The Self-Service Relay

Description

GOAL
Build a process that automatically analyzes new customer requests, searches for matching FAQ articles, scores a solution probability, and depending on the result either sends a self-service reply or creates a support ticket with full context.

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 branching and the data flow.

TRIGGER
Type: new customer request by chat or email.

STEPS
1. AI step: request analysis. Use exactly the "Request Analysis" working prompt (see the Working Prompts section). Output: JSON with main_concern, keywords, category, urgency.
2. Search the FAQ: query your help center or knowledge base with the keywords and collect the top matches with links.
3. AI step: confidence scoring. Use exactly the "Solution Probability" working prompt (see the Working Prompts section). Output: JSON with solution_probability, rationale, recommended_articles, recommendation.
4. Branch: if the recommendation is "self-service" and the solution_probability is at or above the threshold (starting value 80 percent), run the Self-Service Reply AI step. Use exactly the "Self-Service Reply" working prompt (see the Working Prompts section) and send the resulting email to the customer.
5. Otherwise: create a support ticket, with the original request, the analysis JSON, the articles found and the rationale from the confidence score as context.
6. Both branches: log category, chosen path (self-service or ticket) and confidence value in the CRM.

PROMPTS
Take over the three working prompts from the section of the same name unchanged and in full.

DATA FLOW
Trigger to steps 1, 3 and 5: the original request.
Step 1 to step 2: keywords, for the FAQ search. Main concern additionally for step 3.
Step 2 to step 3: matches found.
Step 3 to step 4: recommended articles; solution probability and recommendation drive the branch.
Steps 1 and 3 to step 6: category, chosen path, confidence value.

ERROR HANDLING
No FAQ matches: create a support ticket directly (branch 5) and check keyword extraction and search parameters.
No valid JSON from an AI step: retry once, then create a support ticket as a safe default.
Unclear request or multiple concerns: create a support ticket instead of a self-service reply.
Request in another language: draft the self-service reply in the language of the request, and create a ticket if in doubt.
Sending the reply fails: check the email integration and additionally create the case as a ticket.

PLACEHOLDER CONVENTION
Map all placeholders in square brackets inside the working prompts dynamically from request, analysis and FAQ data, as described in the data flow. Not dynamic are the tone and length of the self-service reply, and the confidence threshold: set these yourself before activation, starting conservatively at 80 percent. List all placeholders that require manual configuration separately at the end.

SELF CHECK BEFORE COMPLETION
Check, before you report the process as finished:
1. Are all steps connected correctly, in particular the self-service or ticket branch and the CRM logging in both branches?
2. Are all three working prompts inserted in full and unchanged, and are all dynamic placeholders mapped?
3. Does the confidence threshold trigger correctly, and are the JSON outputs parsed cleanly?
4. Are all points listed that require manual configuration: the request channel, access to the knowledge base and help center, email sending, the ticketing system, CRM access, tone, length and the threshold?

Working prompts

You are a customer service assistant. Analyze the following customer request and extract the key information.

Customer request:
[CUSTOMER_REQUEST]

Tasks:
1. Name the main concern of the request in one sentence.
2. Extract three to five relevant search terms for the FAQ search.
3. Categorize the request (for example technical issue, billing, product question, account management).
4. Assess the urgency (low, medium, high).

Respond in the following JSON format:
{
  "main_concern": "...",
  "keywords": ["...", "...", "..."],
  "category": "...",
  "urgency": "..."
}

You are a quality reviewer for customer service. Assess whether the FAQ articles found sufficiently answer the customer request.

Customer request:
[CUSTOMER_REQUEST]

Main concern:
[MAIN_CONCERN]

FAQ articles found:
[FAQ_ARTICLES]

Tasks:
1. Check whether the FAQ articles fully address the main concern.
2. Assess the relevance and completeness of the information.
3. Assign a solution probability from 0 to 100 percent.
4. Justify your score in one to two sentences.

Scoring criteria:
80 to 100 percent: the FAQ answers the question fully and precisely.
50 to 79 percent: the FAQ contains relevant information but may not be complete.
0 to 49 percent: the FAQ is insufficient or irrelevant.

Respond in the following JSON format:
{
  "solution_probability": 85,
  "rationale": "...",
  "recommended_articles": ["Article link 1", "Article link 2"],
  "recommendation": "self-service" or "support-ticket"
}

You are a friendly customer service representative. Draft a helpful self-service reply for the customer.

Customer request:
[CUSTOMER_REQUEST]

Recommended FAQ articles:
[RECOMMENDED_ARTICLES]

Tasks:
1. Greet the customer in a friendly manner and confirm their concern.
2. Briefly explain how the FAQ articles help.
3. List two to three relevant FAQ articles with links and a short description.
4. Offer to open a support ticket if further help is needed.
5. Use a [TONE] tone and keep the reply [LENGTH].

Structure: personal greeting, short summary of the concern, two to three FAQ articles with link and short description, note on further help, friendly closing.

Respond directly with the finished email, no JSON.

JSON export