I think the key thing is that they shouldn't be estimating things to which they
can calculate the exact answer (reasonably in the circumstances) when they first start estimating at least. Later, when they're comfortable with estimating, that's fine, but they do need to see that estimating lets them do things they couldn't do otherwise!
Here's a question DS8 (well, he's DS9 now, but he was DS8 then) recently did [ETA with no calculator allowed, obviously]:
In 1954, a total of 6527mm of rain fell at Sprinkling Tarn and this set a UK record for annual rainfall. The tarn has a surface area of 23450m^2. Roughly how many million litres of water fell on Sprinkling Tarn in 1954?
A 15 B 150 C 1500 D 15000 E 150000
Here's another; this one is even better, actually.
A newspaper headline read ‘Welsh tortoise recaptured 1.8 miles from home after 8 months on the run’. Assuming the tortoise travelled in a straight line, roughly how many minutes did the tortoise take on average to ‘run’ one foot?
[1 mile = 5280 feet]
A 3 B 9 C 16 D 36 E 60
ETA: this is one of the areas DS finds hardest; picking the right kind of approximation to get you the amount of precision you need is a bit of an art. If this is typical, and it sounds as though it is, I suppose that's why teachers resort to setting such "easy" estimation questions. But I really doubt it achieves anything.