Measuring Shopping performance
Retail measurement fails in a specific way: the number that moved is real, and the reason attributed to it is not. Most of the fix is deciding what to look at before you change anything.
The metrics worth watching
- Search impression share. The share of auctions you were eligible for and appeared in. The clearest signal of whether budget or bid is the constraint.
- Benchmark CPC and benchmark CTR. What comparable advertisers pay and get for similar products. The only external reference point Google gives you.
- Click share. Clicks received against clicks available. More useful than CTR for judging competitive position.
- Per-product conversion value. Where the money actually is, and usually concentrated in far fewer products than anyone expects.
- Disapproved and expiring items. A performance metric wearing a diagnostics badge.
Reading a before-and-after
- Fix the window before you make the change, and make it a whole number of weeks.
- Compare the same campaigns and the same products, excluding anything added or removed in the period.
- Check the seasonal comparison too, against the same weeks last year, before concluding anything about a change made in November.
- Look at CPC and impression share together. A flat CPC with rising impression share is a win that a CPC-only reading calls a failure.
- Change one thing at a time, or accept that you are measuring a bundle.
After a CSS switch this matters more than usual, because Smart Bidding will convert part of the saving into volume. Judging the switch on average CPC alone systematically understates it.
The concentration problem
In most catalogues a small fraction of products produces the large majority of revenue, and account-level averages are dominated by them. Segment by product group before drawing conclusions: an average CPC that rose because your best product got more expensive is a different story from one that rose across the board.
What not to measure
Position. Shopping does not have one in any useful sense, because results are a grid that varies by device, query and surface. Impression share is the metric that replaced it, and it answers the question people are actually asking.