Technology Stack
Challenge
The travel retailer was investing significant resources in manually reading, assessing, categorizing and redirecting thousands of emails per week to the correct team for processing. This process was slow, resource intensive and open to human error. ​During peak times, this could cause increased time delays on an already high priority process.Â
Solution
Blackbook used an open-source deep learning transformer model and fine-tuned it to data provided by the client. ​
The model was deployed in UiPath AI Centre, where it categorises emails to one of the eleven email classes with an initial balanced accuracy of 77%. Any emails that fall below the model’s confidence score are validated by a human and captured for model re-training.​
Outcome
The travel retailer is now able to respond quickly to emails as they are automatically triaged by UiPath AI Centre and processed by the relevant team. The model can cope under sudden increases in email load and with the model re-training, its accuracy continues to increase over time. The re-training pipeline also allows for the customer to add or remove email classes over time in line with strategic business changes.Â