PPCHow to Avoid 7 PPC Analysis Pitfalls

How to Avoid 7 PPC Analysis Pitfalls

Red Herrings abound when you're looking at your AdWords data. Information overload is a risk. The last thing you want is to see a relationship between a couple of your PPC metrics, make a decision, and later rue that you didn't investigate further.

Red herrings abound when you’re looking at your data in AdWords. Information overload is a real risk. The last thing you want is to see a relationship between a couple of your PPC metrics, make a decision, and later rue that you didn’t investigate further.

With that in mind, here are seven common mistakes people make when looking at their PPC campaigns and how you can easily avoid them.

1. Data Mining

iron-ore-miningBig data is a hot topic at the moment, with good reason: the availability of enormous datasets gives us a lot of potential that we didn’t previously have to analyze relationships, personas, and activities. But with it has come a problem: data mining. Data mining is a trendy term, since we have so much data to mine!

Data mining is a dirty word. Econometricians learned this 30 years ago, but analysts like us are only just starting to think it’s cool.

It’s tempting to say, “We’ve got all this data. Let’s burrow around and see what presents itself that we didn’t already know.” It’s even more tempting if you have the ability to make serious decisions as a result.

“Knowledge is power” isn’t just a saying or a sci-fi claim. If you’re Facebook and you learn how people interact, you can target better ads and get more clicks. If you’re an advertiser and you find a time of day that doesn’t drive any sales, you can save money. The more you know, the better your eventual profits.

The problem: we’re approaching it backwards. We can’t start with the data and find relationships. We need to start with the theory and use the data to attempt to refute it. If it can’t be refuted, your theory lives to fight another day.

With the size of datasets you’re looking at in AdWords you have a lot of entries for a lot of metrics. If you load up your data into a chart and look for trends, you’re going to find some. It’s practically bound to happen.

If you have 10 metrics, you have more than 3 million possible relationships between them. If there is a 5 percent chance of accidentally finding a relationship where none exists, then you’re going to find more than 100,000 relationships. If you’re just looking for the best one, it is unlikely to be one of the limited number of true, meaningful relationships that exist.

To fix this problem you need to start with the hypothesis. “I think that my weekend traffic is less valuable,” “My click-through rate rises as my average position increases,” or “The average order value drops after my phone lines close.” These are all acceptable theories worth investigating. But choose your theory first, decide what data you need to answer your question, and download the report from AdWords.

2. Averaging Your Averages

averaging-averages

I see this all the time from novice Excel users. They create a pivot chart and include click-through rate (CTR), or ad position, or CPC. Clients send me these kinds of charts all the time with questions. This simply won’t work.

When AdWords sends you calculated metrics (anything that is a ratio/percentage, or an average) you then can’t sum or average up these values, without accounting for weighting.

Scenario: Campaign A has 100 impressions in average position 1, campaign B has 1,000 impressions in average position 5. If you were to average up your average positions, they’d come out as position 3, but the true average was 4.6. Most people are sensible enough to realize this isn’t the correct technique, but if people are using pivot tables and have perhaps 50 rows of data, it can be easy to not realise the problem.

The fix for this is easy. If you’re using pivot tables, use the “calculated metric” section to create your calculated metric based on the raw data. If you’re creating the table yourself, just make sure you run your calculations on your totals, not on your raw columns.

2a. Believing Your Overall Ad Positions

It’s even more common to see reports indicating average position at ad group or campaign level. This is a meaningless statistic.

Unless you also know your standard deviation and your skew, an average ad position doesn’t tell you enough to make any judgements. An average position of 3 could be caused by half in position 1 and half in position 5, or equally by three quarters in position 2 and one quarter in position 6.

Fix this by making sure that stats which are obviously only keyword-appropriate (e.g. average position) are only analysed at keyword level.

3. Sorting by CPA

cpa-1-per-click

When you’re trying to identify your worst performing ad groups or keywords it can be easy to sort by CPA from high to low and weed out the poor performers. This is most common among novice AdWords users or those very short on time.

If you’re sorting by CPA, you’re missing all keywords without conversions.

Sort by cost and look for high cost or high CPA keywords, and find the problems that way. Don’t let non-converting keywords continue without attention.

4. Using Impression Share on Broad Match Campaigns

Impression share is meaningless when you don’t know which search queries you’re advertising to. Since you can’t see your impression share at search query level, you need to make sure that your campaign targets the search queries you think it does.

Either ditch the broad match or ditch the idea of meaningful impression share metrics.

5. Including Too Many Conversion Types

If you’re tracking multiple types of conversion, please remember that they aren’t all going to be worth the same value to you. AdWords is bad at presenting different conversion types differently.

When Google took away the ability to separate the conversion types alongside click data, we asked them why. They said “Advertisers found it too confusing.” No they didn’t. Advertisers who shouldn’t even have been using the feature might have, but for the rest of us it was yet another valuable piece of data gone.

Don’t optimise your campaign to CPA if each conversion type is worth a different amount. Either:

  • Choose one primary conversion type to optimize toward, or
  • Use the API to extract the data in a meaningful way and separate these.

6. Making Decisions Too Soon (Or Too Often)

clicks-0

Only using one date range to make decisions is dangerous: you either have too little data, or you mask recent changes. But there’s another side to this, which is campaign managers who persistently make changes on tiny amounts of data.

The most frequent example is seen when launching new creatives. If you’re deciding between two ads with 100 impressions each, you really don’t know which has the better CTR yet, no matter what the numbers say.

Be patient. Wait for data to accrue. I can’t give you hard numbers or say “once you have x impressions you can decide on y.” You’ll have to use your own judgment. But whenever you think it’s time, wait a bit longer.

7. Writing Off First Page Bid Estimates

below-first-page-bid

This is a dangerous tactic, usually done by experienced account managers.

The problem arises in this form: “I always ignore first page bid estimates. I see them for keywords with average positions in the banner!” This isn’t a fault of the first page bid estimate, this is a fault of the campaign manager’s understanding of average position.

Average position is weighted by impressions, not by searches. On any search where you didn’t not appear on the first page it won’t contribute to your average position. The metric is implicitly going to show you an average position on the first page, even if most of the time you weren’t.

Because you don’t have impression share at a keyword level, you can’t know if your keyword is on the first page any reasonably amount of the time or not. Your first page bid estimate is the best guess you have.

It’s not without its problems. Sometimes it can fluctuate wildly in short periods of time (more than could be explained by competitors running out of budget, etc.). Sometimes it can absurdly high.

Normally, bad data is worse than no data, but in this case it’s not true. The slightly questionable first page bid estimate shouldn’t be written off, it’s your best indicator. The top page bid estimate is even better.

Others

This isn’t an exhaustive list, but every one of these problems is rife among marketers running AdWords campaigns. Some are more prevalent in novice campaign managers and some still persist in the more experienced community.

The best thing you can do is to ask somebody else to glance over your work when you’re analyzing your campaigns. If you can find somebody who can sense check this stuff it’ll go a long way towards catching you doing something you know you shouldn’t.

Resources

The 2023 B2B Superpowers Index
whitepaper | Analytics

The 2023 B2B Superpowers Index

8m
Data Analytics in Marketing
whitepaper | Analytics

Data Analytics in Marketing

10m
The Third-Party Data Deprecation Playbook
whitepaper | Digital Marketing

The Third-Party Data Deprecation Playbook

1y
Utilizing Email To Stop Fraud-eCommerce Client Fraud Case Study
whitepaper | Digital Marketing

Utilizing Email To Stop Fraud-eCommerce Client Fraud Case Study

1y