For RPA with intelligence, you want to identify a process that humans are doing that requires some additional decisioning to be made, or provides an opportunity for some intelligence to be applied. Here are a couple examples:
What if you have a general inbox, firstname.lastname@example.org, where various people reach out with different needs? Someone is reading the e-mails as they come in and forwarding them on to the appropriate departments; billing, customer service, sales and so on. With a history of e-mails and where they were routed, a multi-class project could be created in DataRobot. RPA could monitor the inbox, and as requests arrived, score the e-mail through a DataRobot model to decide where to forward it. If the confidence in a class is not strong enough to meet a routing threshold, then it can be forwarded on to a human monitored inbox for manual handling. The manual routing could also serve as input to a future model training set to deploy an even more accurate model as time goes on.
Another usage of RPA could be taking documents received and inputting them into an existing tracking and processing system, like various forms that support insurance claim processing. Rather than just entering data from received documents , a model could be trained in DataRobot based on historic observations of fraud. The new documents could be evaluated for fraud, and when a threshold has been met indicating likely fraud, the claim could be forwarded on to Subject Matter Experts for further investigation to determine if the claim warrants intervention.