Not every search means the same thing. We judge each one by what the searcher actually wanted, not just where it ranks.
A person ready to buy and a person just learning can land on the exact same search position and click very differently. This model tells them apart before deciding what normal performance looks like, so your opportunities are measured against the right expectation, not a blended average.
Picture two people searching at the same time, both landing their search at position 3 on Google.
One types "buy motorcycle gloves." They are ready to act and will click through to a product without much hesitation.
The other types "how to clean leather gloves." They are just learning. They might read a snippet and move on, even from a great position.
If you judge both of these against one shared normal click rate for position 3, you get a number that is wrong for both of them. It makes the buyer's page look fine when it might not be, and makes the learner's page look like a failure when it is actually behaving exactly as expected.
Before this model calculates anything, it first works out why someone likely searched by using the actual words in their search:
Once every search is sorted this way, the model builds a separate, fair benchmark for each combination of position, brand versus non brand, and intent. A ready to buy search at position 3 is only ever compared to other ready to buy searches at position 3, never blended in with casual browsing traffic.
Without this layer, a CTR opportunity model can only tell you that a keyword is underperforming its position. With it, you also learn what kind of miss you are looking at, and that changes what the actual fix should be:
It uses the same underlying math as a standard CTR model, but the fix it points you toward is far more specific.
Generic CTR advice treats every underperforming keyword the same way, usually recommending a title tag rewrite as a catch-all fix. That works sometimes and completely misses the mark other times. A page failing to convert someone who is ready to buy needs a different fix than a page failing to satisfy someone who is just researching.
This model exists to make sure the recommended fix actually matches the problem, so time gets spent on the right change rather than a generic one.