An uncalibrated sensor produces numbers that look authoritative and mean very little. Calibration is not a one-time factory step, it is a process with a schedule.
Establish a reference you trust
Every calibration is relative to something. Decide what your reference is, know its own accuracy, and keep it separate from the equipment you use daily. A reference that lives in the same toolbox as the working meters will eventually be used as one.
Offset first, then gain
Correct the zero point before the slope. An offset error applied through a gain correction produces an error that scales with the reading and is much harder to spot later. Two point calibration across the range you actually use beats a single point at full scale.
Characterise temperature drift
Most sensors move with temperature, and industrial installations swing more than a lab. If you can, sweep the ambient temperature during characterisation and record the coefficient. If you cannot, at minimum measure at the extremes you expect and state the resulting uncertainty rather than pretending it is zero.
Calibrate the chain, not the part
The number your firmware reports is the product of the sensor, the analogue front end, the reference and the conversion. Calibrating the sensor alone leaves the rest uncorrected. Do it end to end, through the same code path the product uses.
Store the coefficients on the device
Calibration data belongs in non-volatile memory on the unit it describes, along with the date and the reference used. Coefficients kept in a spreadsheet will be separated from the hardware within a year.
Decide the cadence and write it down
Set a re-calibration interval based on the drift you measured, and record every calibration event. When a measurement is later questioned, that history is what lets you answer with evidence instead of an opinion.