Before we begin
Support tickets are chock full of valuable information that you can use across every department in your business. For example, we use support tickets to catalog bugs, reference recurring issues, find support article topics, and identify Fin AI (Intercom’s AI agent) success rate.
But in this specific Claybook, we’re going to show you how to import support conversations and classify the complexity of a ticket as “very easy,” “easy,” “medium,” “hard,” or “very hard.”
Scoring support tickets by difficulty might sound like overkill—but it’s actually a game-changer.
When you classify tickets from very easy to very hard, three big things happen:
- You work smarter. Harder tickets get routed to senior agents, simple ones get handled (or deflected) faster, and your team’s time is better spent.
- You spot patterns. Repeated bugs, complex features, and low-AI-deflection areas become easier to flag and fix.
- You track what matters. CSAT means more when you know how hard the ticket was. And performance benchmarks finally have context.
In short: it’s not just about labeling tickets—it’s about leveling up your entire support operation.
How it works:
- Import Intercom support conversations via webhook and HTTP API
- Clean support tickets of any irrelevant content that may cause AI to hallucinate
- Analyze customer support conversations for specific criteria
- Use AI (and our customer support inputs) to score ticket difficulty
Steps
Copy this template
First, copy this template in Clay.
If you’re already a user, you’ll be directed to your workspace. If not, you’ll be prompted to create a free Clay account.
The interactive demos on the right will walk you through each step of this Claybook.
Step 2:
Import Intercom conversations into Clay
This happens in two phases: first, connect to Intercom via webhook (column one) and import conversation IDs from Intercom; second, using HTTP API and the conversation IDs, import all support ticket transcripts into Clay too.
Step 3:
Clean support responses of any irrelevant information (blue columns)
Next, we analyze all support responses and clean them of any irrelevant information that doesn’t directly address the customer’s issue, like “Happy to help!” or “Can you send the table link?”
This leaves you with a “Cleaned Support Response” column focused only on useful, issue-solving content. It helps reduce noise and avoid AI hallucinations later on.
Next, we use formulas to calculate:
- How many cleaned messages there are
- Where that count ranks (percentile-wise) vs. other tickets
- Total word count of the convo
- Word count percentile
Add, delete, or customize the formulas and AI prompts in this step (all blue-colored columns) to fit your business and industry.
Step 4:
Analyze customer responses (yellow columns)
Next, we extracted client messages into CLIENT_RESPONSE, counted them, and ranked their volume—more messages often meant higher complexity. We then flagged frustrations, troubleshooting attempts, and topic count to identify harder tickets. Last, we used AI to infer client experience and receptiveness, which impact resolution time.
Add, delete, or customize the formulas and AI prompts in this step (all yellow-colored columns) to fit your business and industry.
Step 5:
Look for any bugs (green columns)
Next, we used only the support agent’s messages (SUPPORT_AGENT_RESPONSE) to detect bugs, avoiding false flags from clients mislabeling issues. The Bug Detection column flags the bug and shows the triggering message. If flagged, Bug Functionality reads the full convo to identify the broken enrichment and summarize what the client was doing. Splitting this into two steps reduced hallucinations and improved accuracy.
Customize the AI prompt and outputs in this step (all green-colored columns) to fit your business and industry.
Step 6:
Score ticket difficulty (violet columns)
Last, we used ChatGPT-4 and Claude 3.7 to rate ticket difficulty (Ticket Complexity 1 and 2). If their ratings were close, we averaged them. If they differed by 2+ levels—or disagreed on “medium”—we flagged it. Binary Response Complexity forced a choice between the two. If confidence was low, Ticket Difficulty Reasoning Model reviewed both explanations and chose the stronger one.
Add, delete, or customize the formulas and AI prompts in this step (all violet-colored columns) to fit your business and industry.
Copy the template
Start classifying support ticket difficulty so you can level up your entire support organization.
Simply copy this template, follow the steps above, and let Clay work its magic.
Need help?
If you have any issue, you can ask for help in Slack or schedule time with a Clay expert.
Learn with our resources
Check out Clay University, read our GTM blog, or try out our top templates to transform your growth ideas into outreach in minutes.
Clay University
Master Clay in a few core lessons. Uplevel your data enrichment, craft better outbound messages, and book more meetings.
Go to market blog
Our best content on growing your GTM motion, from outbound sales automation to AI lead generation.
Integrations
Browse through all the integrations you can connect to Clay in seconds and discover what you can do!
Clay Experts
Hire a Clay Enterprise expert or find a Clay partner to help with your next project
Clay's Chrome extension
Clay’s Chrome Extension makes it wicked easy to find leads almost anywhere on the internet