A 1000 kg car moving at 20 m/s rear-ends a stopped 1000 kg car, and their bumpers lock. Momentum before is 1000×20=20000 kg·m/s, so the pair moves off at 200020000=10 m/s. Kinetic energy before was 21(1000)(20)2=200000 J; after, it is 21(2000)(10)2=100000 J. Half the kinetic energy went into crushed metal, heat and noise. Total energy is conserved; kinetic energy is not.
Crumpling metal absorbs energy, but it cannot absorb momentum: every push between the cars comes with an equal and opposite push, so their momentum changes cancel. Sticking together is the extreme case. Objects that bounce apart can still be inelastic: a rubber ball dropped on a floor that rebounds to 70% of its starting height has lost 30% of its kinetic energy in the bounce.
Contrast an explosion, a recoil or two skaters pushing apart. These look like a sticking collision run backward, and momentum is still conserved, but kinetic energy increases, supplied by chemical energy or muscles. If the kinetic energy after is larger than before, the event is not a collision losing energy but a release of stored energy, and neither the elastic nor the inelastic model describes it.
Think of it like this. An inelastic collision keeps two ledgers. The momentum ledger always balances exactly. The kinetic energy ledger shows a shortfall, and the missing amount turns up in other accounts: heat, sound, bent metal. Sticking together produces the largest shortfall that a balanced momentum ledger allows.
Core idea. Momentum is conserved in every collision, but in an inelastic one some kinetic energy becomes heat, sound and deformation; when the objects stick together the loss is as large as momentum conservation permits.