The rule that decides between them: ask how much the answer may change before your decision changes. That allowed slack is the precision you need — no more, no less.
Approximate answer is good enough — buying rice for a school picnic.
80 students × about 150 g cooked-rice serving ≈ 12 kg
buy a 15 kg bag — an error of 2 kg either way changes nothing
A 15% error here costs a little extra rice, and nothing else. Estimating is not laziness; measuring each student's appetite would be pointless work.
A very exact value is needed — the dose of a medicine for a child.
paracetamol dose = 15 mg per kg of body weight
for a 20 kg child: 15 mg kg⁻¹ × 20 kg = 300 mg exactly
double this, repeated, and the liver is damaged
Here a 100% error is dangerous, so the child's mass and the syrup's concentration must both be measured properly.
Why it happens: precision costs time, money and equipment, so science spends it only where it changes the outcome. The picnic decision is tolerant — many answers lead to the same action, buy one bag. The medicine decision is sensitive — a small change in the number changes the outcome from cure to harm. The same distinction runs through all of science: an estimate is enough to decide whether an experiment is worth doing, but the final measurement it produces must be as exact as the instrument allows.
More pairs to think about: approximate — how many buses for a school trip, how much paint for a wall, whether a bridge design is even in the right range. Exact — gold weighed at a jeweller's, the diameter of a machine part, the timing of a spacecraft's engine burn.