being pulled in two opposite directions. The problem is that when you group your keywords into Ad Groups, you are trying to achieve two things, and often, these are in competition. On one hand, you want to split your keywords into as many Ad Groups as possible. Every keyword potentially has a different conversion rate, and people searching for it may respond to different adverts. On the other hand, unless you have enough data to determine which advert people respond to in an Ad Group, and what the conversion rate is for that Ad Group, you can’t select the more effective adverts, or bid according to the value of the clicks. So there is one force pushing you to split your keywords into as many Ad Groups as possible, and another force pushing you to combine them. Often, when we take over an account from another agency, or one that has been managed in-house, we find Ad Groups with very similar keywords in, and very low traffic volumes. There’s no way to manage these Ad Groups effectively, and there’s no historical data that shows that the performance is different, or that different adverts work for the different keywords. The best solution to this balancing act is, perhaps, the most obvious one. In general (and there are exceptions to this) very similar keywords perform very similarly. They are searched for by very similar people, who respond to very similar adverts. So group your keywords tightly around themes. It can be very helpful if you use a consistent naming convention in your Ad Group names, so that you know exactly what each Ad Group does and does not include. This also makes it easier to write negative keyword lists to ensure that your Ad Groups don’t overlap, but that’s another story… Here’s an example. You are selling toasters – what Ad Groups do you need? A sensible Ad Group structure may look something like this: <em>Generic – 2 Slice, Generic – 4 Slice, Generic - Generic, Hinari – 2 Slice, Hinari – 4 Slice, Hinari – Generic, Tefal – 2 Slice, Tefal – 4 Slice, Tefal – Generic,</em> Etc. Immediately from this, you can see which brands you are advertising, and which adjectives you are using to split your campaign up. In the above example, you may want other Ad Groups for Compare, Cheap, Designer etc – I’ve just given a simple example. You can also see how easy it is to stop your Ad Groups overlapping – simply put ‘Hinari’ and ‘Tefal’ in as negatives in any Ad Group starting ‘Generic’, and ‘2 Slice’ and ‘4 Slice’ as negatives in the Ad Groups ending with ‘Generic’, and you can be sure that every impression is being picked up by the right Ad Group (and hence, bid and advert). The keywords in the above Ad Group are fairly predictable – though I’d strongly advocate using Exact, Phrase <strong>and</strong> Modified Broad Match (more about this in a moment). For ‘Hinari – 2 Slice’, your keywords would probably be something like <em>Hinari 2 slice toaster, Hinari 2 slice toasters, Hinari two slice toaster, Hinari two slice toasters, 2 slice Hinari toaster, 2 slice Hinari toasters, two slice Hinari toaster, two slice Hinari toasters.</em> Yes – I know that on Modified Broad Match, some of these are the same, but I find it’s the easiest way to ensure that you haven’t missed anything. It’s possible that the performance of the keywords above won’t be the same, or that they will perform differently on different match types. In these cases, you should be able to see this from the data – the decision to split your keywords into different Ad Groups should be based on performance data or based on tailoring the adverts for relevance, not based on a very old theory to maximise your Quality Score. <strong><em>“Erm… What?”</em></strong> A long time ago, before Google explained how the Quality Score worked, a man named Perry Marshall came up with a method for optimising your account that is still widely used now. If you find a keyword that ‘performs well’ in terms of click through rate, you move it into a separate campaign. The idea is that the new campaign will generate a very good campaign Quality Score, and that as a result, you’ll be rewarded with cheaper clicks. And that’s true, up to a point. But there are a few tiny little problems. Firstly, how do you decide which keywords are performing well? Your keywords’ Quality Score isn’t determined by your click through rate, but your click through rate normalised by position. You should get a better click through rate in a higher position, so simply because a keyword has a high click through rate, this doesn’t mean it’s going to have a higher Quality Score. Even if you do identify keywords that are exceptional in terms of their Quality Score, there’s still a big problem. If you remove keywords with very high click through rates from a campaign, its performance may improve, but what happens to the ones that are left behind? The click through rate of the rest of the campaign will fall just as sharply, and the other keywords will disappear down the page as fast as your ‘good’ keyword will move up the page. Is it worth it? Well – if one keyword generates 90 per cent of your clicks, improving its position at the cost of the other keywords seems reasonable – but this is exactly the situation where the impact will be the smallest, since the old campaign will have almost the same click through rate as the new one. In order for the new campaign to have a substantially higher click through rate than the old one, you need to be leaving a lot of traffic in the old campaign – and then the impact on them is as serious as the impact on the ‘good’ keyword. Of course, anyone looking at the performance of the keyword they migrate to the new campaign will be giddy with delight as they see the advert move up the page. But did any of them even check the impact on the ones that got left behind? If you’ve been paying attention, you’ll see another problem. By taking a ‘good’ keyword away from all the other keywords that are very similar, you’re taking away your ability to manage those keywords. You’ll have less data, and hence a less accurate picture of the conversion rate, so your bid adjustments will be less accurate, and this <strong>has </strong>to cost you money. In Perry’s defence, when he invented the Peel-And-Stick method, less was known about how the Quality Score worked – and, to be fair, the algorithm was probably much simpler than it is now. But there is no doubt that sacrificing tangible benefits like tailoring your bids to reflect the value of the clicks, and being able to test adverts quickly and effectively certainly outweighs the intangible benefits of improving your Quality Score. Don’t get me wrong – Quality Score is important – but it shouldn’t be the basis of your optimisation strategy. The perfect Adwords campaign will get only potential buyers (no non-potential buyers) onto your website, with a cost per click that balances volume with profitability to maximise profit. Clearly, Quality Score is relevant here, as high Quality Scores result in your advert appearing in higher positions for the same cost per click, so your ‘sweet spot’ moves up the search results. I’ve drifted quite a bit from my original point, that you need to find a balance between grouping your keywords together so that you can manage their bids together, and separating them so that you can tailor the adverts. Clearly, in the earlier structure, writing adverts that are relevant (and help your Quality Score) is easy. If you’re advertising Hinari 2-slice toasters, an advert with the headline <strong>‘Hinari 2-Slice Toasters</strong>’ is likely to perform well. It’s also going to be similar to your keywords, so Google will give you a Quality Score kick. In terms of the bidding, you’re probably wondering why I haven’t mentioned keyword-level bidding. I would <strong>never, ever</strong> advocate keyword-level bidding. It’s completely unnecessary and irrational. If two keywords need different bids, then the people searching on them must be different – so why assume that they’ll respond to the same advert? This is the basis of my peel-and-stick method – it’s based on performance, not pie-in-the-sky Quality Score hacks. If a keyword is performing differently to the others in the Ad Group, move it to its own Ad Group (along with any related terms that don’t have enough data). For example, it’s possible that people searching for <em>2 slice toasters</em> convert differently to people searching for <em>two slice toasters</em> – they may have totally different demographics. If so, you’ll probably see something similar on all of the 2-slice (and 4-slice) Ad Groups – so you should split them. If it’s just happening on one keyword, it’s likely to just be random variation. In general, you should be able to come up with a rational explanation for why keywords should perform differently – it’s easy to over-react to random variation, and once you do, it’s hard to put things back together again. I’ll end this blog with one final thought. Many bid management tools make bid adjustments at keyword-level. The arguments are that they can make changes far more rapidly, and across as many keywords as you like. By now, you should see why these are very silly sales points. Firstly, you are limited on how quickly you can adjust your bids by the data. If you are only getting one sale per day, you can only assess your conversion rate every month or so – so being able to make a million changes a day is pointless – what would they be based on? I appraised an account recently, and one Ad Group had picked up 4 clicks in the previous 30 days. There had been over 50 automated bid adjustments in the same period! Secondly, if you bid at keyword level, what percentage of your keywords will ever have enough clicks for you to make bid adjustments? The Pareto Principle (the 80:20 rule) applies to keywords as much as anything, and if you’ve got a long tail of keywords with keyword level bidding, you’re leaving a lot of profit on the table due to inaccurate bid adjustments. To return to my original point, if you manage an Adwords account, you are subject to the same opposing forces that I started with. The approach outlined above is my solution to it, and it works well. It’s probably not the only way to work, and you may have a totally different one that you’re comfortable with. The critical point to take away is that these forces exist, and that you should bear them in mind when you contemplate your account structure.

As a PPC analyst, I often feel myself being pulled in two opposite directions. The problem is that when you group your keywords into Ad Groups, you are trying to achieve two things, and often, these are in competition.

On one hand, you want to split your keywords into as many Ad Groups as possible. Every keyword potentially has a different conversion rate, and people searching for it may respond to different adverts.

On the other hand, unless you have enough data to determine which advert people respond to in an Ad Group, and what the conversion rate is for that Ad Group, you can’t select the more effective adverts, or bid according to the value of the clicks.

So there is one force pushing you to split your keywords into as many Ad Groups as possible, and another force pushing you to combine them. Often, when we take over an account from another agency, or one that has been managed in-house, we find Ad Groups with very similar keywords in, and very low traffic volumes.

There’s no way to manage these Ad Groups effectively, and there’s no historical data that shows that the performance is different, or that different adverts work for the different keywords.

The best solution to this balancing act is, perhaps, the most obvious one. In general (and there are exceptions to this) very similar keywords perform very similarly. They are searched for by very similar people, who respond to very similar adverts. So group your keywords tightly around themes.

It can be very helpful if you use a consistent naming convention in your Ad Group names, so that you know exactly what each Ad Group does and does not include. This also makes it easier to write negative keyword lists to ensure that your Ad Groups don’t overlap, but that’s another story…

Here’s an example. You are selling toasters – what Ad Groups do you need? A sensible Ad Group structure may look something like this:

Generic – 2 Slice, Generic – 4 Slice, Generic - Generic, Hinari – 2 Slice, Hinari – 4 Slice, Hinari – Generic, Tefal – 2 Slice, Tefal – 4 Slice, Tefal – Generic, Etc.

Immediately from this, you can see which brands you are advertising, and which adjectives you are using to split your campaign up. In the above example, you may want other Ad Groups for Compare, Cheap, Designer etc – I’ve just given a simple example.

You can also see how easy it is to stop your Ad Groups overlapping – simply put ‘Hinari’ and ‘Tefal’ in as negatives in any Ad Group starting ‘Generic’, and ‘2 Slice’ and ‘4 Slice’ as negatives in the Ad Groups ending with ‘Generic’, and you can be sure that every impression is being picked up by the right Ad Group (and hence, bid and advert).

The keywords in the above Ad Group are fairly predictable – though I’d strongly advocate using Exact, Phrase and Modified Broad Match (more about this in a moment).

For ‘Hinari – 2 Slice’, your keywords would probably be something like Hinari 2 slice toaster, Hinari 2 slice toasters, Hinari two slice toaster, Hinari two slice toasters, 2 slice Hinari toaster, 2 slice Hinari toasters, two slice Hinari toaster, two slice Hinari toasters.

Yes – I know that on Modified Broad Match, some of these are the same, but I find it’s the easiest way to ensure that you haven’t missed anything. It’s possible that the performance of the keywords above won’t be the same, or that they will perform differently on different match types.

In these cases, you should be able to see this from the data – the decision to split your keywords into different Ad Groups should be based on performance data or based on tailoring the adverts for relevance, not based on a very old theory to maximise your Quality Score.

“Erm… What?”

A long time ago, before Google explained how the Quality Score worked, a man named Perry Marshall came up with a method for optimising your account that is still widely used now. If you find a keyword that ‘performs well’ in terms of click through rate, you move it into a separate campaign.

The idea is that the new campaign will generate a very good campaign Quality Score, and that as a result, you’ll be rewarded with cheaper clicks. And that’s true, up to a point. But there are a few tiny little problems. Firstly, how do you decide which keywords are performing well? Your keywords’ Quality Score isn’t determined by your click through rate, but your click through rate normalised by position.

You should get a better click through rate in a higher position, so simply because a keyword has a high click through rate, this doesn’t mean it’s going to have a higher Quality Score. Even if you do identify keywords that are exceptional in terms of their Quality Score, there’s still a big problem.

If you remove keywords with very high click through rates from a campaign, its performance may improve, but what happens to the ones that are left behind? The click through rate of the rest of the campaign will fall just as sharply, and the other keywords will disappear down the page as fast as your ‘good’ keyword will move up the page.

Is it worth it? Well – if one keyword generates 90 per cent of your clicks, improving its position at the cost of the other keywords seems reasonable – but this is exactly the situation where the impact will be the smallest, since the old campaign will have almost the same click through rate as the new one.

In order for the new campaign to have a substantially higher click through rate than the old one, you need to be leaving a lot of traffic in the old campaign – and then the impact on them is as serious as the impact on the ‘good’ keyword.

Of course, anyone looking at the performance of the keyword they migrate to the new campaign will be giddy with delight as they see the advert move up the page. But did any of them even check the impact on the ones that got left behind? If you’ve been paying attention, you’ll see another problem.

By taking a ‘good’ keyword away from all the other keywords that are very similar, you’re taking away your ability to manage those keywords. You’ll have less data, and hence a less accurate picture of the conversion rate, so your bid adjustments will be less accurate, and this has to cost you money.

In Perry’s defence, when he invented the Peel-And-Stick method, less was known about how the Quality Score worked – and, to be fair, the algorithm was probably much simpler than it is now. But there is no doubt that sacrificing tangible benefits like tailoring your bids to reflect the value of the clicks, and being able to test adverts quickly and effectively certainly outweighs the intangible benefits of improving your Quality Score.

Don’t get me wrong – Quality Score is important – but it shouldn’t be the basis of your optimisation strategy. The perfect Adwords campaign will get only potential buyers (no non-potential buyers) onto your website, with a cost per click that balances volume with profitability to maximise profit. Clearly,

Quality Score is relevant here, as high Quality Scores result in your advert appearing in higher positions for the same cost per click, so your ‘sweet spot’ moves up the search results. I’ve drifted quite a bit from my original point, that you need to find a balance between grouping your keywords together so that you can manage their bids together, and separating them so that you can tailor the adverts. Clearly, in the earlier structure, writing adverts that are relevant (and help your Quality Score) is easy. If you’re advertising Hinari 2-slice toasters, an advert with the headline ‘Hinari 2-Slice Toasters’ is likely to perform well. It’s also going to be similar to your keywords, so Google will give you a Quality Score kick. In terms of the bidding, you’re probably wondering why I haven’t mentioned keyword-level bidding. I would never, ever advocate keyword-level bidding. It’s completely unnecessary and irrational. If two keywords need different bids, then the people searching on them must be different – so why assume that they’ll respond to the same advert? This is the basis of my peel-and-stick method – it’s based on performance, not pie-in-the-sky Quality Score hacks. If a keyword is performing differently to the others in the Ad Group, move it to its own Ad Group (along with any related terms that don’t have enough data). For example, it’s possible that people searching for 2 slice toasters convert differently to people searching for two slice toasters – they may have totally different demographics. If so, you’ll probably see something similar on all of the 2-slice (and 4-slice) Ad Groups – so you should split them. If it’s just happening on one keyword, it’s likely to just be random variation. In general, you should be able to come up with a rational explanation for why keywords should perform differently – it’s easy to over-react to random variation, and once you do, it’s hard to put things back together again. I’ll end this blog with one final thought. Many bid management tools make bid adjustments at keyword-level. The arguments are that they can make changes far more rapidly, and across as many keywords as you like. By now, you should see why these are very silly sales points. Firstly, you are limited on how quickly you can adjust your bids by the data. If you are only getting one sale per day, you can only assess your conversion rate every month or so – so being able to make a million changes a day is pointless – what would they be based on? I appraised an account recently, and one Ad Group had picked up 4 clicks in the previous 30 days. There had been over 50 automated bid adjustments in the same period! Secondly, if you bid at keyword level, what percentage of your keywords will ever have enough clicks for you to make bid adjustments? The Pareto Principle (the 80:20 rule) applies to keywords as much as anything, and if you’ve got a long tail of keywords with keyword level bidding, you’re leaving a lot of profit on the table due to inaccurate bid adjustments. To return to my original point, if you manage an Adwords account, you are subject to the same opposing forces that I started with. The approach outlined above is my solution to it, and it works well. It’s probably not the only way to work, and you may have a totally different one that you’re comfortable with. The critical point to take away is that these forces exist, and that you should bear them in mind when you contemplate your account structure.