AI solution design · Documents and workflows
Enquiry triage and drafting
Every email and web enquiry is read, categorised and answered with a draft reply, then passed to the right person for approval.
Talk about something like thisThe situation
A shared mailbox and a web form receive sales and service enquiries. Somebody reads each one, works out who should answer it, looks up the customer and writes a first reply. Busy days mean slow replies, and the odd urgent message gets buried.
How it works
Step by step
- 1
Read
Each new message is read and given a category, an urgency and a short summary.
- 2
Look up
The customer and any open orders or tickets are found in your systems.
- 3
Draft
A reply is written in your usual tone, using the lookup and your standard answers.
- 4
Route
The draft goes to the right person with the summary. Urgent messages are flagged.
- 5
Approve
A person edits and sends. Over time, the simplest categories can be allowed to go out automatically, if you decide that.
What it is built from
- Microsoft 365 mailbox and web form
- Where the enquiries arrive
- Azure OpenAI
- Classifies, summarises and drafts
- Semantic Kernel (.NET)
- Co-ordinates the steps and the lookups
- CRM or order system API
- Customer and order details
Safeguards
- Drafts need approval until you say otherwise
- Sensitive categories, such as complaints and legal threats, always go to a person
- Every step is logged, and a person can see why a message was categorised as it was
- The agent can only read the systems you give it, and change nothing
What you end up with
- Faster first replies
- Urgent enquiries surfaced at the top
- A log showing what was categorised, drafted and sent
Where to start
Run it in suggestion mode on one mailbox for a few weeks. You will see exactly how it behaves before it is allowed to send anything.