Continuing our series on AI management options (kicked off by the HBR article Managing AI Decision-Making Tools), the next option is Human in the Loop For Exceptions (HITLFE).
This option is probably the most common in our clients. Our typical client is aiming to get a reasonably high rate of straight through processing – automated decisions – while still allowing for some human intervention.
There are those who think of HITLFE as having the computer recommend and then letting the human decide whether to override or not. We’re not fans of this approach as it has two issues. First, it puts unrealistic pressure on the human decision-maker who might have to accept or reject the computer recommendation very quickly and without having the option to investigate every part of the automated decision. Second, it assumes that the decision must be automated or not as a lump rather than considering how human and automated decision-making can be integrated. We prefer to have human decision-makers handle exceptions in a more specific and directed way.
For instance, we have several clients working to improve their loan and mortgage origination. Central to these is the ability to automate approvals. Verifying income is one of the most complex pieces of this process as there are so many kinds of income with different verifications. We designed and built automated decisioning that can verify many kinds of income. When it cannot, it can create a case for an adjudicator. Rather than just asking them to review the whole application, though, it asks specifically for a review of the part of the income that could not be automatically verified – the exception. Once the review is complete, the adjudicator submits their assessment of the income and the rest of the automated decision is re-executed with the new income included. This allows a detailed explanation of why a mortgage was approved to be kept – logged by the automated system – and engages humans to handle exceptions in a focused, efficient way that ensures they don’t waste their time on repeating tasks already considered in the automated system. Other clients have used a similar approach to insurance underwriting, engaging underwriters to assess specific risks when an automated assessment is not practical or desirable.
This kind of exception handling with humans is great for more complex decisions, but when simpler decisions are being made and when 100% automation is a requirement, what you need is a Human on the Loop (HOOL) and that’s the third kind of system.
If you missed part one of the series where we discussed “Human in the Loop (HITL)”, click here.
If you have any additional questions on this article or other related topics, drop us a line – we’d love to connect.