Impact of NLP in AI-powered Bank
summary
NLB deploys two email classifiers in production since March: a 10-topic classifier and a sentiment pipeline built from 5,000 Slovenian emails with manual labels. Models use a staged “waterfall” of smaller classifiers—best-performing approach is stacked SVM waterfall—plus language detection and hand-written priority rules (spam low, negative/reclamation high). Results reported: about 90% auto-topic assignment to correct agents, improved sales win rate (~30% uplift) and customer score NPS 57.
AI-written summary · about
chapters (11)
- 0:00NLP in AI-powered banking
- 3:23Why NLP matters
- 5:31NLP in banking ops
- 6:54From idea to models
- 10:33Training data challenges
- 12:40Topic classification approach
- 17:19Sentiment analysis design
- 21:31Contact center impact
- 25:09Contact center gains
- 27:44Future NLP developments
- 30:18Q&A
related talks
32mHarnessing Large Language Models for Enhanced Customer Experience
55mDemystifying Natural Language Processing across several industry verticals
35mBank evolution to data science-driven business model
31mWhen humans teach machines
37mNatural Language Processing for Email Routing
17mNatural Language Processing in Automation
06mAdvancements in Natural Language Processing (NLP) and Text Analytics
15mMachine Learning for All