Skip to main content
How tickets are analysed

Sentiment and tonality alerts

How QuantumOps scores the mood of a ticket conversation, when it raises an alert, where alerts appear and how to give feedback on one.

Written By Chris Scaminaci

Last updated About 2 hours ago

QuantumOps scores how the customer and the support team sound in a ticket conversation. When a score is low enough it raises an alert, so a manager or a customer success manager can step in early. This page explains the scores, when an alert is raised, where alerts appear, how to read the message in Slack or Teams and how to give feedback. Administrators also find the settings that tune alerts.

Before you start: anyone can read an alert. Acting on one happens in the Customer Success Center, for people with the Administrator, Service Desk Manager, Dispatcher or Customer Success role, when the Customer Success module is on. Tuning alerts needs the Administrator or Q-Director Configuration role.

What the two scores measure

Each ticket gets two scores from 1 to 10. A score of 1 is the worst and 10 the best.

  • Sentiment describes how the customer comes across. Its category runs from Hostile, Frustrated and Disappointed through Neutral to Cooperative and Pleased.
  • Tonality describes how professional and cooperative the exchange is. Its category runs from Abusive, Unprofessional and Strained through Neutral to Professional and Excellent.

The analysis also gives a Severity, a Risk Category, key phrases, warning indicators, recommendations and whether escalation is required. The ticket's own dialog shows all of them. See Ticket details, sentiment and sanity scores.

QuantumOps scores a ticket after a customer-visible update, when it analyses the ticket and when the ticket closes. It reads the latest update against the earlier conversation. Private agent notes are ignored when judging the conversation, and a message that was already scored is not scored twice. Excluded ticket types and clients are never scored.

When an alert is raised

A ticket alerts when its sentiment or tonality score is at or below the alert threshold. Both thresholds are 5 until an administrator changes them. A higher threshold gives more alerts.

  • Per client or contact. A calibration can move the thresholds for one client or one contact by up to two points, either way. See Tune alerts.
  • Once per new alert. If an active alert for the ticket was raised in the last four hours, a new low score only refreshes that alert. It does not send a second message.
  • Which type. When both scores are low, the alert counts as a sentiment alert. The alert type reads Negative Sentiment or Unprofessional Tonality.

Where alerts appear

  • On Home. The Alerts column of the ticket list, the alert counts on the Agent workload card and the Sentiment Alert Active and Tonality Alert Active badges in the sentiment dialog. See Home: the QA Pipeline.
  • In HaloPSA. The Ticket Dashboard status bar reads ATTENTION REQUIRED with Sentiment and Tonality badges; see Read the status bar. The Client, User and Agent dashboards count alerts in their Alerts or Active Alerts figures; see The Client Dashboard, The User Dashboard and The Agent Dashboard. The Ticket Triage Dashboard weighs alerts in its priority score; see Read the priority score.
  • In the Customer Success Center. The alert feed lists the ticket so that someone with access can dismiss it, silence it, e-mail the customer, send a CSAT survey or create a Customer Success ticket. See Working the alert feed.
  • In Slack or Teams. QuantumOps posts one message per alert to the alert channel that an administrator set up in Q-Director, on the Messaging tab. See Q-Director: Slack, Teams and agent mapping.
  • As a Q-Notice with buttons. This is optional. See Get acting buttons in Slack or Teams.

Read the alert message in Slack or Teams

The message is headed with the alert type, for example Negative Sentiment Alert, with an emoji for the severity. It shows:

  1. The ticket number, the time (in UTC), the client, the user and the subject.
  2. Analysis Scores: the sentiment and tonality scores out of 10, each with its category.
  3. The Severity and the Risk Category. The Teams card heads the second field Risk.
  4. The AI's Analysis, the Key Phrases it found (Key Phrases Detected in Teams), the Warning Indicators and the Recommendations.
  5. ESCALATION REQUIRED, with the kind of escalation, when the score calls for one.
  6. Two link buttons: View Ticket opens the ticket in HaloPSA and Open in Customer Success opens the Customer Success Center.
  7. When feedback buttons are on, Was this analysis helpful? with Good Analysis, Poor Analysis and Exclude Similar.

A message posted to a Slack or Teams webhook can only carry links, not buttons that act on the alert. Use the Customer Success Center to act, or add the Q-Notice below.

Get acting buttons in Slack or Teams

A Q-Notice can post the same alert as a Customer Success card with buttons to Reprocess, Dismiss, Silence, send a CSAT survey or create a Customer Success ticket. Adding the rule needs the Administrator or Q-Notice Admin role.

  1. Open the Q-Notice Center from the sidebar and go to the Auto-Rules tab. See Q-Notice auto-rules.
  2. Select New from template: Sentiment alert card.
  3. Review the rule that opens. It listens to the event Sentiment / tonality alert raised and posts to Slack and Teams. Check the destination and the conditions, then create the rule.

Nothing is created until you save the rule. See Acting on alerts from Slack and Teams for how the card works.

Give feedback on an alert

The feedback buttons open the Sentiment Analysis Feedback page in a browser. The page needs no sign-in, so anyone who has the message can use it.

  1. Select Good Analysis, Poor Analysis or Exclude Similar in the message.
  2. The page shows Alert Details (ticket, client, alert type and time) and the Analysis Scores.
  3. Fill in what applies:
    • After Poor Analysis, use What should the scores have been? (Optional) to enter an Expected Sentiment Score and an Expected Tonality Score, from 1 to 10.
    • After Exclude Similar, tick Exclude all tickets in this category or Exclude all tickets from this client.
    • After any of them, add Additional Comments (Optional).
  4. Select Submit Feedback. The page says Thank You!.

If the link is no longer valid, the page says Alert Not Found.

Feedback is recorded against the alert. It does not change scoring by itself. The Customer Success Center counts the Exclude similar requests, helpful ratings and not-helpful ratings on its Settings tab. A request to exclude a category or client is recorded for review, not applied. An administrator adds the exclusion under Exclusions in Q-Director, on the Classification tab, in Sentiment & Tonality Analysis: Excluded Ticket Types and Excluded Clients.

Tune alerts

These settings are in Q-Director, on the Classification tab, in Sentiment & Tonality Analysis. Save with Save Changes. See Q-Director: sentiment and categorisation.

SettingWhat it does
Sensitivity LevelLow gives fewer alerts, Medium is the default and High flags subtle concerns. It changes how the AI scores, not the thresholds.
Sentiment Alert Threshold (1-10) and Tonality Alert Threshold (1-10)Alerts fire at or below the value. The default is 5.
Custom Analysis InstructionsYour own guidance for the AI, for example password reset tickets should never trigger sentiment alerts.
Excluded Ticket Types and Excluded ClientsTickets that are never scored.
Enable Feedback Buttons in AlertsAdds the feedback buttons to the Slack and Teams messages. On by default.

Client & Contact Calibration is a separate card below them. Select Add client or contact, pick a Client, leave Contact (optional) empty to cover every contact of the client or pick one contact, and set a Sensitivity from Much less sensitive to Much more sensitive. Each step moves both thresholds by one point. Add a Baseline note for the AI to say what is normal, such as always terse; do not flag brevity. A note never hides threats or abuse. When a contact and its client both have a row, the contact's Sensitivity applies and both baseline notes are used. Save with Save calibration.

Sentiment alerts on call recordings are separate. See Call recordings.