A little OT perhaps, Automation is much much faster for straightforward claims. This is a definite advantage for passengers and the operator, decisions can be made almost instantly and passengers have been known to receive compensation (when using instants methods such as BACS or PayPal) the same day, or shortly afterwards, and these Northern Free Tickets can be on your doorstep in a week in many circumstances.
I did quite a lot of data analysis on this in 2019 (over a hundred hours) with lots of data, some freely available and some requested and noticed a lot of trends. I never got to drawing up a detailed set of conclusions due to other commitments but plan to obtain more data this year and next to include some comparative data whilst doing so.
The primary issue is a risk of the system paying out on invalid claims. The way to get around this is to bias the system design to only accept the most definite absolute cases and choose what to do with the rest (which will only be a small percentage). The most customer friendly approach would be to refer those to a human who could look through and contact the passenger for more information to clarify something before making a decision. Of course, the cheaper option is to just reject them outright. This is the option many TOCs take. It’s then down to the passenger to appeal that decision where it MAY then be looked at by a human. This provides an extra opportunity to thin out the claims further as many will just accept their defeat even if they were eligible. These systems ARE improving but we still see obvious systemic issues such as
* rejecting missed connections where the arrival time and departure time were 0 minutes or more (regardless of minimum connection time).
* using timetable data instead of running data when a train is cancelled.
* only paying out on the leg of the journey that was delayed, even with a through ticket.
* using running information from a train that did not call to calculate compensation.
And many others.