• House was right.

    Everybody lies. Averages are no exceptions. Averages are icebergs. You see their tip, but have no idea what’s under the water.

    Unless you dive deeper.

    Pick a keyword. An average one. Nothing special. Look at its average CTR.

    • CTR: 5%
    • Quality Score: 7
    • Impressions: 4000
    • Clicks: 200

    Nothing to call home about, right? So how is that keyword doing? Good, bad, average?

    If you stay at this level, you’ll never know. So let’s deconstruct that keyword’s CTR.

    How does a keyword get its CTR? Someone types a search term into Google’s search box. A keyword is triggered. If that keyword makes it into the auction (its Ad Rank is high enough), the system pulls an ad from that keyword’s ad group, and shows it. That’s an impression. For the keyword, the ad, and the search term.

    Another search term is typed, and the same keyword gets triggered. If it makes it into the auction, the keyword gets another impression, just as the search term, and the ad.

    So actually, a keyword’s CTR is, in fact, the sum of clicks of all search terms that triggered it, divided by the sum of their impressions.

    Therefore that 5% CTR can in fact be composed of:

    •  120 clicks per 2000 impressions for S1 (search term 1), CTR 6%
    •  80 clicks per 2000 impressions for S2 (search term 2), CTR 4%

    Again, 4% and 6% are not too far apart, they both look pretty much the same.

    Let’s go one level deeper. Let’s say you have two ads in that ad group. How does a search term record an impression? All alone? Not really. An ad gets triggered by it. And if we put on a pair of our customers’ shoes, we realize that the decision to click an ad or to skip it is not related to the search term alone. It’s related to the search term / ad pair. If the ad answers the search query, if it is related to it, a click is recorded.

    So those 6% and 4% CTR for search terms S1 and S2 can, in fact, be, the sum of clicks for the S1A1, S1A2, S2A1 and S2A2 search term / ads pairs, and their impressions.

    They could be:

    •  110 clicks per 1000 impressions for S1A1, CTR 11%
    •  10 clicks per 1000 impressions for S1A2, CTR 1%
    •  10 clicks per 1000 impressions for S2A1, CTR 1%
    •  70 clicks per 1000 impressions for S2A2, CTR 7%
    Just as they are in the picture below.

    The average lie. Three levels of CTR.

    Amazing, right? Average at keyword level, still average at search term level, completely apart at search term + ad level. What works for one search term in terms of ads does not work for the other, and vice versa. Please note that if the same CTRs would have been split differently, you would have seen the bad performance either at search term level, or in your ad report. But when one search term works well with one ad and bad with the other, and the other way around, you can only see it when you look at the pairs.

    So should you start looking at all the search terms and all the ads, in all the ad groups through all your campaigns? No. That would be way too much. But for loose match keywords, more general keywords (if you have to use them), for keywords attracting many impressions and many different search terms, in ad groups where you have more than one ad, and where the ads you’re testing are quite different, it’s something worth looking at.

    All you have to do is to look for those keywords, click on them, then choose “See search terms – Selected”. Press “download” and add the Ad Id as a segment. You’ll end up with a table which will contain search terms, Ad Ids, CTR and other metrics.

    Sort by search term, and go through that table, and check for variations in CTR as the Ad Id changes for the same search term (or create a pivot table if you wish). Consider splitting that ad group into several ad groups, and make good use of negative keywords whenever you see that some search terms do great with some ads and bad with others.

    Make sure you are the master of your search terms, and that you always pair them with the right ad.

    We’ve now come to see that there are situations when even inside the same ad group, search terms can perform radically different, according to the ad they’re paired with. And that what matters is in fact exactly what the customer uses when making the decision whether to click or not: the search term, and the ad paired with it. Not the keyword, because that’s invisible to the customer, and all it does is attract one search term and another, not the ad (alone), and not the search term (alone). But the team. As always.

  • Stop making excuses and start tracking your leads and conversions, even if they happen over the phone or through e-mails.

    It’s pretty amazing how features which are available since long, long time ago are still neglected by both advertisers and their customers.

    And how customers and advertisers keep throwing money out the window without the faintest idea about the profitability of their “investment”.

    I’m not going to insert the famous Wanamaker “half the money …” quote here, because:

    • By now everybody knows it
    • It’s false (in many cases). It’s way more than half.

    Here’s a quick way to track conversions which start on the website and end on the phone (or in e-mails).

    Instead of hiding behind the “our leads/conversions are happening over the phone, and we cannot ask our customers which ad they clicked and what they searched for” poor excuse, talk to your webdeveloper, the person doing the advertising, and the one answering your phone / e-mail.

    Here’s what each has to do:

    • the advertiser has to tag his campaigns (auto-tagging, specific tagging, it doesn’t matter)
    • the web developer has to detect an AdWords visit and
      • switch to the AdWords phone number (show that number for AdWords visitors)
      • switch to the AdWords contact e-mail (send contact e-mails to that address)
    • the person answering phones / e-mails has to log them, together with their source

    Piece of cake, for any of them worth their salt. They’re not perfect, but you have an indication. You can tell if on one day you’re doing better or worse. And if you don’t have many campaigns, and you can use more extensions, you can track each campaign. You can even generate a code on the website and have the customer give it to you over the phone, or include it in the e-mail, and then you can track everything down to search term and ad level if you want to.

    If you can’t do that, the you can talk to your customer. You, the advertiser, can call him, as often as you can afford, and ask him how things went since your last talk.

    You can ask him how many phone calls he received, and how many orders / leads he got. You can ask him what kind of orders they were, and you can deduct which campaigns went well and which nose-dived. If you and your customer work as a team, you can do it.

    Not long ago, I convinced a friend of mine, the very first person whose account I managed, to use a different phone number for AdWords calls. And he agreed. Shortly after that, I was walking with him not far from his office, when I heard a phone ringing. He smiled, reached into his pocket and told me “AdWords”. And he went “Yes, that’s us. Sure, how many would you like? Ok, this is the price, you can pick them up anytime”. That was a live conversion, for the campaign related to the product he was talking about.

    He gave up carrying three phones (he already had two when we started using the AdWords phone) abouth a month later. By that time advertising costs were insignificant compared to his AdWords-related revenues. And it was him who made that conscious decision, I did not push him.

    He now talks to his customers and asks them how they found out about him. And I talk to him, several times a week, sometimes daily. And he tells me what he sold, how many, what the approximate revenue is, how many customers were new and how many were return customers. In my turn, I tell him what his AdWords expenses are, which products spend more, which less, which cities are more active and which less. That’s verbal tracking. And it works. In any case, it works much better than no tracking at all. I learn about his business, he learns about what I do. And together we managed to turn a  secondary product into a main revenue source.

    A few days ago he asked me, for the first time “can you see how many people have downloaded my price list?”. I smiled, but he could not see me. I was tracking downloads and hits on the contact page since I set that website up.

    “Yes, I can. About X% of your visitors download your offer. That’s about Y a day. Most of them download it from product page A, then C, then B, and you also have some downloads from D and E.” They’re from city A, B, and C. “C, are you kidding me? Are people from C visiting my website and downloading my offer?”. “Yes they are”. “Amazing”.

    Amazing, indeed. Yet so simple, so affordable and so effective.

  • Are you the master of your search terms?

    Or do they control you?

    Here’s one way you can detect what situation you’re in. Read on.

    Putting keywords in an ad group, and writing ads for that ad group is something that we do because we want to show a certain ad for a certain search query. An ad that matches that query (answers it), is relevant for that query, and entices the searcher to click on it.

    But are we making sure that for the search terms that we attract we are showing the right ad?

    Of course, if we have several ads in an ad group (and we should), and the ad group is set to rotate the ads more evenly (and it should be), a certain search term may be matched with any ad in that group.

    But that’s ok. Because everything placed in an ad group (keywords, negative keywordsads and through them destination URLs, be they ad or keyword destination URLs) should be centered around the same theme. One theme to rule them all.

    What we do not want to happen is to show, for any search term, an ad which is not related, or one which is less related than the ones we intended to show.

    And it can happen. We can have an ad group with more general keywords, which runs for a while, builds a little history, and then, even if we have another ad group – more relevant for certain search queries – we see that ads from the former ad group still show.

    The more we work on an account, the higher the chances of something like this to happen. Sure, we can diagnose our keywords, when we create a more specific ad group, to see if a search term triggers ads from the right ad group, but we don’t do it every time. And sometimes we need to wait quite a bit before the changes are active. And we cannot think of all the search queries which may match certain keywords to see if, for all of them, the right keywords are picked and the right ads are shown.

    Search terms, keywords and ads

    If you’re interested to see how Google AdWords’ system picks a certain keyword when more of them match a search query, here’s the document describing the algorithm.

    What I wanted to find though was some kind of report which could tell me, periodically, if the search terms which triggered my ads stayed all within one ad group, or if they spilled across more than one.

    And that report is a basic search terms report, adjusted a little bit in Excel.

    I prefer to do it by generating an account-wide report, and to make sure that search terms don’t creep into ad groups across several campaigns.

    Create a search terms report and create a pivot table containing the following:
    – Search term and ad group as row labels
    – Ad groups as values.

    Set the value field settings for “Ad groups” to “Count”. This should show you the number of ad groups which contain a keyword that triggered a certain search term.

    Then sort by count of ad group, descending. You’ll probably see that some keywords appear in three ad groups, or more. And when you look at the ad groups, you’ll see that they’re not different, but that some search terms are reported three times in the same ad group.

    That’s because of the match type. Not the match type of that keyword, but the match between the search term and the keyword that was triggered by it. A broad match blue widgets keyword can attract search queries such as cheap widgets, which is a “Broad match” for it, cheap blue widgets, which is a “Phrase match” for it (the search terms include the keyword), and blue widgets, which is an “Exact match” for that broad match keyword (the search query and the keyword are identical, letter by letter).

    So even if a search term only triggers keywords and ads from one single ad group, you may see it reported up to three times in the same ad group, due to the fact that in your downloaded report you may have three different rows in the same ad group, each containing a match type between search term and keyword.

    Search terms and ad groups pivot table

    Not good. You don’t want to scroll through rows and rows of data in your pivot table only to see if those search terms were linked to the same ad group several times or to more ad groups one time.

    So what you need to do is to go to your report and select the search terms and the ad group columns, and filter that list, in place, for unique search term – ad group pairs (remove duplicates). Then create a pivot table which contains, again, search terms and ad groups as row labels and count of ad groups as values. Sort the table by count of ad groups, descending.

    If, at this point, you see more than one ad group for a search term, it means that search term was able, through certain keywords, to trigger ads from different ad groups, and even campaigns. It also means you’re not in control. You can, if you want, see which keywords attracted those search terms.

    In this situation, through negative keywords and bid variations – where appropriate – you can control which ad groups show ads for certain keywords, and master your search terms.

    If you want to see where (in which ad group) certain search terms performed better, you should use the first version (before removing duplicates) of the pivot table, and see, per search term, which ad group (through its ads) suits a certain search term better. You can do that by including CTR, conversions and other relevant metrics in your pivot chart.

    And you should tweak your settings until you make sure that each search term gets to be paired with the best ad, every time someone enters it in Google’s search box.

  • You’re running an E-commerce shop, or are in charge of advertising it, and you’re looking at a loooong list of products. You have cheaper products, and more expensive products. Popular products and less popular products. High-margin products and low-margin products. And there are a lot of them. Analysis paralysis is what comes to mind.

    “I have over 8000 products on stock”, that’s what I heard from a shop owner. “I’ve tried to advertise some of them, which I knew were popular, but it did not work. In the end my cost per conversion exceeded my profit.”

    Well, truth be told, you may not be able to advertise all of them without some serious automation in place. But what you can certainly do is to pick the lowest hanging fruits.

    What you need for this is, of course, Google Analytics, your product database, and a bit of math. Nothing fancy, nothing scary.

    And it all starts with profit. You want to spend less for advertising than you’re making by selling a certain product. You want a popular product, one that people search for, otherwise your ad will not show. And you want, of course, a high conversion rate.

    Assuming that you’ve been running your website for a while before making the decision to advertise your products, you can filter, in the Content – Pages section of Google Analytics, all your product pages. Usually categories have some URLs, static pages have other URLs, and product pages have easily-recognizable URLs.

    Let’s look at Interspire, a popular shopping cart software. A product URL is something like /products/some-words-here.html. Which means that if you filter for pages matching the ^/products/.+\.html regular expression, you’ll only have your product pages showing in your content report.

    Product page views

    Export that URLs list, and go into the admin area of your E-Commerce solution and match those URLs to the corresponding products (SKUs).

    You’ll end up with a list of products, and their respective pageviews, for a certain period of time.

    Get the sales for that products, in the same period of time. Then get your margin, per product, from wherever you have it.

    You should end up with a table showing you views, sales, margin and margin per view, for that list of products.

    Now, a product is:

    • popular, if it has many views
    • high-converting, if the sales/view ratio (Bryan Eisenberg calls its inverse the “look to book” ratio) is high
    • highly profitable, if the margin is high

    When choosing your lowest hanging fruits, you’re looking for popular products with a high historical margin per view ratio. That’s the metric which combines conversion rate (sales per view), and margin.

    And that’s it. Sort descending by margin per view and by views, and pick popular products with a high margin per view, and go advertise them.

    Don’t know what the maximum CPC should be? You do know. It’s margin per view, which you may want to adjust a little to cover for the costs for running the business. That’s because the margin, i.e. the difference between the purchase price and the sales price, is not sheer profit.

    So take your monthly costs for running the business, take your total margin and try to compute an adjustment factor for margin/view, then use it as your maximum CPC.

    And if, through advertising and bidding a certain percentage of your margin per view you manage to attain the same conversion rate as your historical conversion rate, then your cost per acquisition -or CPA – should not exceed your profit per sale.

    Once you’ve launched your campaign, all you need to watch for is the cost per conversion, which should stay under the profit per sale. The initial, target cost per conversion, is (max CPC) / (historical conversion rate). If you manage to get clicks for less, or if you manage to convert better, through more precise targeting, then your CPA will be lower than the expected one.

    That’s it. No more analysis paralysis. No more “I think these products are profitable”. Your data knows which are profitable and which show potential.

    Furthermore, if you analyze the performance of your products from this angle, you’ll quickly discover which products are totally uninteresting, so you can maybe decide to get rid of the stock at a lower price and replace them with something else.

    Bonus: Don’t know which keywords to use when advertising those products you’ve just picked?

    Oh, but you do.

    Filter your product landing pages by the same regular expression as above (depending on your shopping cart solution), and add “Keywords” as the secondary dimension. Export that list and see which keywords are related to those products (which keywords were used in search by people who landed on those pages). Maybe use the keyword list splitter to separate more general keywords from the more targeted ones. The nice part about this technique is that if you’ve shown up in organic searches for that keyword, and got clicks, those keywords are relevant for that page. So the chances of seeing the dreaded “Keyword relevance: poor” in your keywords diagnostics are very, very low.


  • What keywords got me those search terms?

    Or, in other words, how broad is broad match?

    Any serious AdWords manager is watching the search terms attracted by her keywords religiously.

    And any serious AdWords manager has already been frustrated by the fact that he cannot see, in one table, side by side, the search term, the match type of the keyword, and the keyword which was triggered by it.

    Ok, it’s a known fact that you can go to your keywords tab, and select one keyword, and click on “See search terms”, so you can see only the search terms attracted by that particular keyword.

    But would you do it a hundred times a day? Click, “See search terms”, click, “See search terms”, click, oh, no, I’ve already clicked that before, back to square one …

    Disclaimer: the solution I’m going to provide only works for search terms which attracted at least one click.