Click Residual: A Query Success Metric

Walter Underwood defines click residual on his blog Most Casual Observer: the difference between the number of clicks a query should get and the number it actually gets. He developed the metric at Netflix.


Definition

exp = att × ctr
res = clk − exp
  • att — query attempts. An attempt can be a submitted query or a displayed results page, so the metric also works for character-by-character (search-as-you-type) search.
  • ctr — the average click-through rate across all queries
  • clk — actual clicks for the query

A negative residual marks a query where people searched but didn’t click much. The post first printed the second formula with the sign reversed; Underwood fixed it after a reader pointed it out.

Why Not Just Low CTR

Ranking queries by low CTR mostly surfaces long-tail queries, where fixes barely move overall results. Sorting by residual puts the biggest customer impact first. That covers both moderately common queries with very low CTR and very popular queries that underperform by only a few points.

Netflix Examples

  • “batman” was the top underperformer one week. The Dark Knight doesn’t contain the word “batman”; adding an alias fixed the query after the next indexing run.
  • “cj7” came from fans looking for a Stephen Chow film not yet released in the US. The movie metadata team added an entry so subscribers could follow its availability.

People