Automate customer ticket handling: from an idea to a production ready model
summary
End-to-end system classifies incoming customer tickets into operational categories and triggers replies or agent routing, built in ~6–7 weeks then iterated; initial text-only model hits ~60% accuracy until adding 150 metadata fields and a two-step classifier that combine text distribution and account/fraud metadata and flips many labels. Implementation uses Azure cloud with models developed in R, Microsoft Translator for language normalization, ~1,000 tickets ingested per day, retrained every 4–6 weeks, and GDPR-safe anonymization.
AI-written summary · about
chapters (12)
- 0:00Start
- 2:18Who Is a Python Developer
- 7:37Create a Roadmap
- 8:34Rapid Prototyping
- 11:11Education of Business
- 17:08Technology Assessment
- 19:45Text Classification
- 21:10Ramp Up Setup
- 22:28The Increase of Value for the Client
- 22:51Optimize Specific Cases
- 24:16Distribution of Knowledge
- 30:34Use Automated Inbox Replies in Mail Clients
related talks
46mHandling Support Tickets with AI Without Burning the Planet
30mDriving a Higher Level of Customer Support with Machine Learning
01h 39mShipping complex AI applications
30mAI-based Assistant for Business Process Outsourcing
32mHarnessing Large Language Models for Enhanced Customer Experience
33mAutomating Resolution of IT Support with Chatbots
51mTaking Models to the Next Level with Azure Machine Learning
08mAI-Driven Innovation for Enhanced Food Delivery Operations