I'd suggest thinking through whether consumption patterns differ substantially between HHs and individuals. This seems likely.
One could argue an individual will order both on a standalone basis and in a group (HH) and so the smallest unit to model would be per individual.
Then you could always segment this into different segments of individuals e.g.
1 - young, solo living
2 - couple
3 - young family
4 - older couple
Within each of these segments you can then assume order frequency and also taking into account # individuals / HH
Hope that makes sense