Building predictive algorithms from exceptions

Most operations leaders tell us exception handling is a burden. We take a different perspective. We see outliers as a unique opportunity to boost the intelligence of AI. OnCorps is actively running systems that queue and learn from exceptions. Our goal is not simply to handle outliers more efficiently, but to use these cases to improve our algorithms.

How the solution works

Learning from exceptions

Learning from the observations of users is critical to improving the predictive performance of the AI. We do this by testing hypotheses on observable conditions or events that are easy to identify and potentially more predictive. To do this, we ask users to click on parameters they observe, then correlate these variables with the outcome we are predicting.

What the solution does

Our algorithms unlock efficiencies and reduce risk

Reducing false positives

Once the checks are run against transaction datasets, we typically find that adjustments are needed to reduce the number of exceptions produced.

Learning from observations

Each exception is queued to the right person. Our AI shares all relevant data and asks the user to observe and describe what they see.

Intelligent workflow

Our system is capable of applying rules to assign certain exceptions to the right expert. In this way, leaders can be certain the right people are seeing exceptions.