The graph shows that the manually-calibrated board (Competitor 1, blue line) exhibits extreme errors up to 0.3% of full-scale as a result of temperature swing, while the autocalibrated board (Diamond Systems model no. DMM-32-AT, red line) exhibits a maximum error of no more than 0.03%, a 10-fold improvement.
Note that the error curve for both boards is unpublished and unpredictable and therefore uncorrectable (unless you conduct this test yourself).
A 0.3% error essentially limits your A/D measurement accuracy to about 8 bits, regardless of the A/D board's resolution or any other published specification. Is 8-bit performance acceptable for your application? If not, you need an A/D board with autocalibration.