Use customer sentiment to guide thoughtful support follow-up

Customer sentiment is an AI-derived signal about the tone of eligible customer activity. Use it to find messages that deserve closer attention, prepare an account review or identify a pattern worth investigating. In B2B support, a score becomes useful when you connect it to the customer organization, business impact, ongoing cases and the next commitment. Read the original message before acting: a sentiment score is an interpretation, not a customer survey answer or a verified prediction about the relationship.

Distinguish sentiment from other customer signals  ·  Read the current activity score bands  ·  Check generation and display prerequisites separately  ·  Review a scored customer message in its case context  ·  Build a focused review list with supported sentiment fields  ·  Worked example: repeated equipment issues create frustration  ·  Use sentiment responsibly in a supervisor or account review  ·  Troubleshoot missing or unexpected sentiment

Distinguish sentiment from other customer signals

A polite message can describe a serious production outage. A frustrated message can concern a lower-impact issue with a poor communication history. Use sentiment to notice something worth reading, then apply your actual service policy. Do not use it as the sole basis for evaluating an agent or assuming a customer will leave.

Read the current activity score bands

Stored activity sentiment uses a 0–100 scale. The Case activity display uses the following bands. These describe the current activity-label presentation; view filter pickers can use their own descriptive scale labels, so use the numeric boundaries when you need an exact range.

A label is not a precise measurement of a person’s feelings. Sarcasm, technical terminology, quoted conversation, terse writing and multilingual context can affect an automated interpretation. If the score conflicts with the message, act on the verified operational facts and keep the uncertainty visible.

Check generation and display prerequisites separately

1. Confirm you can access the intended case and customer Activity. Sentiment does not expand case visibility.
2. Have the authorized administrator review Settings → Account → Feature Modules → Health Scoring when activity sentiment indicators are absent. Health Scoring gates the current display; enabling a display module alone does not guarantee that a score will be generated.
3. Review the effective AI Configuration for the case’s Division, including the applicable global fallback and enabled automatic customer-activity enrichment. Confirm the account has the required AI entitlement and processing capability.
4. Check whether the customer Activity type participates. Custom types require the applicable sentiment setting; outgoing replies and internal notes should not be assumed to follow the same customer-scoring path.
5. Use a representative eligible customer Activity with usable content to verify the actual outcome. Some processing is immediate in a particular intake path, while other work is scheduled and can retry. Allow for processing and investigate failures rather than promising universal real-time coverage.

Review a scored customer message in its case context

1. Open the correct case and find the relevant customer Activity. Confirm the author, time and whether it is the latest message or an older point in the conversation.
2. Read the original message first, then inspect its sentiment and emotion indicators where available. Identify the customer’s stated problem, requested action and any explicit deadline.
3. Read enough earlier activity to understand promises and repeated questions. Check the current case owner, queue, priority, unresolved tasks and escalation state.
4. Review the Company and Contact context that you are authorized to use. Determine whether this is one isolated concern or part of several related cases; do not assume multiple Contacts at one Company have identical priorities.
5. Separate confirmed facts from the AI interpretation. Record the actual issue and next action, such as “customer reports repeated missed updates,” rather than recording an unsupported claim about their feelings or renewal intent.
6. Choose the appropriate action: a clear customer reply, an internal specialist task, a supervisor review or a formal escalation. Confirm the responsible owner and next-update time.
7. After acting, verify the saved case activity and any task/assignment. A score changing later is not proof that the promised work or customer outcome was completed.

Build a focused review list with supported sentiment fields

1. Define the review question, for example “Which eligible customer messages in this team’s current work need a communication review?” Choose the appropriate view type and permitted scope.
2. For an Activity view, use the available Customer Sentiment Score column and Customer sentiment from, Customer sentiment to or Customer sentiment between controls. A Case view uses its available Sentiment field and case-level interpretation.
3. Add useful context columns such as the available case identity, customer, owner, date and activity type. Keep enough context to open the underlying record and verify why it matched.
4. Use a deliberate numeric range and time window. Do not mix an Activity’s own score with a related Case’s aggregate merely because their column labels sound similar.
5. Save the view, find a known matching record and check a known excluded record. Inspect blank/unscored values separately rather than treating them as neutral feedback.
6. Open a small representative set of results and decide on actual support actions. Review the list’s scope and data availability before sharing counts with management.

Company and Contact views can expose Sentiment and Recent sentiment. Recent sentiment uses a 30-day case-created cohort; it is not automatically “all messages received in the last 30 days.” Likewise, a survey view’s Case Sentiment refers to related case sentiment, not a new survey answer. State the population and date basis alongside a report.

Worked example: repeated equipment issues create frustration

Suppose a customer Contact writes that this is the third interruption and they have already supplied the requested logs. A negative score may help the message stand out, but the operational concern comes from the text and case evidence.

1. Check the current interruption’s impact, affected equipment and customer deadline. Confirm what was actually supplied and avoid requesting the same information again without explaining why.
2. Read the related Company history and authorized cases to determine whether the incidents share a verified cause. Similar wording alone does not establish one recurring defect.
3. Review the last promised update and whether it was met. Ask the specialist for a concrete diagnostic next step while keeping one owner accountable for the customer reply.
4. Reply with confirmed facts and a specific next-update time. Acknowledge the repeated effort without presenting an unverified root cause or release promise.
5. Escalate through the agreed process if the business impact or stalled progress warrants it. Record evidence, receiving responsibility and the communication plan.
6. Once the solution is verified, improve the relevant knowledge article and check that the customer can access it. In the next account review, assess the actual reliability and follow-through, not only the sentiment trend.

Use sentiment responsibly in a supervisor or account review

1. Choose comparable customer cohorts, activity types and periods. Record how many relevant records are actually scored and how many are missing.
2. Read examples from both concerning and positive results. Check whether a handful of messages or one long-running case dominates the apparent pattern.
3. Compare the score with customer comments, SLA/service evidence, current commitments and survey feedback. Keep those measures distinct rather than inventing a single satisfaction percentage.
4. Identify a process improvement the team can verify, such as clearer handoffs, fewer repeated evidence requests or dependable update ownership.
5. Assign an owner and review point. Reassess the underlying customer outcomes as well as the sentiment signal after the change.

Use individual examples for coaching with context and care. The model may interpret the customer’s product frustration rather than an agent’s service. Automated scores alone are not a fair account of agent performance.

Troubleshoot missing or unexpected sentiment

Continue learning

• Review Feature Modules and display availability
• Configure the effective AI policy
• Configure Activity types and eligible behavior
• Build and verify an operational view
• Choose field-specific view filters
• Review Company health with supporting evidence
• Manage a clear escalation handoff
• Interpret actual customer survey feedback