When it comes to our retirement accounts, we all expect that the person managing our funds will be paying close attention to the yield of the assets, maximizing our potential upside, and consistently rebalancing our investments.

Yet when it comes to our paid search campaigns, we don’t always approach them with that same level of rigor or analysis.

Why not?

To be fair, it’s non-trivial to simply get all the data you need into one place to make the right decisions. On top of that, you might not even know what to look for once you do have all your data in one place.

This guide serves as an introduction to the Portfolio Management Model of paid search campaign optimization. By the end of this article, you’ll be able to look at all your paid search keyword investments to quickly reveal insights as to where you can “rebalance” your paid search investments.

The Basic Concept

In this model, we treat our keywords like the “assets” in our fund (in this case, our paid search campaigns). We will be analyzing our keyword data to determine the different tiers of performance that our keywords fall into – you can think of these performance buckets as a “keyword asset class.”

We can then summarize our account by these buckets – and look at how much of our money is going to keywords in the different performance categories, as well as how “saturated” we are in those categories (i.e. how high our impression share for those keywords is).

Getting and Preparing Your Data

First, we need to get our search keyword data. For the sake of simplifying this analysis, we will only deal with “active” keywords in our account.

Start by getting a keyword report from your Google Ads account for all of your campaigns that includes these important metrics:

We’ll go ahead and export a .csv to use Excel for now (although you could certainly get fancy and fetch this data via API + use a programming language like Python for all of this!).

Making Impressions Share More Flexible

There are two problems with the impression share metric we need to solve:

1. Sometimes it includes symbols for very high or very low impression share numbers

   1. E.g. “>90%” or “<10%”

2. It can’t be averaged together directly when attempting to “roll up” your keyword data

   1. E.g. Imagine you have a campaign with 2 keyword targets

      1. Keyword A has 90% impression share and 9 impressions

      2. Keyword B has 10% impression share and 100 impressions

      3. The campaign impression share is NOT 50% (average of 90% and 10%)

      4. The campaign impression share is instead closer to ~11%

To make this easier, we’ll instead calculate a “market impressions” metric that represents the total eligible impressions that were available for each keyword. So if our keyword had 9 impressions and a 90% impression share, we would calculate a “market impressions” value of 10 (i.e. we captured 90% of 10 available impressions).

Once we have a “market impressions” metric, we can slice and dice our data however we want to and still get back to “impression share” by calculating the quotient of “impressions” to “market impressions.”

What should we do about the greater than or less than symbols? In this case, we are just going to close our eyes and delete them! In all seriousness, it won’t impact our calculations too much to replace “<10%” with “10%” – and this rounding estimate will save us a big headache.

Dealing with Statistical Significance

We are almost ready to crunch some numbers, but we have one additional challenge – how do we deal with statistical significance?  This is especially important if we have a lot of keywords in our portfolio that have 0 conversions. How many clicks should we buy before we turn these off? 5? 10? 500?

There are plenty of different mathematical approaches to calculating these types of thresholds, but for our purposes – we are going to keep it (somewhat) simple. Do you remember this question from high school or college statistics?

“What’s the probability of flipping a coin 10 times and getting 0 heads?”

To calculate this, we first take a deep journey into the world of independent events and probability, then we come back out and calculate this probability by multiplying the probabilities of each individual outcome occurring in succession.

50% x 50% x 50% x …

= (1/2)^10

= 1/1024 chance of flipping a coin 10 times and getting 0 heads

In our keyword data, we are going to do something similar by asking a slightly different question:

“What’s the probability of getting X many clicks and seeing 0 conversions for this keyword?”

To answer this, we are going to start with a “target cost per conversion” for our keywords (we’re hoping you have a target in mind – if not, figure this out before spending any more money!).

For example, if we want a $100 cost per conversion and a keyword has an actual CPC of $5, then we need a 5% conversion rate to hit our goals.

If a keyword had 150 clicks, 0 conversions, and a $5 CPC, we would ask this question: “what is the probability of this keyword getting 150 clicks with a 5% conversion rate and observing 0 conversions?”

We then pick a probability at which we want to address these keywords. For example, we might only want to pause off this keyword if there is a 5% or less probability of observing this outcome.

We will spare you the math, but you can generalize this function and use it to calculate a “statistical click threshold” for each keyword target by using that keyword’s actual CPC and a target cost per conversion as follows:

Viewing Your Portfolio Performance

Once we have adjusted impression share metrics and accounted for statistical significance, we simply need to divide our targets up into buckets.  You can create your own performance thresholds, but here is an example:

• Non-converters + Insignificant

• Non-converters + Significant

• Converters +  200%+ Target Cost per Conversion

• Converters + 150-200% Target Cost per Conversion

• Converters + 100-150% Target Cost per Conversion

• Converters + 50-100% Target Cost per Conversion

• Converters + <50% Target Cost per Conversion

You can use multiple methods to label your targets this way.  Once you have, you can use something like a Pivot Table in Excel to summarize the following stats by performance category:

• Count of keyword targets

• Spend

• Impression Share (Sum of Impressions / Sum of Market Impressions)

You’ll quickly see how you may want to shift budgets between these targets once you get your data organized into this format:

What To Do Now?

It’s easy to see how you want to move your money around between these buckets, but it’s much harder to actually do it.

Sometimes keywords are limited because of the campaign budget, but sometimes it’s the bid.  Other times, the keyword that is performing well is mixed into an ad group with a keyword performing poorly – and this then requires you to restructure the account.

We strongly suggest working through these changes on a keyword-by-keyword basis, starting with your top-spending, poorest-performing terms.

Once you get in the habit of looking at your data this way, you’ll be able to quickly spot issues in any account and easily prioritize your optimization of current paid search investments.