AI assistants
AI enquiry triage setup
When enquiries from forms, email and messages land in one place, someone has to read each and decide where it goes: sales, support, accounts, a specialist, or urgent attention. AI enquiry triage reads incoming messages, suggests a category and urgency, extracts key details and routes them to the right person or queue. Done well, it gets enquiries to the right place faster and lets staff spend time replying rather than sorting. The design questions are which categories exist, what happens when the AI is unsure, where a person must check, and whether any routing amounts to a decision about someone that needs legal safeguards.
This page covers categories, routing, confidence, human checks, automated decision safeguards, testing and review. LetGetGrow builds customer-facing chatbots and internal AI assistants, with the scope, data sources, cost and timeline set out in a custom quotation.
Illustrative diagram
- Enquiry arrives
- Classify and extract
- Confidence check
- Route or to a person
Define the categories first
Look at a few hundred past enquiries and group them: new business, existing customer support, billing, complaints, partnerships, job applications, spam. Write a short description and examples for each. Clear, non-overlapping categories make triage more accurate, whether done by a person or an AI.
Categories, urgency and routing
| Category | Urgency signals | Routes to |
|---|---|---|
| New business enquiry | Requested date, budget mentioned | Sales queue |
| Support for existing customer | Service down, cannot log in | Support, high priority |
| Billing question | Invoice reference | Accounts |
| Complaint | Words of dissatisfaction | Manager review |
| Unclear or mixed | Low confidence | Person to triage manually |
When the AI is unsure
Every triage system misclassifies some messages. Set it to route enquiries it is not confident about to a person for manual sorting, rather than guessing. Start with a cautious threshold, so more go to people, and loosen it as testing shows where the system is reliable.
Extracting key details
Beyond category, triage can pull out the company name, order or account reference, dates and a one-line summary, so the person receiving it can act quickly. Show the original message alongside, so nothing is lost in the summary.
Setting up triage
Analyse enquiries
Group past messages into categories.
Output: Category definitions
Routing rules
Where each category and urgency goes.
Output: Routing table
Configure
Classification, extraction, thresholds.
Output: Test triage
Test on past enquiries
Compare with how staff sorted them.
Output: Accuracy by category
Go live with checks
People review low-confidence and samples.
Output: Live triage and review routine
Example: triage for an IT support company
Illustrative example — not a client case study
Situation: An IT support company receives sales enquiries, support requests and invoices in one shared inbox.
- Past messages are grouped into five categories with written definitions.
- Support requests mentioning outages are routed to the on-call queue with high priority.
- Anything below the confidence threshold goes to the office manager to sort by hand.
What this shows: Urgent support reached engineers faster and nothing was silently misrouted. This example is illustrative.
Triage checks
Design
- Categories written with examples
- Routing table agreed
Safety
- Low-confidence cases to a person
- Original message always kept
Testing
- Accuracy measured per category
- Complaints and urgent cases tested
Automated decisions about people
Routing a message to a queue is usually an administrative step. But if triage effectively decides something with a legal or similarly significant effect on a person, such as rejecting a job application or refusing a service, data protection rules on solely automated decision-making apply. The ICO's draft guidance describes safeguards including a person being able to obtain human intervention and contest the decision, with a human review that can change the outcome. Keep such decisions with people, or take advice.
Reviewing triage
Have staff flag misrouted enquiries, and review a sample each week. Patterns show where category definitions need sharpening or new categories are needed.
Data protection
Triage processes the personal data in every enquiry. ICO guidance says AI use usually requires a data protection impact assessment. Check where messages are processed, retention and supplier terms, and mention the processing in your privacy notice.
Spam and abuse
Triage can filter obvious spam, but keep a way to review filtered messages so genuine enquiries are not lost.
Channels
Enquiries may arrive by web form, email, social messages and phone notes. Triage can work across channels, but start with the busiest one, prove the categories and routing, then add others. Each channel may phrase things differently, so test each separately.
Measuring the effect
Record time from arrival to first reply, how often enquiries are moved between teams and how much staff time goes on sorting, before and after triage. These show whether it is helping.
Questions about AI enquiry triage setup
What is AI enquiry triage?
Reading incoming enquiries, suggesting category and urgency, and routing them.
How are categories chosen?
From groups of real past enquiries, with written definitions.
What if the AI is unsure?
Route the enquiry to a person to sort manually.
Can it make decisions about people?
Significant decisions need safeguards; keep them with people or take advice.
How is it tested?
Against past enquiries, comparing with how staff sorted them.
Will spam filtering lose real enquiries?
Keep a way to review filtered messages.
Sources referenced
Platform features, guidance and rules change. We check the current versions before committing to an approach. This page is general information, not legal advice.
Request a Custom Quote
Ask about AI enquiry triage setup
Thank you — your enquiry has been sent.
We have received your message and will reply by email. If you need to add anything, email info@letgetgrow.com.