AI reply drafting: agents edit, they don’t write from scratch
An experienced agent knows what to answer. A newcomer spends 10 minutes searching the documentation, then 10 minutes writing the reply. 20 minutes per ticket. AI reduces that to 4 minutes — the agent just reviews, updates the customer name and hits Send.
How AI generates the draft
Ticket context
Ticket #2245 — Kovospol s.r.o.
Agent: Mária K.
Customer text: "The April invoice has an incorrect VAT ID.
Invoice number FAP-2026-0124."AI suggests a reply
AI DRAFT REPLY:
"Dear Sir/Madam, thank you for letting us know.
We have verified invoice FAP-2026-0124 — the VAT ID was
indeed recorded incorrectly [AI note: please insert correct VAT ID].
We will issue a credit note and a corrected invoice
within 24 hours. The documents will be sent to
info@kovospol.sk.
We apologise for the inconvenience.
Kind regards,
[agent name]
Modulario Support"
Source: KB article #127 "Correcting billing details"
AI confidence: 89 %
[Send] [Edit] [Reject and write manually]Contextual auto-fill
AI automatically populates:
- Customer name from CRM record
- Contact email from CRM
- Invoice number from the ticket text
- Relevant KB article as a link
The agent just checks accuracy and clicks Send.
Building the knowledge base from overlooked tickets
When an agent writes a reply that AI did not suggest, the system proposes adding it to the knowledge base. Over time the pool of replies grows and AI drafts become more accurate.
