Classify enquiries using AI

The problem

You get dozens of enquiries a day by email or web form. Staff spend time reading each one to figure out which service it's for and how urgent it is. You want a faster, more consistent way to triage them.

The solution

Use Claude or ChatGPT to classify enquiries. You paste in the enquiry, and the AI tells you which service it's for, how urgent it is, and whether it needs escalation. Once you've got a prompt that works well, you can process batches quickly or hand it off for automation.

What you get

For each enquiry: a service category, urgency level, and any red flags (safeguarding, complaint, press). You can do this one at a time or in batches. The classification is consistent because the AI applies the same rules every time.

Before you start

• A clear list of your services and what each one handles
• Defined urgency criteria (what counts as urgent, normal, low priority)
• A list of red flags that always need human review (safeguarding keywords, complaint indicators)
• A Claude or ChatGPT account

When to use this

• You want consistent triage without writing detailed procedures for every scenario
• Staff disagree about how to route certain enquiries and you want a tie-breaker
• You're training new staff and want to show them how experienced colleagues think about triage
• You want to test AI classification before investing in automation

When not to use this

• Every enquiry is complex and genuinely needs human judgement from the start
• You get very few enquiries and sorting them isn't a real problem
• Your categories are fuzzy and even staff disagree about routing

Steps

1. 1

   Document your current triage logic

   Write down how staff currently decide where an enquiry should go. What questions do they ask themselves? What keywords indicate each service? What makes something urgent? What always needs escalation? You're teaching the AI to think like your best triage person.

2. 2

   Create your classification categories

   Make clear lists: your service categories (with descriptions and 2-3 example enquiries each), urgency levels (with specific criteria), and red flags (keywords or patterns that always need human review). Be specific: "Mentions suicide or self-harm" is better than "seems serious".

3. 3

   Write your classification prompt

   Create a prompt that includes your categories, urgency criteria, and red flags. Ask the AI to return: service category, urgency level, any red flags spotted, a one-line summary, and its confidence level. Include 2-3 example enquiries showing how they should be classified.

4. 4

   Test on real enquiries

   Take 20-30 recent enquiries that staff have already triaged. Run them through your prompt and compare the AI's classification to what staff decided. Where does it disagree? Sometimes the AI is wrong and you need to adjust the prompt. Sometimes it spots inconsistencies in how staff routed things.

5. 5

   Refine until accurate

   Adjust your prompt based on what you learn. Add examples for edge cases. Clarify category boundaries. Keep testing until you're confident the AI gets it right at least 90% of the time. The goal is useful, not perfect.

6. 6

   Use it in practice

   Start using the prompt for real triage. You can paste enquiries one at a time, or batch several together. Keep an eye on accuracy and refine the prompt as new types of enquiry come in. Once you trust it, consider automating with the follow-on recipe.

Tools

Resources

Part of this pathway