Okay, what about breaking the deaths down by location factor and using a weighted average calculation.
For example:
Motorways: increase in deaths has a -1% effect on their revenue retention
A-roads: -5%
B-roads: -10%
Residential: -20%
These could be set as maximum swings with stepped targets for each. For example, if deaths on A roads go up 4%, they lose 2% revenue, a 10% increase or more in A road deaths brings the full 5% penalty.
This would allow such motorway catastrophies over which they have little control to have a minimal effect on their finances, whereas the roads that they should have a major input into making safer would carry larger penalties as they are much less likely to be hit by a 'one-off' event that skews the figures.
It would also give much more focus to those roads that are more likely to have pedestrians on them rather than those lovely profitable motorway bridges.
Complex I know, but workable.