Before a designer or user can decide what to show in a search or matching application, it is imperative to understand whether it is best to Undermatch or Overmatch.

It comes down to which case causes more or less problems for the business.

If it is simply a case of reducing the cost of mailing by avoiding duplicates then undermatching is good. Yet if it was important to avoid annoying the recipient, then overmatching would be good.

If it is a matter of not letting a known terrorist into a country or on to a plane, then overmatching is essential and, as in all security systems, a necessary consequence will be that some innocent people get inconvenienced by the process.

In a statistical process the consequences of undermatching can not be measured, but experiments can be designed to measure the amount of overmatching in the results.

In all designs it is necessary to know whether one would rather miss things, or rather find some things you did not want to find. Once one accepts that error and variation in the data is normal and unavoidable, then it is true that absolutely correct matching cannot be achieved, and it becomes necessary to decide if the "maybe true" answers should be seen or hidden from view. This is a fundamental business decision.