Despite having huge range of beautiful applications, weight sensors come with a lot of hurdles for their implementation. One lies in its architecture. One has to be extremely careful of how the architecture comes into play for performance. The part of the sensor, which works to transduce the weight, has to be in freely suspended condition with holding the entire load at its own. If it comes to touch with any other part, it cannot distort proportionately with the reaction of the load. Hence, the transduced value of volt shall never reflect the load.
The other one is its non-linear characteristics. The millivolts generated from a particular load do not linearly correspond to that generated from another load. Therefore, to get rid of this one, one has to interpolate through multiple readings of the millivolts (or ADC value) from multiple known value of loads from tedious formulae.
Another huge problem is the induction loop. The arms of wire left open, could be prone to EMI and external noises that could interfere the infinitesimal voltages of the sensor. Hence, we could come up with corrupted ADC values. Methods are to be takes, this induction loop has to be minimized with twisting of negative and positive wires, and they have to be shielded with appropriately.