Know the moment a hot reply lands
AI reads every inbound reply as it arrives, pushes genuine buying interest straight to your phone, and files the out-of-offices where they belong.
The problem
How it works by hand
When outbound and marketing run at any real volume, the danger is not rejection, it is a buying signal buried under out-of-offices, unsubscribes and "not right now". Nobody reads two hundred replies a day with full attention, so the one that says "can you call me this week" gets the same treatment as the autoresponders. By the time someone spots it, the moment has passed.
A worked example
What a working version looks like
Every reply, across email and LinkedIn, hits an AI classifier the second it arrives. The model reads the actual content and sorts it: genuine interest, a question, a referral to a colleague, an objection, an out-of-office, an unsubscribe. Hot replies trigger a priority push notification to your phone with the prospect's name, company and what they said, so you can respond inside minutes instead of whenever the inbox next gets cleared. Everything else is filed and logged against the contact in your CRM: unsubscribes honoured automatically, out-of-offices scheduled for a later retry, objections tagged for review. The classifications are auditable, you can check what the model decided and why, and correct it.
The exact tools change per business. The shape does not.
What it needs
Honest inputs, nothing exotic
- 01Access to the inboxes and channels where replies land
- 02Your CRM, so every classification is logged against the contact
- 03A definition of "hot" for your business, we tune the classifier to it
- 04Whose phone gets the priority push
The payoff
What you get back
No buying signal sits unread overnight, however many replies the system is handling. Your salespeople stop wading through autoresponders and only ever see the messages worth a human, with the context already attached.
Do it yourself
How you would build this yourself
No course, no upsell. This is the order we would build it in, with the tools named, and a prompt to start from.
- 1
Get every reply into one stream first. Forward the sending inboxes to one address, or read them directly with the Gmail API.
- 2
Label ten real replies yourself: interested, question, referral, objection, out of office, unsubscribe. Those ten examples teach the classifier more than any clever prompt wording.
- 3
Handle unsubscribes first and automatically: into a suppression list before anything else happens. This is the step that keeps you compliant and your domain clean.
- 4
Push the hot ones somewhere you actually look: a Slack channel or a phone notification carrying the sender, company and full reply text.
- 5
Audit weekly: read the log of what the model decided, correct the misses and add them as new examples. Twenty minutes a week keeps it sharp.
Build me a reply classifier for my sales inbox using the Gmail API and the Claude API. 1. Watch the inbox for new replies to my outreach 2. Classify each one as: interested, question, referral, objection, out of office, or unsubscribe. I will paste ten real replies below with the label I would give each, use them as examples in the prompt 3. If it is an unsubscribe, add the address to a suppression list in a Google Sheet before anything else 4. If it is interested or a question, send me a Slack message immediately with the sender, their company and the full reply text 5. Log every classification and the model's reasoning to the sheet so I can audit it weekly Walk me through the Gmail API consent screen setup, that is the fiddly bit. Plan first, then build. My ten labelled replies: [paste them here]
Copy it into Claude Code, fill the brackets, and it will plan the build with you before writing a line of code.
We would rather show you how than bill you. The whole ladder of free help, answers, guides and the weekly build-along, is on the do-it-yourself page.