Every few months I end up in the same argument. A lab run on a template scores well, the field data for the same URLs does not, and somebody has to decide which number the next quarter of work gets planned against.
The gap is usually explainable. A lab run uses one device profile, one connection, and a cold cache, on a page where nobody has interacted with a consent banner and no third party has loaded for a signed in user. Field data carries every device, every network, and every one of those interactions.
What I do now is stop treating them as two measurements of the same thing. Lab data tells me whether a change I am about to ship helps. Field data tells me whether the page is fast for the people actually using it. A lab improvement that does not move field data over the following month was a change to something that was never the constraint.
Two things I am less certain about.
The first is how long to wait before calling a field change real, given that the window is twenty eight days and a deploy lands in the middle of it.
The second is what to do with a template that has too little traffic to produce field data at all, which on a large site is most of the long tail. Lab data is all there is, and lab data is exactly the number I just argued not to plan against.
How do you handle the second one?
Swiss Knife SEO ModeratorAdministrator