# Leaving it to Chance on Pay Per Click is the Best Approach by Far!

#### Paul C

PPC

The simple answer is no. Search behaviour and the performance of your campaign can be modelled mathematically using probability theory to make informed decisions that will maximise your profit. Google supplies you with figures which can help you to optimise your campaign, but the interpretation of these figures can be a daunting task causing most people to simply guess.

By doing the analysis properly you can almost completely eliminate the guess work and make correct decisions based around science rather than chance.

## Pay-Per-Click Decision Making

Consider the following statistics on a keyword.

PPC Case 1

Impressions Clicks Ave. Cost/Click Conversion Ave. Profit/Coversion
200 4 1.00 0 0

Given these statistics, should you make a change? Should you amend the price, change the landing page, change the ad copy, or perhaps drop the keyword altogether?

You would probably say that there are not enough clicks on which to base a decision and we have only spent 4.00 so far. You should also consider that when you do make a change, if you want to make a subsequent decision in future, you need to start your statistics from the time you made the change because you dont want your next decision to be polluted with previous statistics.

So, given the above statistics, we probably do not have enough data do make any concrete decisions.

Consider then the following PPC situation.

PPC Case 2

Impressions Clicks Ave. Cost/Click Conversion Ave. Profit/Coversion
2000 40 1.00 1 20.00

#### Should you make a change now?

If we focus on the statistics of clicks to conversions to profit before we decide whether our click through rate is good enough (impressions to clicks), you can see that we have paid 40.00 for a return of 20.00. Therefore we have lost money on this keyword so far.

However, we still only have one conversion. If we get another conversion soon and we assume we get about 20.00 per conversion, the keyword suddenly starts to look promising.

Consider then the following PPC management situation.

PPC Case 3

Impressions Clicks Ave. Cost/Click Conversion Ave. Profit/Coversion
200000 4000 1.00 100 20.00

We are now losing more significant money on this keyword. We have paid 4000.00 for the keyword and we have only made 2000.00 profit. We have a large enough sample size to say that probably this keyword will never recover and start performing if we do not intervene.

In all three cases, the proportions of clicks to conversions are very similar. So how do we know at what point to intervene.

Most Pay Per Click advertising managers actually make too many decisions. In doing this they lose any hope of ever discovering the real underlying performance of their keywords.

## Pay Per Click Decision Making

If we toss a coin twice and both times it comes up heads, do we conclude that the coin is either biased, or is a double headed coin? Probably not. Four tosses, four heads, we become a bit more surprised, but probably not suspicious.

If we toss it 20 times and it comes up heads every time, we are probably going to be more suspicious. The question is, how suspicious?

To put the question scientifically, what are the chances of achieving 20 heads in 20 tosses, if our coin is completely normal and fair? If we can put a number on this probability, we can determine the exact likelihood of the coin being normal. We can determine the exact likelihood of us needing to swap the coin for one that is performing to expectation.

We can apply similar statistical models to place probabilities on the performance of a keyword. This is essential in PPC decision making. With large volumes of keywords and masses of data, you cannot simply go down the list and make changes on gut feel.

Take PPC Case 2 above as an example.

Impressions Clicks Ave. Cost/Click Conversion Ave. Profit/Coversion
2000 40 1.00 1 20.00

If we assume that orders (conversions) are on average worth 20.00, with the average cost per click at 1.00, we need a PPC conversion rate of 1 in 20 (or 0.05) in order to break even. We have actually achieved so far a PPC conversion rate of 1 in 40 (or 0.025).

We can see that the keyword looks to be not performing. But how sure are we that this is a keyword that will not perform.

To put the question scientifically:

What are the chances of getting the results we got in PPC Case 2, if indeed the long term and underlying performance of the keyword is to break even?

My guess, looking at the figures, would be that actually these conversion results are not too bad for such a small sample, and that if indeed the long term outlook for this keyword is break even, there would be a fairly high chance of getting this poor start to the sampling.

To justify this conclusion we have to do the maths, which unfortunately means we have to get a bit technical about it so I’ll try to keep it simple. We can estimate the standard deviation of the long term performance of the keyword, around the average (mean) value of the break even conversion rate of 0.05. The standard deviation tells us important information. The average or mean conversion rate is what most PPC managers will base decisions on, but the standard deviations gives us an exact mathematical indication of how we expect the results to deviate around this mean. In other words, if we know the standard deviation, we can calculate how likely it is that our keyword is not performing.

If something has a probability of 0.05 of happening, and we take a sample of 40 (the 40 clicks from PPC Case 2), statistics will tell us that we can expect a standard deviation of the results of 0.0345. So with a sample size of 40, we expect the resulting probability to deviate around the mean (of 0.05) with a standard deviation of 0.0345.

Using probability statistics, we can calculate that the probability of getting these actual results in Case 2 is: 0.24 or 24%.

Therefore for PPC Case 2, the probability of getting these results (1 conversion in 40 clicks) when actually the keyword will eventually break even is 24%.

This is a high probability.

We can conclude that we should not start changing this keyword based on these statistics.

We have to apply this very same statistical modelling to thousands of keywords in order to keep your pay per click campaigns on track.

Unfortunately PPC management is not as simple as applying standard formulae to individual keyword statistics. Additional clues in the campaign will guide a skilled manager in making much better decisions. For example, the multiplying word buy when attached to keywords invariably makes it perform better. For example, buy heaters is a better keyword than heaters. This makes sense, but unfortunately the search volumes on this deeper keyword, buy heaters, are such that we would be waiting much longer for statistical significance.

At Search Laboratory, we have specialist systems that enable trained staff to take advantage of statistical significance across a group of keywords. In this example, we can analyse the effect of the multiplier buy across all product keywords.

Grouping and slicing the keyword statistics in this way allows us to maximize the profit of a campaign for a client. This is a mathematically difficult thing to achieve, when faced with masses of keywords, varying amounts of data against each keyword, and random behavior of searchers.

## Search Laboratory Pay Per Click Optimisation

At Search Laboratory, we take a highly statistical approach to the management of pay per click bid management. We do not leave the performance decisions on keywords to the gut feel of the PPC management executive. We apply a consistent and auditable strategy to all campaigns in order to achieve the maximum cost effectiveness for your campaigns.

It is important to apply statistical methods to each area of pay per click management to determine, as well as keyword performance, optimum rank for a keyword, maximum price in relation to each rank, pair wise comparison results (for example, which landing page of two possibles is the most effective) and other such decisions.

Do not leave your pay per click performance to chance.

Insights

Insights

"