Amazon Connect weeknotes, w/e 3 July 2020

It’s been a little quieter on this project this week; it feels like it’s settling down somewhat. However, we’ve still been tinkering with the system to optimise things and I think it’s paying off. Over the last week, the stats are as follows:

  • 2407 calls were offered (that is, received by the system)
  • 673 calls were partially deflected (that is, the caller used the bot but still spoke to an agent)
  • 124 calls were fully deflected (that is, the caller only used the bot or payment line)

We now have some more useful data in the new report as well. The deflected call rate is now 33.1% – a huge increase that properly reflects calls that are routed to the payment line as well, so not touching an agent. Even stripping out the payment line, the rate is still 16.8%, a significant increase on last week. Of those calls handled by the bot:

  • 346 callers used the balance option
  • 60 callers used the transactions option
  • But only 156 callers passed identification & verification 

That’s a pass rate of 38.4%; not fantastic but it has improved since we changed the messaging to highlight the need for the payment reference. We’ve also implemented a retry loop during this week, so callers get two attempts with each piece of information before being passed to an agent, so next week should see another improvement in this rate. 

Feedback remains steady. Although the response rate has reduced, residents are firmly in two camps. Those who have had a trouble-free experience love it, those who have had problems with the bot don’t love it. We’ve been able to do some investigation on some of the poor experiences and noticed a minor issue with the Transactions API, which has been amended. 

During the week I had a chat with colleagues at Durham County Council, who have implemented voice capture to route calls and provide recorded information, though they don’t provide personalised information like we are. It was a useful conversation and I’m reassured that we are doing the right things in this pilot, especially around reason capture as that really drove their strategy. Their deflection rate is high, and it’s a good benchmark. 

I’m going to be off for a couple of weeks but I’ll write in the middle of next week about the outcome of the service assessment and potential next steps. 

+ posts

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.