AutoResponder: AI Replies for Email and Website Chat
How we answer inbound replies from NDIS providers and migration agents in seconds, while keeping a person in charge of anything uncertain
The Problem
We run two directories, NDIS Provider Review and MigrationAgent.review, and reach out to the businesses listed in them. The replies were the valuable part, and they were piling up: the same questions about listings, claim links and calls, arriving at all hours and spread across campaigns.
Slow Replies Lose Interest
A provider who replies at 8pm and hears back the next afternoon has often moved on.
Repetitive Questions
Most replies asked whether their listing existed, how to update it, or whether someone could call them.
Answers Need Live Data
A useful reply has to look up the sender's actual listing, not paste a template.
Some Replies Need a Human
Complaints, pricing pushback and anything unusual should reach a person, not a bot.
How the AutoResponder Works
Each inbound message, from an email reply or our website chat widget, goes to a bot configured for that channel and campaign. The bot is an AI agent with written instructions, hand-off rules and a small set of tools.
Configurable Bots
Each bot has its own instructions, hand-off rules, model and tools, and is routed to specific campaigns or the website chat.
Tools That Do Real Work
Bots look up the sender's listing in the directory, create a private claim link, and book a phone call as a task in Close CRM.
Auto or Draft Mode
A bot can send on its own, or hold every reply as a draft for a teammate to approve or edit. Our migration bot runs in draft mode.
Human Hand-off by Design
When a hand-off rule matches, the bot is paused, or a run fails, the conversation moves to a person with the reason saved. The teammate can reply and hand it back.
Results in the First Three Weeks
Between 5 and 27 October 2026 the bots handled 67 live conversations. They wrote 65 of 93 replies (48 sent automatically, 17 held as drafts for approval); 21 were handed to a person by a hand-off rule and 6 failed runs went to a person automatically. On 27 October our model provider retired the model the bots used. Every run failed, every failure went to a teammate instead of the customer, and we moved the bots to the new model the same day.
Fast First Replies
Median time for the bot to write a reply, at any hour
Less Repetitive Typing
Of replies written by the bot, sent or drafted
Broken Replies Sent
During the model outage, every failed run went to a person instead
Ready to Transform Your Operations?
Leads replying faster than your team can answer? We'll set up bots that reply in seconds and know when to hand over.