You’re offline. This is a read only version of the page.
49
When trying to analyse trends on the topics our customers are discussing with us over Omnichannel for customer service chat widgets the ootb AI model consistently underperforms.
As the model is now embedded and no longer in the standalone Insights app there is no way to train it. Additionally as the transcripts are unable to be easily exported there's no way to manually perform the tasks without technical support.
This reduces the trust in the data and reporting provided by the Omnichannel historical analysis dashboard.
Has anyone been able to adjust or improve their results in any way?
Is anyone aware of any way to troubleshoot beyond contacting MS support who have been unable to provide any insight?
Hi Bill,
I've read the Git and unfortunately it just provides the same information that is available in the docs. Also for clarity the feature I'm specifically interested in is Topic clustering for conversations not cases.
As the process and model is managed then it is difficult to see the full process. I expect that one of the primary keys linking the 'tables' is broken or causing duplicates.
I noticed in my system the percentage was also very low. I don’t have enough data for the model to produce any meaningful topic data.
I’ll ping someone internally to see if there’s any suggestions.
Send me a PM with your company and contact info. I’ll see if I can locate a partner or CSM that’s assigned to your org for assistance (if that’s desired).
Hi Bill,
Good to hear you're getting some value from this feature.
The Historical dashboard and topic view is blank in the Customer Service Hub as that data is based on the case level entity.
However the topic view in the Omnichannel historical analytics dashboard does contain topic data. This data however is wildly inaccurate.
Over 180 days and over 20K conversations 70% of the topics are either 'undefined' or blank. The next most significant topic contains a unique identifier which signals that there is an issue with the data going in to generate the topics. It is not possible that say a 1000 conversations have all been in regards to product 1d4hyy69g (as an example).
I've had a bit of a dive into the solution data and unfortunately have been unable to isolate the problem. It did help me to follow a few conversations from transcript through to topic match. Those conversations were clearly matched incorrectly however the model is managed by MS since being removed from Customer Service Insights application and therefore there is very little detail about the input/output at a customer level.
I'll review the Git and see if that helps.
Thanks!
When you browse to the historical dashboard and view "topics" is that data providing any value?
If you look at https://make.powerapps.com and then open up Solutions, there are various Solutions that connect to Omnichannel. The model that's providing that functionality likely lives inside one of those solutions.
There is a GitHub project out there that connects to this which may give you an idea where the models live for training - github.com/.../configure-topics-clustering-cases-cs.md
#1
Muhammad Shahzad Sh...
51
Most Valuable Professional
#2
Ramesh Kumar
42
#3
David Shaw_UK
27
Share
Report
All responses (
Answers (
