How to average impression share across days without getting the wrong answer

This is our argument, not Google's. A plain average of daily impression share gives every day the same weight. A day on which the campaign was eligible for a handful of impressions moves the result as much as a day on which it was eligible for thousands, so a month with a few quiet days at a high share can report a better figure than the campaign actually achieved.

Google defines impression share as impressions divided by total eligible impressions, and says eligible impressions are estimated using many factors, including targeting settings, approval statuses and quality.

Our arithmetic, from that formula: rearrange it and each day's estimated eligible impressions are its impressions divided by its impression share. Add up the impressions over the window, add up the estimated eligible impressions, and divide the first total by the second. That is the share for the window, measured the way Google measures a single day.

The division holds only where the impressions and the share describe the same traffic. Google says search partners won't be included in impression share data, that on Shopping campaigns only your first impression for each search will qualify for impression share, and that Performance Max impression share is impressions from Search + Shopping over total eligible impressions from Search + Shopping. So we do the division inside one Search campaign, on an impression count that leaves search partners out.

A worked illustration, not a client result. Day one: 100 impressions at a 50 per cent share, so 200 eligible. Day two: 900 impressions at a 75 per cent share, so 1,200 eligible. The plain average of the two shares is 62.5 per cent. The window's share is 1,000 impressions over 1,400 eligible, about 71 per cent. The busy day was the better day, and the plain average let the quiet one drag it down.

Google defines Search lost IS (budget) and Search lost IS (rank) as percentages of time that your ads weren't shown on the Search Network, due to insufficient budget and to poor Ad Rank in the auction respectively.

This is Riibon's choice, not a Google rule: when we roll either loss column up across days, we weight each day by the same estimated eligible impressions. A day the campaign could barely have shown on should not carry the same say in the month's loss figure as a day it was eligible for thousands of impressions. Weighting a percentage of time by an impression count is a proxy, and we use it knowingly.

Google says that a dash in place of impression share might be due to insufficient data, such as low traffic or new keywords, and that data processing for these metrics can also take 24 to 48 hours to update. Its data page adds that all impression share metrics are updated within 1 to 2 days.

What we see in the data, as opposed to what Google documents on the pages we cite: when we read impression share through Google's API, a share below 10 per cent or above 90 per cent arrives as a fixed placeholder value rather than as the measured share. We leave those days out of the weighted figure rather than treating the placeholder as a reading, and we count them, so the report can say how many days were withheld. The cost is a bias we accept and disclose: leaving out days above 90 per cent pulls the window's figure down, and leaving out days below 10 per cent pulls it up, so a window with many withheld days carries that count beside it.

Our practice for the gaps: a dash is not a zero, a placeholder is not a measurement, and a window with no measured day returns no figure at all rather than a mean of nothing. The most recent day or two of any window may simply not be reported yet.

Google says impression share metrics are reported separately for each campaign type and aren't aggregated across the account. Our reading: a Search campaign's share and a Display campaign's share are measured against different pools, so we roll up within one Search campaign and do not blend campaign types into one account figure.

Our practice, as a checklist. Rebuild every multi-day impression share from impressions and estimated eligible impressions, never from the daily percentages. Weight the lost IS columns by the same eligible impressions. Leave out dash days and placeholder days, and say how many of each there were. Return nothing for a window with no measured day. Roll up within one Search campaign, and never across campaign types. And read the share beside the absolute impression count, because Google says changes to your bids, quality or Google's ad systems may change the set of auctions in which the system estimates that you were competitive, so the pool can move as well as the share.