PPC expert

Is Your Account Managed by a PPC Expert? A Scientific Approach to Find Out.

In this piece we hope to remove some of the subjectivity associated with evaluating the skill of a PPC manager. By definition, any analysis involves collecting and analyzing data, more specifically, data related to the type and quality of optimization decisions over time. This kind of information is supposed to be contained in a PPC audit, so we will start with a brief discussion of the basics of a ‘standard audit’ – the kind of thing you’d likely receive if you paid somebody to audit your account today. After discussing some of its limitations we will then introduce a new audit framework that aims to better illustrate how well somebody manages an account. In other words, here’s your key to separate the real PPC experts from the poseurs.

A Typical PPC Audit and Why It Isn’t Good Enough

A standard PPC audit is essentially a glorified checklist whose objective is largely to determine if somebody is spending an acceptable amount of time optimizing an account, if the optimization decisions appear to follow some kind of logic and if the optimizations occur at some acceptable time interval. It seeks to answer questions such as “Does the campaign/ad group structure make sense?” “How often is bidding happening and do the bid adjustments appear to be formulaic?” “How often are negative keywords implemented?”

The answers to these types of questions are typically stored in some kind of table or spreadsheet and then a score is assigned to each. When the audit is completed the scores are totaled and the account gets some kind of grade. Some companies use letter grades like B+ and other use some kind of numeric grade such as 8.5/10.

While it may seem somewhat disparaging to characterize this type of audit as a checklist, it is the necessary first step in any PPC management evaluation and its importance should not be understated. If somebody hasn’t changed a mobile bid modifier in the entire account in the last six months, we’ve learned all we need to know about the management of that account. But what do you do when all these boxes are checked? Are you sure the account is being well-managed?

Let’s face it, anybody with a few years’ experience as a PPC practitioner will be able to take care of the basics. An account audit that encompasses only this basic information can’t easily differentiate between ‘great’ account management and ‘average’ account management, it only sees ‘bad’ and ‘not bad’. Over time we found ourselves increasingly subject to and purveyors of audits that didn’t provide much clarity around how good somebody actually is at managing PPC accounts because the audits failed to explain the quality of optimization decisions.

So we built a better audit.

What Differentiates PPC Experts From the Rest?

There are two concepts that companies can use to enhance the usefulness of their PPC audits:

  1. Convergence of certain KPIs (key performance indicators) across account entities (such as campaigns or devices) and
  2. Correlation of aggregate KPIs over time.

The remainder of this post will look at these concepts. When customized for business-specific objectives, these two concepts will provide a foundation from which better audits can be constructed, and will seek to answer the question: is your account being managed by a PPC expert?

Enter convergence, which we will define as “the trending of a KPI to a specific target”. A standard objective for a PPC account is to meet or exceed some kind of profitability metric. We work with e-commerce businesses so as an illustration, we’ll use a common e-commerce objective: ROAS (return on ad spend), calculated as “revenue divided by ad spend”.


Let’s say our target is 500% ROAS. For the sake of simplicity, we will assume that all aspects of the account are going to be managed to achieve a 500% ROAS target, that the account received a decent grade on its standard audit (say B+), and this account, on the whole, does meet its 500% ROAS objective. But assume you notice that over the last 12 months, mobile ROAS has consistently produced a 350% ROAS. And assume you also notice that during this same time period, the top 20% of income earners (assuming you have added household income tiers to your AdWords campaigns) have consistently produced a ROAS of 800%.

If you were to plot the moving average of mobile ROAS in this illustration, you would not see the line approach 500%; instead, you would see it has bounced around 350% for the last 12 months. This represents a persistent over-investment in mobile traffic. The opposite is true for household income tiers – there was persistent under-investment.

Our position is that all of the optimization segments in the above illustration – devices, locations, audiences, campaigns, ad groups, everything – should see their ROAS converge over time to 500% if the account is being managed by an expert.

If there are segments that consistently over-perform or under-perform relative to a given KPI objective, then revenue maximization for a given budget is not occurring. In this example, the ideal state of the account would be one where all segments are generating ROAS at exactly 500%.

Obviously this perfect 500% ROAS scenario will likely never actually happen, which is why we like to use convergence (or lack thereof) over time, and not a specific achieved target, to gauge the success of optimization decisions. Expertise is, then, the ability to adapt optimization decisions to the unique characteristics of each of these segments within a PPC account. Whether or not the adaptation is occurring is shown by the existence of convergence of segment performance toward the account’s aggregate goal.

Next, consider correlation of aggregate KPIs.

We’ll stick with the same scenario as before (500% ROAS goal). Correlation is defined by a value between -1.0 (perfect negative correlation) and +1.0 (perfect correlation). While this value can be useful, we are more concerned with how that value trends over time.

A good way to determine whether or not your account manager is getting better at optimizing your account is to build scatterplots such as the sample below.

To build a scatterplot, plot revenue (sales) on the Y axis and ad spend on the X axis (secondary Y axis represents orders, which adds another dimension to the scatterplots and can be used to augment the interpretation of correlation as defined in this exercise). Each point represents a day. Here is a group of quarterly scatterplot charts for our illustration:

PP experts Seasonality of ROAS

The important thing to take away from the above charts is the trend over time.

From Q1 to Q4 the R-squared value (the square of the correlation coefficient) improved from 0.35 to 0.96 (the correlation coefficient improved from 0.59 to 0.98). Because we are comparing revenue and ad spend, a correlation coefficient of 1 would represent a time period where ROAS was exactly 500% every day, and leads to the conclusion that the distribution of ad spend was where it should be. In other words, in this illustration revenue maximization occurred because exactly the “right” amount of money was spent each day.

Clearly, it is unlikely that anybody will generate the exact same ROAS every day, the key is that only the expert account manager will evolve their optimization decision-making processes to generate more-consistent aggregate performance.

Is Your Account Manager Getting Better?

What sets true PPC experts apart from the rest can be summarized as the improvement in the quality of optimization decisions over time. To understand whether or not this happening, we suggest exploring ways to incorporate convergence and correlation into a standard PPC audit. If there is convergence of certain KPIs across account segments, and if there is consistent improvement in the correlation of certain aggregate KPIs, then you can rest easy with the knowledge that those tasked with the management of the account are doing a good job.

If not, you know whom to call!

  1. Jesse says:

    Hi Peter, did you build the charts in this post with excel or with a program like R or Matlab? Trying to replicate the color gradient.

    • Peter Erickson says:

      Hi Jesse,
      The charts in this post were created using R. The name of the R package used to create these specific charts is currently eluding me – if I find it I will make sure to reply to this post. Hope this helps!

Comments are closed.