Genesys Cloud CX

 View Only
Discussion Thread View
  • 1.  Predictive Routing comparison test

    Posted 09-04-2022 19:20
    Hi, i ran predictive routing comparison test on couple of our queues and noticed that only a very small subset of interactions got routed as per below:
    I would have expected the numbers to be higher if the predictive routing is turned on 50% of the time and turned off 50% as per description of the comparison test. How do i really know that my handle time went down due to predictive routing or if we just had a good week?
    #Routing(ACD/IVR)

    ------------------------------
    Anish Sharma
    Datacom Connect Limited
    ------------------------------


  • 2.  RE: Predictive Routing comparison test

    Posted 09-05-2022 01:38
    I'm wondering if your  queue's Predictive Routing timeout should be extend to give more time targeting an agent based on AI routing.

    https://help.mypurecloud.com/articles/configure-timeout-for-predictive-routing/

    ------------------------------
    Michael Joseph Carreon
    Dexcom, Inc.
    ------------------------------



  • 3.  RE: Predictive Routing comparison test

    Posted 09-05-2022 17:23
    Thanks for the link. That is potentially one of the considerations we need to take however, increasing timeout would mean longer wait time for callers which could result in more abandons. I would have thought that since predictive routing goes through the historical data, it would automatically suggest an optimum timeout

    ------------------------------
    Anish Sharma
    Datacom Connect Limited
    ------------------------------



  • 4.  RE: Predictive Routing comparison test

    GENESYS
    Posted 09-09-2022 04:57

    Hi Anish

    Thanks for posting - great to get this feedback.  A few comments on this from my side.

    >> How do i really know that my handle time went down due to predictive routing or if we just had a good week?

    Per https://help.mypurecloud.com/articles/test-predictive-routing-for-your-queues/, we usually recommend that the comparison test run for at least two weeks to get enough data to be statistically significant, though you may need more at low volumes.  This is the reason we have a 90 day free trial, for you to run long enough to give confidence that there's enough data to achieve a valid comparison.

    On your point on wait time:

    • In general, we see that predictive routing can indeed increase wait time. This depends on the timeout value, but generally the average increase tends to be a few seconds.  But (when optimising for handle time) we also see that where predictive is used, the overall time the customer spends on the phone (wait + talk time) is reduced, which should provide an overall better customer experience.  The assumption here being that customers are willing to wait a little, if they're dealt with more efficiently. 
    • We have a new report in development which will show (similar to the A/B report for the target KPI) the impact of predictive routing on other metrics such as wait time, SLA and abandon rates.  See the screenshot below for a sample/preview.  You can then adjust the timeout accordingly, and understand the impact of any changes.

    And on the timeout in general:

    • The first thing to say on the timeout itself, is that if there is a customer surplus situation where customers are waiting beyond the timeout and even the SLA, then predictive routing will never come into the picture, and the split between predictive and non-predictive will always skew towards non-predictive.  In these scenarios, that's difficult to avoid.  If you have customers waiting a long time, you presumably don't then want to add to that by holding interactions even longer, or by answering interactions out of order (for example, where an agent coming available might be a better fit for an interaction which has been waiting less time in queue).
    • That said, your comment on the system suggesting a timeout is valid and should be feasible.  I've raised https://genesyscloud.ideas.aha.io/ideas/AI-I-135 to track this - feel free to vote and comment.



    ------------------------------
    David Farrell
    Genesys - Employees
    ------------------------------



  • 5.  RE: Predictive Routing comparison test

    Posted 09-26-2022 18:38
    Thanks for the response David. I have couple more questions:

    1. Is the predictive model updated over time e.g if a campaign is being run this week, the type of queries and the pool of agents who can reduce the handle time can change. Will the model update itself to factor in this pattern change?
    2. What mechanism exists to prevent the pool of agents being hammered with calls. Iam thinking of situations where you have two pools of agents, one that results in lower handle time (which predictive routing will favour) vs another pool who dont contribute as much. The pool that predictive routing favors could potentially keep those agents very busy compared to the other pool or will the predictive model apply some level of fairness to the routing?
    3. With respect to return on investment with turning predictive routing on, since there is a fee for each call being routed via predictive, is there any assurance that any call routed via the predictive engine will result in lower AHT? Iam thinking of a scenario where 80% of calls may get routed ($$) but the AHT doesnt reflect that investment.

    ------------------------------
    Anish Sharma
    Datacom Connect Limited
    ------------------------------



  • 6.  RE: Predictive Routing comparison test

    GENESYS
    Posted 09-30-2022 09:49
    Hi Anish

    1. Yes, the model is re-trained regularly (usually weekly) to take into account the latest data, changes in queue membership, performance, workloads etc.
    2. Yes, as with any target expansion mechanism, if there are agents who are regularly highly ranked and targeted first, they may get proportionally more calls.  This could be mitigated where possible by applying predictive to queues where performance is important (perhaps calls are more varied or complex) and leaving other queues with Standard routing.  In this way, agents will get 'pulled' to the queue where they can add most value, and other agents not performing as well will still get calls from other, non-predictive queues.  If this isn't feasible, we have a feature in beta which adjusts the predictive agent rankings for those agents who are most/least busy, to prevent them having too many/too few interactions https://help.mypurecloud.com/articles/workload-balancing/  The thing to bear in mind is that using this feature will 'trade off' some of the benefits of predictive routing.
    3. There are no assurances. However, we've tried to 1. make it as easy and transparent as possible to see what benefit is delivered (via the reports below), and 2. make it easy to turn on/off.  Where a queue doesn't show enough RoI, you can easily switch that queue back to an alternate routing method, and no further charges will be incurred.

    Hope that helps.

    ------------------------------
    David Farrell
    Genesys - Employees
    ------------------------------



  • 7.  RE: Predictive Routing comparison test

    Posted 10-01-2022 03:07
    Great responses. Thank You.

    ------------------------------
    Anish Sharma
    Datacom Connect Limited
    ------------------------------