Two years ago customer services asked us to automate the approval of small damage claims. If a customer's claim was under two hundred and fifty pounds, a credit note was issued without anybody looking at it. Claims above the limit went to a team of three, who took around five working days. The automation removed about seventy percent of their workload.
Over the following year the number of claims rose by a little over two and a half times, while deliveries rose by six percent. When we plotted claims by value there was a cliff: hardly anything between two hundred and fifty and three hundred pounds, and a tall spike between two hundred and twenty and two hundred and forty nine. Several trade customers were submitting three or four claims for a single damaged delivery, one line at a time. Most of them were not being dishonest. One of our own agents had been advising customers to split larger claims because, as she put it, it was the only way to get them paid this month.
The automation had done exactly what we built. What we had not considered was that a threshold, once people learn where it is, becomes a published rule for avoiding a person. The limit had been chosen to separate small claims from large ones. In practice it separated customers who knew about it from customers who did not, and the ones who knew were those who called us most.
The limit now applies to the total claimed per delivery and per customer each month, not per form. One automatic approval in twenty is sampled by the claims team, chosen at random rather than by value. A weekly chart shows the distribution of claim values, and a new spike just under any limit is treated as an alert in its own right. The other half of the fix was the queue. The team now decides larger claims within two days, which removed the reason most people had for splitting them.
Claim volumes settled at about one and a half times their old level, which we think was genuine demand all along.
Automating a decision below a line turns the line into something people work towards. Before we set an automated limit now, we ask who will find out where it is and what they will do about it.
– Serguey Shinder
Top comments (0)