When do you need to update the algorithm in Data science?
You can think of it in terms of input , process, output.
Input : When attributes of the input data changes
e.g. the range of values are different, the prediction values/classes are beyond what the model was originally trained on, data stability has changed etc
Process : When the model construct needs to be updated - for example earlier you had national level models, now you need state level models, or if a newer version of the algorithm is available, or the model parameters needs to be fine tuned etc
Output : When the model measurement metrics go for a toss - example accuracy, precision , recall, AUC, ROC , false positives, true negatives, senstivity , specificity etc are degrading
Note that the output metrics are indicators while input and process modifications are the root causes that need to be tweaked to get the output metrics back on track.