Estimate Your Own Lost Impressions Share Data
Back in January, Google announced their intention to begin offering ad group level impression share metrics
within the ad words interface. Alongside this they announced their plan to update the algorithms they use for their calculations. This was to provide a more accurate assessment of campaign impression share metrics.
As part of the changes, the calculations of impression share data will now only update once per day and unfortunately for those who live by GMT or similar, these once per day updates take place at noon Pacific Time (GMT-8). So that means our campaigns will not show impression share data for yesterday until 8pm today. This is probably too late to make bid or budget changes based on this info.
So if it is important for you to get this data sooner, what’s the answer? Well, there's not much you can do with lost impression share due to rank, but with lost impression share due to budget, you can estimate your own using a model you are comfortable with. Even when this data comes from Google, it is an estimate based on trends, so we can make our own estimates based on the historical data in our account.
OK so to get an estimate, which is sensible, I would suggest that there are two things we need.
Firstly, we need some sort of model which represents a ‘typical’ distribution of impressions throughout the day for the account. To do this, grab some top level impressions' data for your account and segment this by hour of the day. Then find the percentage of impressions each hour that contributes to the daily total. We should also add the cumulative total percentage for reasons which will become apparent later in the calculation.
Bear this in mind: the longer the date range is that you pull this data from, the more accurate model you are likely to get. However, you should ensure this data is ‘typical’ in that it was with the same budget and has no changes which would affect impressions, such as major keyword changes.
Next we need to collect the impression data for the day that you are trying to calculate, remember to segment by hour of day. Now, if we have significant lost impression share, the later rows may be blank. Let’s compare:
So we can see that the budget ran out a little after 10pm. For the calculation, I’m going to assume it ran out at 10pm and ignore the 1,699 impressions after this.
So to calculate, I’m going to sum the impressions received up to 10pm which is 105,094. Using our distribution model we can estimate that this relates to 91.5% of impressions we receive in a day and so, to calculate 100% of impressions, divide our total by 91.5 and multiply by 100.
(105,094 / 91.5) * 100 = 114,857
And finally, to get to get a lost impression percentage:
(114,857 – 105,094) / 114,857 = 8.5% lost impressions share.
There are a few things to bear in mind if you are going to do your own calculations this way:
- It will only work if your campaign delivery method is set to accelerate, and was set to accelerate for the duration from which you take your impressions model data.
- The validity of any result is very dependent on the quality of your impression distribution data.
- If you have campaigns that perform differently, produce a different model for each campaign.
- If you have days of the week which perform differently, produce different models for each day.
Have the changes to the impression share data affected your campaigns? Leave a comment to let me know.