aharper | 2019-05-23 13:37:45 UTC | #1
I was not sure where to post this question. Please move around if needed.
Customer is wanting to know the specifics for the algorithm for EWT. In my understanding, when working with another client and Genesys, recently, there had been some modification around the way EWT calculates. While I know it is "estimated" and I know it looks at certain items listed in the documentation, it still leaves a few things unmentioned. "Other system behaviors" - see below excerpt.
"These predictions take into account the queue length, agent availability, handle time, as well as other system behaviors and variability at a given moment in time, i.e. when a request is made."
What exactly is the EWT using now for its calculation? Is it the last 15 interactions? I do not think it is still looking back to the queue from 7 days ago at the exact time, correct?
I just need to help my customer understand what this is looking at as the number appear to be very skewed from time to time.
I have explained that if a interaction sits for a long time, say, in wrap up, or they have many different types of interactions in one queue, that can also impact the EWT, however, it is only calls.
Only looking for a bit of clarity on this subject.
I appreciate your time, Angie
tim.smith | 2019-05-23 18:57:02 UTC | #2
It sounds like you've already found the Overview of Estimated Wait Time page. My understanding is that our EWT calculation is considered a trade secret and we intentionally don't publish the exact details for that reason. However, if you open a case with PureCloud Care, they may be able to provide some additional details in a private setting.
aharper | 2019-05-23 20:00:11 UTC | #3
Thank you Tim, I had a case opened and they told me to post here.
bayu | 2019-05-24 14:32:06 UTC | #4
Angie,
Thank you for your question. In calculating the EWT and PiQ, we are indeed looking at-the-moment state of the queue (how many agents are presently active, how many interactions are currently waiting in front of the line etc.).
Our formula leverages queuing theory, and as such we need to know a steady-state server time (that is, how long a typical agent would take to handle an interaction), and that is where the 7-day average is used (for estimating AHT). The reasoning behind it is that we want to take in the average AHT so that it doesn’t swing too wildly especially at the beginning and end of the day where there’s no much data point to refer from. While this provides a good proxy for a typical handle time, it will have trouble when there is a sudden spike in handle time (e.g. where the last several interactions takes much longer than the "typical" handle time). We are actively looking into this issue.
There are also some complexity that affects EWT when agents are activated in multiple queues, where our formula tend to under-predict. This is due to the fact that we are over-estimating the availability of agents (e.g. when activated on 2 queues, on average, we should expect only half their time is available in handling a queue). We are also actively working to resolve this issue.
system | 2019-06-24 14:32:11 UTC | #5
This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.
This post was migrated from the old Developer Forum.
ref: 5221