Now Assist for CSM - Complaint Case Summarization

PriyamvadhaM · yesterday

Help agents triage and resolve complaint cases faster with structured, context-aware AI summaries

Complaint Case Summarization is a generative AI skill that produces structured summaries of complaint cases directly in the CSM Configurable Workspace. Unlike base Case Summarization, this skill surfaces complaint-specific context, including the product or location at the center of the complaint, related parties involved, key actions taken, and SLA urgency ranked by time remaining—so agents can understand the full complaint picture without reading through the entire activity stream.

It is especially valuable during complaint intake reviews, escalations, cross-team handoffs, or when returning to an active complaint after a period of inactivity.

Key Capabilities

Click-to-generate summaries: Agents can trigger summaries with a single button in the CSM Configurable Workspace or Now Assist Panel.
Complaint-aware output sections: In addition to Issue and SLA, summaries include a dynamic product or location section and a Related Parties section, which is context important to complaint cases.
Related Party visibility: All related parties linked to the complaint case are surfaced in the summary with their type and responsibility, giving agents an immediate picture of who is involved.
SLA prioritization by urgency: SLAs are sorted by business time remaining in ascending order, so the most time-sensitive obligations always appear first.
Integrated directly into the workspace: The summary appears inline, with options to copy, provide feedback, or save to the record.

Implementation

Install or update the Now Assist for CSM store app.
Install the Now Assist for Complaint Case (CSM) app - the Complaint Case Summarization skill ships within the Now Assist for Complaint Case (CSM) scope.
Clone and ground the Complaint Case Summarization skill for your use case using the Now Assist Admin Console or Now Assist Skill Kit within the correct application scope.

Prerequisites

Now Assist for CSM is activated
Now Assist for Complaint Case (CSM) is activated
Product or location fields populated on complaint cases for the dynamic header to render correctly
Related party records configured and linked to complaint cases if surfacing relationship data
Users have appropriate roles assigned to access the skill

Customization with Now Assist Skill Kit

Administrators can customize or ground this skill using Now Assist Skill Kit or Now Assist Admin. For broader guidance on customizing an out-of-the-box Now Assist skill in CSM, refer to this 20-minute CRM AI Academy walkthrough on our ServiceNow Community YouTube channel.

An example customization for this Now Assist Skill could be adding additional input fields to the prompt, from the associated complaint case record. For instance, the {{sn_complaint_case.cause}} field.

Navigate to Now Assist Skill Kit and ensure the correct scope is selected (Now Assist for Complaint CSM)
Open the Complaint Case Summarization skill
Go to the Prompts section
Clone the default prompt for the relevant case state variant (New, WIP, or Resolved)
Modify the cloned prompt to your requirements: for example, adding a custom complaint category section or adjusting SLA output formatting or adding an input field like mentioned above.
Save the cloned prompt
Run tests to validate the changes across product and location complaint scenarios
Enable 'Make it default' and finalize using the lock icon
Publish the skill and select which finalized prompts to include
Changes take effect when CSRs (Customer Service Reps or Human agents) open complaint cases in the workspace.

What Makes Complaint Case Summarization Different

Complaint Case Summarization generates structured summaries of complaint case records directly in the CSM Configurable Workspace. The skill processes the full record context: Case fields, Activity stream, Related list data, and extensibility fields, with prompts specifically designed for the complaint lifecycle.

Out of the Box Prompt Variants

The skill ships with three state-based prompt variants: New, Work in Progress, and Resolved. Each variant is published across many supported LLM providers. (Azure OpenAI, Amazon Bedrock, Google Cloud Vertex AI, Now LLM Generic)

The sections returned differ by case state:

New: Issue, Product or Location, Related Parties, Service Level Agreement
WIP: Issue, Product or Location, Related Parties, Key Actions Taken, Service Level Agreement
Resolved: Issue, Product or Location, Related Parties, Key Actions Taken, Service Level Agreement, Resolution

Dynamic Complaint Entity Section

One of the distinct elements of this skill is the dynamic complaint entity header: i.e., the prompt uses a <COMPLAINT_TYPE_HEADER> placeholder that resolves at runtime:

If a product is present on the complaint case, the section renders as "Product" with the product name as the value.
If a location is present instead, the section renders as "Location" with the location name.

Note: If there are other complaint types that make sense for your use case, make sure the field for it exists on your complaint case type table. In the prompt for this skill, add additional condition(s) under the <COMPLAINT_TYPE_HEADER> block.

Related Parties Section

The Related Parties section is unique to the Complaint Case Summarization skill. For every related party record linked to the current complaint case, the prompt outputs a separate entry in a JSON array in this format:

"<PARTY_NAME>: <TYPE> (Responsibility: <RESPONSIBILITY>)"

Key rendering rules:

Each related party is a separate array item: they are never combined into a single string.
If Responsibility is empty, it is excluded from the output rather than shown as blank.
If all three fields (type, name, responsibility) are missing, the section returns N/A.

Note: The skill surfaces related party data already linked to the case record and does not infer or generate relationship data. Ensure related party records are configured and associated before enabling the skill for production use.

Service Level Agreement Section

The SLA section in Complaint Case Summarization is tailored to match the urgency and importance that's usually characteristic of complaint cases. For each SLA, the output format is:

SLA: <SLA NAME>, State: <SLA STATE>, Start date: <SLA_START_TIME>

If Stop Time is available, Stop time: <SLA STOP TIME> is appended. If Breach Time is available, Breached on: <SLA BREACH TIME> is appended.

SLAs are then sorted by<BUSINESS_TIME_LEFT> in ascending order so that agents prioritize the most at-risk obligations immediately.

Key Actions Taken Section (WIP and Resolved Variants)

The Key Actions Taken section in complaint prompts includes explicit conciseness guidance.

Similar actions should be combined where possible (e.g., multiple tasks created and closed can be expressed in a single sentence).
Case task numbers must be preserved exactly as they appear in the system; do not add extra characters.
Logs, stack traces, JSON outputs, and source code are excluded from consideration.

Resolution Section (Resolved Variant)

The Resolution section follows the same anchor-based sourcing as the base Case Summarization skill. The prompt instructs the LLM to look near the end of the activity stream (<|ACTIVITY_START|>) for confirmed resolution content, which is where the most recent resolution-confirming entries appear chronologically.

Registered Inputs

Out of the box, the skill uses nine inputs:

short_description, description, priority, state, and activities are the core Case fields. activities is typed as JSON Array and carries the full activity stream with timestamps, entry types, and agent versus customer attribution.
other_fields is typed as JSON Object and enables extensibility via {{other_fields.all_remaining_attributes}}, which injects any additional case fields passed at runtime.
relatedlists_data is typed as JSON Object and enables injection of related lists via {{relatedlists_data.all_remaining_attributes}}—this is the mechanism that surfaces Related Party data.
tablename and sysid identify the record but do not appear in the prompt body directly.
sections is used only by Dynamic Sections variants.

Key Best Practices

Ensure complaint cases have product/location fields populated before the skill is invoked; the dynamic header section returns N/A if neither field is present.
Configure and link related party records before enabling the skill in production.
Do not modify the <COMPLAINT_TYPE_HEADER> placeholder syntax in cloned prompts. It is a runtime resolution token and must remain intact for product vs. location logic to work correctly.
Preserve formatting exactly, especially when referencing fields using {{}} notation. Changing or removing these breaks how the platform passes data into the prompt at runtime.
Avoid modifying the JSON output schema structure. The platform's rendering logic relies on specific section names and key formats to display summaries correctly in the workspace UI.
Do not add new sections unless the downstream rendering configuration is updated accordingly. Otherwise, additional sections may appear as raw JSON rather than formatted UI content.
Use skill cloning to remove sections not relevant to your complaint taxonomy and test across a representative sample of complaint cases before publishing.
Monitor agent feedback to improve knowledge and field completeness via Now Assist Analytics.

Formal Learning

For more learning on skills and agents in CSM, visit and enroll in our Now Assist for CSM Essentials course.

Measured Success and Outcomes

Outcome	Value to Agents	Key Metric
Faster complaint triage	Reduces time to understand complaint context, related parties, and SLA risk at a glance	Time to summary per case
Consistent handoffs	Ensures clarity when complaint cases transfer between agents or teams	% of complaint cases summarized
SLA risk visibility	Surfaces the most time-sensitive SLAs first, enabling proactive action	SLA breach rate on complaint cases
Related party awareness	Eliminates manual effort to identify who is involved in a complaint at intake	Agent time saved per case
Quality monitoring	Helps managers audit summaries and complaint resolution accuracy via feedback history	Feedback scores (positive vs. negative)

Frequently Asked Questions

Where does the summary appear? In the CSM Configurable Workspace, above the activity stream when the "Summarize" button is clicked, or through the Now Assist Panel either by indicating a specific record or by having a complaint case open in the left side of the screen for context.
What sections appear in the summary? The sections depend on case state. New cases show Issue, Product or Location, Related Parties, and SLA. WIP cases add Key Actions Taken. Resolved cases add Resolution.
Why isn't the Product or Location section appearing? The dynamic complaint entity section only renders when a product or location field is populated on the case record. If neither is present, the section returns N/A. Check that your case creation process captures one of these fields.
Can I add custom sections? Yes, via skill cloning in Now Assist Skill Kit.
Can summaries be edited? Not directly, but agents can copy the summary into work notes or comments and make changes there.
Does the feature require UI customization? No, the button and panel are available out of the box once the skill is enabled.
Do duplicate skills persist after upgrades? Yes. When using a cloned version of the skill, custom logic will carry over to future releases. If it has been further customized, additional testing is recommended post-upgrade. We recommend using Now Assist Skill Kit for prompt-level customization.
How is this different from the base CSM Case Summarization skill? Complaint Case Summarization adds three complaint-specific elements not present in the base skill: a dynamic product or location section, a Related Parties section, and SLA output sorted by business time remaining. It is purpose-built for the complaint case lifecycle.