The problem this solves
Replies are where a campaign turns into pipeline, and they are the part of the process that most often falls into a gap. The sequence tool holds the replies; the CRM holds the contacts; the rep reads the replies when there is time, logs some of them, and the interested prospect who answered on Tuesday hears back on Friday. Referrals - 'talk to my colleague' - are the most valuable replies and the most often lost, because they need a new contact created rather than a field updated.
Classifying replies is a reading job with five outcomes, and each outcome has a different next step. The Personalization Agent reads every reply in the campaign, decides which outcome it is, logs it on the contact in HubSpot so the record is true, and posts the interested ones to #sales the same hour with a suggested next step drafted from the reply. Referrals produce a new contact with the introduction quoted; unsubscribes produce a do-not-contact flag before the next step in any sequence can reach them.
How the mission runs
- Read the replies from campaign <name>. The Personalization Agent reads every reply received in Lemlist campaign <name> since the last run, with the original step it answered and the contact it came from, and separates human replies from auto-responders.
- Classify each reply. Each reply is classified as interested, objection, referral, later or unsubscribe, with a one-line reason drawn from the text. Ambiguous replies are marked for a human read rather than forced into a category.
- Log the outcome in HubSpot. The classification, the reply date and a short summary are written to the contact in HubSpot as a property and a timeline note. Unsubscribes set the do-not-contact flag; referrals create the referred person as a new contact with the introduction quoted and linked to the original.
- Post the interested replies to #sales. Interested replies are posted to #sales one by one with the contact, the company, the reply quoted and a suggested next step - a meeting time, a specific answer, a resource - written from what the reply asked for, so the owner can respond within the hour.
The prompt
This is the exact objective the agent receives. Swap the obvious placeholders for your own domain, segment or channel and run it as-is from the console, Slack, or the API.
What comes back
Every reply from Lemlist campaign <name> classified and logged on the HubSpot contact with a summary, unsubscribes flagged do-not-contact, referrals created as new contacts with the introduction quoted, and the interested replies posted to #sales with a suggested next step each. The summary shows the reply mix by category and the objections most often raised, which is the campaign's real feedback.
Make it yours
- Run it hourly during an active campaign so interested replies never wait.
- Read replies from Instantly or Apollo instead of Lemlist, or from several campaigns at once.
- Draft the full reply to each interested prospect, staged in Gmail, rather than a next step only.
- Post objections to #product or #marketing as a weekly digest of what the market said.
Frequently asked questions
How accurate is the classification?
High for clear replies and honest about the rest: anything ambiguous is marked for a human read rather than guessed, and the reason line lets you check each classification quickly.
Does it pause the sequence for people who replied?
Lemlist usually stops a sequence on reply; the agent confirms that the contact is stopped and flags any still active, and it never restarts a sequence.
What happens with out-of-office replies?
They are recognized as auto-responders and logged as such with the return date where stated, so a follow-up can be timed for when the person is back.
Can it read replies from several campaigns at once?
Yes. Name the campaigns or say all active ones, and each reply is logged with the campaign it came from so the reply mix can be compared across them.