How wrong is a ZIP code?
Most sales tax rate data is keyed on ZIP codes. ZIP codes are postal delivery routes. Tax jurisdictions are not. We measured how far apart those two things actually are.
What the measurement is
We take addresses, resolve each one to a point, and find the rate by testing that point against real jurisdiction boundaries. Then we ask what the ZIP-level rung would have returned for the same address, and compare.
About 1 address in 11 gets a different rate from the two methods. In the worst ZIP codes the gap reaches 2.70 percentage points. On a $400 order that is $10.80 of tax charged wrongly, in one direction or the other.
Why it happens
A single ZIP code can straddle a city line, a county line, and two transit districts. Whoever publishes a ZIP-to-rate table has to pick one answer for the whole ZIP. Usually that is the most common rate inside it, or the lowest. Either way, every address on the wrong side of the line inside that ZIP gets the wrong number, and nothing in the response says so.
What we do about it
We cannot make the ZIP-level rung accurate. Nobody can; the information is
not there. What we can do is stop pretending, which is what the
precision flag is for:
address: the address was placed inside real boundaries. This is an answer.zip: it fell back to the ZIP. This is a floor, and it is wrong about 9% of the time.
A caller who wants to be careful can treat zip as a reason to
flag an order for review, or to collect a little more and reconcile later.
A caller who does not care can ignore the flag. Either way the information
is in the response, instead of in a footnote in somebody's documentation.
Judge us on this
This is the number we would want to see from a rate vendor, and the one none of them publish. If you can show us a measurement where we do worse than this, we want it: tell us.