The question every performance drop needs answered first

A metric can fall for two entirely different reasons that look identical on a chart: the account got worse, or the market did. A recommendation built on the wrong one wastes budget fixing something that was never broken, or cuts spend right as demand recovers on its own.

The check

The check that separates them is not complicated, but it is the first one skipped under time pressure: compare the change month-over-month against the same change year-over-year. A metric down both ways points to something structural in the account. A metric down month-over-month but stable or up year-over-year points to a seasonal pattern the account did not cause and does not need fixing.

A worked example

Illustrative case: conversion rate on an account drops 18 percent from July to August, and the instinct is to look for what broke -- a landing page change, a bidding shift, a competitor move. Checked year-over-year, the same August-versus-July drop shows up in the prior year too, at a comparable size, and the account's traffic is in a category with a well-known summer dip. Nothing in the account needs fixing; the market did what it does every August. Skip the year-over-year check and that same 18 percent drop reads as an urgent account problem instead of an expected seasonal one.

Where the distinction actually has to land

That distinction only matters if it actually blocks a bad recommendation before it reaches someone, not just a note buried in an appendix. A finding that cannot show a plausible cause, or cannot point to real evidence behind it, does not get to become a recommendation -- no matter how clean the correlation looks on its own.

The payoff is boring, on purpose

The result of getting this right is boring, and that is the point: fewer recommendations, not more, each one more likely to survive contact with what actually happened once someone acts on it.

When neither comparison settles it

Not every drop resolves cleanly into one category or the other. A metric can be down both month-over-month and year-over-year while the underlying cause is still seasonal, if the prior year's comparison window itself had an unusual spike. The check is a strong first filter, not a verdict on its own -- it narrows where to look next, and a finding that still cannot point to a plausible, checkable cause after that stays a question, not a recommendation.