Another insurer is computing sentiment for clients from the call center in real time. A key data point for that bot is the polarity score, which measures the emotion expressed in a sentence. This score can be used as another predictor for cross-sell and up-sell models. The polarity can be positive, negative or neutral.
At SAS, we’re building a fraud surveillance bot to better detect fraud, especially anti-money laundering fraud. Since fraud is a rare event, it’s naturally hard to detect. An important goal in this area is to minimize the false positive rate, so the fraud detection bot doesn’t classify legitimate transactions as fraud.
Sometimes an investigator can triage a false positive in just a few minutes but often it can take 8 hours or more – which isn’t a good use of time. By contrast, the bot summarizes variables for classifying fraud and automatically selects a probability threshold cutoff to minimize the false positive rate.
Since the fraud investigative process is labor intensive, banks are hiring more and more investigators to meet demand. A bot that deploys powerful machine learning pipelines to accurately detect fraud can help focus investigative resources where they are really needed.
The bot can adapt to new fraud approaches and attacks through continuous machine learning. Natural language generation provides narrative summaries about why a transaction or set of transactions if fraudulent or not. This speeds the investigative process and helps with compliance.
How to build your own bots
Another goal at SAS is also to build open, extensible software for customers who aspire to build their own bots or virtual assistants. We plan to deliver a natural language interaction (NLI) service that converts keyed or spoken natural language text into application-specific, executable code.
In other words, it automatically maps a user’s command to the correct action, like: “summarize sales” or “score a truncation for fraud.” We’ll also include a natural language generation (NLG) service for developing narrative summaries and templated reports. The goal is to reason on input with the NLI service and explain on output with the NLG service.
This article was republished with permission from SAS.