Happy Falloween!! It’s definitely all Treats here, no Tricks. Here is all the goodness rolling out this month.
Major new features
- New Sentiment score and Needs Attention score
Alerts on Keywords
More Escalation Value charts and tables
Filters alignment across several pages
Automated retraining of ML models for Escalation prediction
Improved detection of Churn Risk and Follow-up signals
Clearer Signal count alignment across cards
Improved score aggregation
The SupportLogic Team
New Sentiment Score and Attention Score
- New scoring - Our improved scoring system will bring more variation to our product's scores. Key changes include:
- Needs Attention scores now consider case priority, not just case interactions.
- Negative sentiment impact on Sentiment scores decays gradually, reflecting case closure.
- Needs Attention scores also decay slowly for urgent cases.
- Positive sentiment in outbound and case notes has reduced influence on Sentiment scores.
- Non-English tickets are filtered out for greater score variance.
- More participants in cases now affect scores, addressing customer preferences
Better visualization and explanations of factors - Ever wondered why the scores dropped or increased for a case. You will now be able to answer that directly from the case details page. When hovering over the scores for a case, this informative graph will show both the trend line scores of the case, as well as key factors that led to the overall score computation.
Alerts on Keywords
Escalation charts and table - Likely to Escalate trends,
Escalation Review Activity table
Predicted Escalations Trend line chart - Following up on last month’s addition, we are adding another chart to the Escalations tab under Operational Metrics, this one will let you visualize the trend line of Escalation Predictions, as well as the percentage of cases that were predicted. This helps connect the dots end-to-end, starting with cases predicted to escalate, cases that receive an escalation request and cases that eventually escalate.
Top 10 Agents by Escalations - We are also adding a table that shows the Top 10 Agents ranked by the number of predicted escalations. This will be helpful to identify the agents who need help managing their cases better, since they're susceptible to more escalations.
Escalation Review Activity table - Finally, we are also adding an Escalation Review Activity table that helps tie all of this together. This table will show the Top users, who are reviewing and actioning on the predictions we make.
You will also be able to search for particular users and review how actively they have been actioning on predicted escalations. By leveraging this along with the team level filters, the entire end to end picture can be painted. Eg. Manager A is actively using the product and actioning on all the predicted escalations, and that’s reflective of their team’s overall Escalations, Escalation Requests going down.
Filters - various improvements site-wide
Global filters are now supported on the My Agents page. With this improvement, if you have global filters applied, they will also be honored on the My Agents page, and the list of agents and cases they are working on will be filtered to that view alone.
Filtering for cases with Engineering Issues as well as Unassigned cases is also available now. These are available both on Global Filters (site wide) as well as Dynamic Filters (page specific).
Creating a new case list on the Backlog page has a whole new guided flow, making it easy to setup lists that you are interested in.
Additionally, it’s now possible to use the entire Customers filter and Agents filter in each individual list on the Backlog page. This makes that page extremely powerful, since each list can be customized to suit your needs. As an example, it’s now possible to setup Backlog lists specifically looking at Top Tier Accounts, or a subset of Agents, or combining both.
The calendar filters on each Customer and Agent insights page will now remember and persist the date ranges selected earlier.
Automatic retraining of Escalation Prediction ML models
We continue to improve our Escalation Prediction model performance regularly. A new model training pipeline has been launched which retrains every model on a monthly cadence, evaluates the output of the new model compared to the existing versions and automatically promotes the new model if meaningful improvements are detected.
Since launching last month, we have already seen improvements of around 8-10% better recall rates for multiple customers, and as much as 90% improvements for some instances.
Increased detection of Churn Risk and Follow-up signals
With continued model improvements, we are seeing significant improvements to these key signals.
Churn Risk signals detected have increased by 130% month over month. Additionally, the Recall rate has been improved to detect 3.8x more signals; while reducing the number of false positives by over 47%.
Follow-up signals detected have increased approximately 13% for all customers. Additionally, the Recall rate has been improved to detect 74% more signals; while reducing the number of false positives by over 75%.
Improved Score aggregation
When analyzing data, users look for trends to identify potential issues. We have updated the calculation method on the Experiential Metrics page to improve clarity. The new method gives more importance to scores in the highest/lowest bins, making important variations more noticeable. This allows for a focused examination of score changes.
Sentiment Score Aggregation Changes
Additional factors on the Agent Insights section for Escalation Review
To improve the accuracy of our predictions, we are adding two more agent specific insights to our model as well as show it under the Agent Insights section when reviewing an escalation:
Case count - number of cases the agent is currently working on
Escalation count - number of escalation, the agent is currently involved in
Was this article helpful?
Thank you for your feedback
Sorry! We couldn't be helpful
Thank you for your feedback
We appreciate your effort and will try to fix the article