Exposure is the right frame and it needs a denominator. Impressions alone rank a publisher's largest exposed set as its most at-risk one, when that set is often the least reachable in value. Two additions to the workbook:
1. Weight the exposed query set by demand and commercial value before ranking it. Without that weighting the output is a priority list a commercial director can dismiss in one sentence. 2. Report a citation measure next to the impression measure. A link surfaced is not a link seen, and the two move independently.
On the log layer: a crawler fetch and an agent-runtime call are different events on different paths, so split bot access by user agent before it stands in for demand. Then validate the split with a one-variable change against a frozen query set.
The layering gets you the signal. The weighting is what makes it a decision.
This was a really interesting read.
Thanks very much, I do my best!
Exposure is the right frame and it needs a denominator. Impressions alone rank a publisher's largest exposed set as its most at-risk one, when that set is often the least reachable in value. Two additions to the workbook:
1. Weight the exposed query set by demand and commercial value before ranking it. Without that weighting the output is a priority list a commercial director can dismiss in one sentence. 2. Report a citation measure next to the impression measure. A link surfaced is not a link seen, and the two move independently.
On the log layer: a crawler fetch and an agent-runtime call are different events on different paths, so split bot access by user agent before it stands in for demand. Then validate the split with a one-variable change against a frozen query set.
The layering gets you the signal. The weighting is what makes it a decision.