There's a huge potential in mining search data, especially if you can group top-of-funnel vs. bottom-of-funnel searches to see where failures are occurring. Segmenting queries like "zm950" against "shoes" or "nike" and seeing where gaps exist against user intent.
When it comes to zero (or near-zero) results, I've had good results using this to identify gaps in the current product offering and what visitors are expecting to be there. Two examples:
1) A seller of custom prescription glasses: top two search queries were "contact lenses" and "sunglasses". They didn't offer the former, they did sell sunglasses (most frames could take a tinted lens as an option) but didn't make it obvious with design, content, or marketing.
2) A seller of cabinet hardware (pulls & knobs): a large proportion of their top 10 search terms seemed to have a door hardware intent. Adding this missing category boosted sales without additional marketing dollars spent (the customers were already there and just bouncing when they realized the site didn't carry what they wanted).
These are all ways to focus on understanding failures instead of trying to optimize successes, which is often finding the local maxima.
When it comes to zero (or near-zero) results, I've had good results using this to identify gaps in the current product offering and what visitors are expecting to be there. Two examples:
1) A seller of custom prescription glasses: top two search queries were "contact lenses" and "sunglasses". They didn't offer the former, they did sell sunglasses (most frames could take a tinted lens as an option) but didn't make it obvious with design, content, or marketing.
2) A seller of cabinet hardware (pulls & knobs): a large proportion of their top 10 search terms seemed to have a door hardware intent. Adding this missing category boosted sales without additional marketing dollars spent (the customers were already there and just bouncing when they realized the site didn't carry what they wanted).
These are all ways to focus on understanding failures instead of trying to optimize successes, which is often finding the local maxima.