And they wont be foolish to pay for
  non-results for long...

Imagine Joe gets re-elected after delivering 30,000 pre-election leaflets, knocking on 3,000 doors, delivering 10,000 reminders on the day of the poll, getting 13,000 views of facebook adverts, employing 2 people to do casework for constituents, calling 7,000 people with a human-operated phone-bank, robocalling 20,000 people, attending 40 community events, getting 300 people rides to polling stations, getting 5 celebrity endorsements, getting 1 visit from a really high profile colleague, and so on...

It's far from obvious to me that you could tell apart the effects of different efforts - especially as you don't know who each voter voted for.

>It's far from obvious to me that you could tell apart the effects of different efforts - especially as you don't know who each voter voted for.

Companies that do those short of campaigns stagger them at different times, do polls before and after, test a new method in this or that area or demographic and check again and so on through election year.

It's not just "apply a bunch of methods at once" -> "check the final election result" as the only feedback.

still in this case the candidate who spent the most lost the election