Genesys Cloud - Main

 View Only

Discussion Thread View
Expand all | Collapse all

Customer First Callback Impacts on EWT

  • 1.  Customer First Callback Impacts on EWT

    Top 25 Contributor
    Posted 10-09-2024 23:09

    I have a few questions with how people are using this and if they are seeing impacts to EWT presented to customers.

    First are you using the same queue as the original call / original callback to send the customer first callback call to? Are you using a flow to update the priority first or direct to queue, if direct to queue how do you ensure it isn't sent to the back of that queue?

    Assuming you are updating priority from flow or DataAction from InQueue Flow, are you finding EWT presented is impacted, as I could be wrong, but my understanding with how it is calculated (https://developer.genesys.cloud/routing/routing/estimatedwaittime-v2) is that if you have multiple interactions that push to the front, this will impact that lower quartile value, and make seem like the longer calls are the anomalies and tell the other customers a lower EWT than what they actually should expect.

    This also assumes you have a high volume of callbacks, with enough 6+ calling out to customers in a short space of time.


    #Reporting/Analytics
    #Routing(ACD/IVR)

    ------------------------------
    Anton Vroon
    ------------------------------


  • 2.  RE: Customer First Callback Impacts on EWT

    GENESYS
    Posted 10-10-2024 08:36

    Good question, looking forward to seeing if people have noted this and had to workaround it



    ------------------------------
    Sam Jillard
    Online Community Manager/Moderator
    Genesys - Employees
    ------------------------------



  • 3.  RE: Customer First Callback Impacts on EWT

    Posted 19 days ago

    Thanks @Anton Vroon. I am also interested to know more about this questions from other community members responses.



    ------------------------------
    GANDHI SHANMUGAM
    ISM
    ------------------------------



  • 4.  RE: Customer First Callback Impacts on EWT

    Posted 19 days ago

    I don't know the answer but looking forward to get additional information on this topic



    ------------------------------
    Gupta Kg
    IT Cloud Integrator (Senior Consultant)
    ------------------------------



  • 5.  RE: Customer First Callback Impacts on EWT

    Posted 19 days ago

    Found the below, hopefully it helps?

    AdjustedAHT = (AWT * NumberOfAgents) / PositionInQueue
    where:
     
    NumberOfAgents: The number of agents activated and online (On Queue) in the queue when the EWT inference was done for the completed interaction at a point in time in the past.
    PositionInQueue: The position in queue when the EWT inference was done for the completed interaction. When conversationId was not given, this would be the (QueueLength + 1).
    AWT (Actual Wait Time): The observable time elapsed (in secs) between the time when the EWT inference was done and the actual time that an agent answered the interaction.
     
    Once these samples of recent observations are made, the formula uses the Median of these sampled AdjustedAHTs to create a single PredictedAHT value to be used in the following calculation to generate EWT inference of the target interaction. In the situation where there is only a limited number of cached records (up to 3), the PredictedAHT uses Mean instead.
     
     
    EWT =  PredictedAHT * PositionInQueue / NumberOfAgents
    where:
     
    PredictedAHT: A representative service time for the interaction to be handled.
    PredictedAHT is the most likely predictor of an interaction service time durations.
    PredictedAHT takes into account the most recent completed interactions and their adjustedAHTs.
    PositionInQueue: Position of the current interaction in queue, i.e. it reflects how many interactions are currently waiting in the target queue waiting to be handled by agents, including itself.
    PositionInQueue is derived based on the actual arrival time and the conversation details which include: media type, current queue length, and how long interactions have been waiting in queue.
    PositionInQueue also takes into account utilization and interruptibility setting at org level, i.e. media types priorities (see: https://help.mypurecloud.com/articles/utilization/).
    PositionInQueue is assumed to enter to the end of the queue, unless it can interrupt other interaction(s) currently waiting in queue.
    When querying for an interaction already waiting in queue, PositionInQueue reflects the position in queue the interaction is currently in. Consequently, each subsequent EWT query will reflect its advancement through the queue.
    Generally, an interaction that has:
    Earlier arrival time has a smaller PositionInQueue, while later arrival time has a higher PositionInQueue.
    Higher priority has a smaller PositionInQueue. while lower priority has a higher PositionInQueue.
    Interrupting media type (e.g. inbound phone) has a smaller PositionInQueue, while the interruptible media type (e.g. email) has a higher PositionInQueue (i.e. driven by the utilization setting).
     
     
    NumberOfAgents: The number of active agents in the queue, i.e. the total number of activated and online (On Queue) agents in the target queue at the time the request is made.
    NumberOfAgents is derived from the real-time observation of the number of agents activated and online for the target queue. This includes agents that are currently handling conversations.
    The algorithm inherently takes into account situations where agents are activated in multiple queues Nq, as it is looking at the efficiency of agents handling past immediate completed interactions.
     
    Additionally, in order to ensure that the formula is robust to outliers, 1.5x Inter-Quartile Range (IQR) distance from median of cached AWTs is used to as a guardrail to make sure the resulting EWT inference reflects the majority of how most recent past interactions are handled. That is, when the raw EWT inference is above the upper range, it is capped to the upper range. Conversely, if it is below the lower range, it is capped to the lower range (with a caveat explained further below).


    ------------------------------
    Adrian Bradshaw
    Enterprise IT Engineer
    ------------------------------



  • 6.  RE: Customer First Callback Impacts on EWT

    Posted 19 days ago

    Thanks for Anton raising this question and Adrians info. In addition, you could consider using strategies like distributing callbacks more evenly across queues or limiting priority overrides to prevent significant distortions in EWT. Another approach could be to implement separate queues for callbacks and regular calls, ensuring that priority interactions don't disproportionately affect regular call wait times.



    ------------------------------
    Matt Calton
    ------------------------------



  • 7.  RE: Customer First Callback Impacts on EWT

    Top 25 Contributor
    Posted 18 days ago

    Thanks Matt,

    I've been trying to get the business to reconsider their priority assignment for a while now, unfortunately not about to happen any time soon. We have calls that get 300+ priority boosts. And we already throttle the amount of callbacks allowed.

    With customer first callback, the callback is already there, this is when it auto dials and putting it back in queue. So you basically have to put them at the front of the queue with prioritization (assuming it is the same queue) or it is a not very good experience. And from the documentation link I shared, and Adrian copied from, it states that it expects new interactions in queue to arrive at the back of the queue not the front.

    So just feels a little like by the very intent of this new feature is in conflict with the expectations of EWT.  



    ------------------------------
    Anton Vroon
    ------------------------------



  • 8.  RE: Customer First Callback Impacts on EWT

    Posted 19 days ago

    We don't use callbacks, so their impact on EWT is not something I have ever thought about. But I think it's a great question nonetheless.



    ------------------------------
    Tim Speakman
    Tech Support Engineer
    ------------------------------