Customer Support Triage
Classify and route customer messages with JEV
Separate plan inquiries, integration questions, and account requests. JEV returns categories and probabilities that your help desk can use to assign the next step.
- Intent classification
- Queue routing
- Human review
We plan to integrate your API next month, with around 100,000 requests per month. Which plan would fit?
JEV · Inquiry category
- Plan inquiry94%
- Integration question5%
- Account request1%
Use cases
A topic for routing. A score for follow-up.
Classify by your team’s queues
Define the categories your help desk already uses. A question about choosing a plan can go to sales; a question about an API error can go to technical support.
Identify concrete buying intent
Evaluate readiness separately from topic. A message with a use case, launch date, and expected volume gives your team different follow-up context from a general inquiry.
Keep ambiguous messages for review
Use the returned probabilities in your routing rules. Messages that fit several categories can stay in a review queue instead of being assigned solely by the top label.
WORKED EXAMPLE
One inbox, three kinds of customer inquiry
A shared inbox receives plan questions, onboarding requests, and account administration messages. They may all mention “API access,” but the next step depends on what the customer is trying to do, not just a keyword in the text.
Define the categories around the teams that handle the work. Send the message to JEV with a choice question about its topic, then let your application use the returned classification and probabilities in its assignment rules.
| Example message | Topic to identify | Possible next step |
|---|---|---|
| We expect 100,000 calls a month. Which plan fits? | Plan selection | Sales consultation |
| Can you help us make our first API request? | Integration guidance | Technical onboarding |
| How do I invite a colleague to our team? | Account administration | Account support |
Use messages your team has already handled to test these categories. Compare the suggested route with the team’s actual assignment, then refine the descriptions where they disagree.
- CUSTOMER MESSAGE
Read the inquiry
- “Which plan fits our team?”
- Launch: next month
- Volume: 100,000 calls / month
- JEV
Ask separate questions
- Choice → Inquiry topic
- Score → Readiness
- Noul → Needs onboarding?
- YOUR HELP DESK
Apply your rules
- Assign a team
- Include follow-up context
- Review unclear cases
How to build it
From incoming message to the right queue.
- 01
Define your routes
List the queues your team uses and describe which inquiries belong in each one. Test the criteria on representative messages.
- 02
Evaluate the message
Send the message as context to JEV. Ask for a choice of queue and, when useful, a separate score or yes-or-no judgment.
- 03
Apply your routing rules
Let your application route clear results. Send ambiguous or consequential cases to someone who can review them.
BEYOND THE FIRST LABEL
Give each team the context it needs to follow up
Describe the distinction between neighboring queues
“I need help choosing API access” and “I need help using my API access” are different requests. Describe sales as plan selection and purchasing, and technical support as using or integrating the API. Clear category boundaries give the model a more useful task than broad department names alone.
Include the customer’s relevant context when it is already available to your application, such as the current message and the previous question in the conversation. Keep the classification focused on the request that needs a response now.
Score readiness separately from topic
A plan inquiry can come from someone browsing options or a team preparing a launch. Use a separate score question with levels such as general interest, a defined use case, and a use case with timing and expected volume. The category still selects the queue; the score adds context for follow-up.
A yes/no probability can answer another narrow question, such as whether the message asks for onboarding help. JEV, a System One model, lets you ask these typed questions over the same context without requiring the customer to choose a complicated form category first.
Review decisions against how the team actually works
Try examples from each queue, including short questions and messages with more than one request. Compare the suggested categories with the team’s own assignments, then refine descriptions that repeatedly overlap.
Use the probabilities to choose where an automatic route is useful and where a person should review the message. Your help desk retains control of assignments and replies. Reviewing misrouted examples can also reveal where your own queue definitions need clarification.
FAQ
Customer Support Triage questions.
Does JEV send replies or move tickets itself?
No. JEV evaluates the context and returns a classification, probability, or score. Your help desk or application decides whether to assign a queue, ask for more information, or draft a reply.
Can I ask more than one question about a message?
Yes. You can evaluate the topic, purchase intent, and follow-up priority as separate questions over the same message. Keep each question tied to a distinct decision your team needs to make.
What should happen when a message fits two queues?
Compare results against examples from your own inbox and set thresholds for automatic routing. If a result is unclear, keep a human review path rather than treating the highest probability as a guarantee.
How can I test this before connecting my help desk?
Try representative messages in the JEV Playground. Define your categories, inspect the answers, and export the request once the results fit your workflow.
Get started
Test your first routing decision.
Bring a real customer message to the JEV Playground and see how it fits your team's queues.