The Feedback Route

Description

GOAL
Build a workflow that retrieves new customer feedback every day, analyzes it per entry by AI for sentiment and topics, produces a structured daily report from all analyses, and automatically escalates critical cases to Slack or Teams, each with a link to the source and a recommended follow up.

Adapt the structure to the concepts of your automation tool (for example nodes, steps, zaps or agent actions). While doing so, preserve the trigger logic, the sequence, the loop per feedback entry, and the data flow.

TRIGGER
Type: time based, daily, for example at 6:00 PM.
Scope: all feedback entries from the last 24 hours from the connected sources, for example Typeform, Zendesk or Intercom.

STEPS (sequential)
1. Retrieve feedback: load all entries from the last 24 hours from the connected sources, including a link to the source. Check the date filter.
2. Loop per entry, AI step: sentiment and topics. Insert the working prompt "Sentiment and Topics" exactly as given (see the Working Prompts section), for each entry individually. Result per entry: a JSON object.
3. Aggregate: consolidate the individual results into a sentiment distribution, the five most common topics with counts, a list of cases with urgency HIGH or CRITICAL, and positive highlights.
4. AI step: report. Insert the working prompt "Report Creation" exactly as given (see the Working Prompts section), with all analyzed entries as input.
5. File and escalate: file the report in the project management tool. Additionally send cases with urgency HIGH or CRITICAL immediately to the team chat, each with a summary, a link to the source, and a recommended follow up.

PROMPT GUIDANCE
There are two fixed working prompts for the two AI steps, see the Working Prompts section. Insert both unchanged and in full into step 2 and step 4.

DATA FLOW
Step 1 to step 2: a single entry as input for the feedback text placeholder in the working prompt, per loop iteration.
Steps 2 and 3 to step 4: all analysis results as input for the analyzed feedback placeholder in the working prompt.
System to step 4: the current date and the reporting period as input for the corresponding placeholders.
Step 2 to step 5: entries with urgency HIGH or CRITICAL, including a link to the source.

ERROR HANDLING
No feedback in the period: send a short notice instead of an empty report and check the date filter.
AI returns no valid JSON: retry the affected entry once, then include it in the report marked as unclassified.
Irony or sarcasm misclassified: add examples to the working prompt and spot check the classification.
Too many escalations: sharpen the threshold, for example only at CRITICAL or a combination of urgency and negative sentiment.
Multilingual feedback: detect the language beforehand. The analysis works independently of language, the report stays in the team's language.
High volume: run the loop in batches and observe rate limits.

PLACEHOLDER CONVENTION
Two kinds of placeholders occur. Placeholders in curly braces are dynamic: map {FEEDBACK_TEXT}, {ANALYZED_FEEDBACK}, {CURRENT_DATE} and {PERIOD} as described in the data flow. [NUMBER_OF_RECOMMENDATIONS] is not dynamic: set this value yourself before activation, for example 3. The square brackets in the JSON output format of the first working prompt are format instructions to the model and stay unchanged. List all placeholders that require manual configuration separately at the end.

SELF CHECK BEFORE COMPLETION
Check, before you report the workflow as finished:
1. Does the analysis run as a loop per entry, and are both outputs served, the report filing and the escalation?
2. Are both working prompts inserted in full and unchanged, and are all dynamic placeholders mapped correctly?
3. Does the escalation trigger only at urgency HIGH or CRITICAL, and does it include a link to the source and a follow up recommendation?
4. Are all points listed that require manual configuration: feedback sources, the report's storage location, the escalation channel, the escalation threshold, [NUMBER_OF_RECOMMENDATIONS], and the LLM access?

Working prompts

Analyze the following customer feedback and provide a structured assessment:

FEEDBACK:
{FEEDBACK_TEXT}

TASKS:
1. Sentiment analysis: Rate the emotional tone as POSITIVE, NEUTRAL or NEGATIVE.
2. Urgency: Estimate the urgency (LOW, MEDIUM, HIGH, CRITICAL).
3. Topics: Identify one to three main topics (for example product quality, customer service, delivery, price, features).
4. Summary: Summarize the feedback in one to two sentences.

If details are missing or ambiguous in the feedback, mark your assessment as an assumption instead of presenting it as certain.

OUTPUT FORMAT (JSON):
{
  "sentiment": "[POSITIVE/NEUTRAL/NEGATIVE]",
  "urgency": "[LOW/MEDIUM/HIGH/CRITICAL]",
  "topics": ["Topic 1", "Topic 2"],
  "summary": "[Short summary]",
  "original_feedback": "{FEEDBACK_TEXT}"
}

Create a professional daily feedback report based on the following analyzed entries:

ANALYZED FEEDBACK:
{ANALYZED_FEEDBACK}

REQUIREMENTS:
1. Executive summary: overview of the number of feedback entries and the overall sentiment.
2. Sentiment distribution: percentage breakdown of positive, neutral, negative.
3. Top topics: the five most common topics with the number of mentions.
4. Critical feedback: list of all feedback with urgency HIGH or CRITICAL.
5. Positive highlights: two to three particularly positive feedback entries.
6. Recommended actions: [NUMBER_OF_RECOMMENDATIONS] concrete next steps based on the findings.

DATE: {CURRENT_DATE}
REPORTING PERIOD: {PERIOD}

Format the report clearly with markdown structure (headings, lists, highlights). Mark assumptions where the data basis or context is insufficient for a clear statement.

JSON export