Product Implementation
Suggested minimum data-set size for model training
Models should have at least 1,000 samples for each label. For example, "good" and "bad" classes used in a single model need at least 1,000 samples each.
Suggested maximum data-set size for model training
Individual .csv files should be no larger than 12 MB. Customers that require larger samples can use incremental training and chuck the file into batches.
Model quality dependence on data volume
Although not impacted by small changes in the data set, the model needs to be redesigned when classification requirements or language styles are updated.
Plan your implementation
Ticket categorization
Once a service ticket arrives, the machine learning algorithm automatically categorizes it into a certain category with a certain confidence level and level of urgency. The system decides where the tickets need to be routed, assigning them to various functional teams where customer service agents can immediately start with the most urgent ones.
Solution recommendation
When a ticket arrives, the machine learning algorithm analyzes the unstructured text and pulls out most similar tickets and closest solution articles. Customer service agents go through them and pick up the most relevant one. Then they respond to the customer without having to do a manual search through their databases.