What this automation does
The workflow watches for deal events, runs a short follow-up cadence in the rep’s voice, and gets out of the way the moment the prospect engages.
- Trigger on an event. Demo completed, proposal or quote sent, meeting no-show, or a deal with no activity for a set number of days starts the matching sequence.
- Check eligibility. The contact must not be on the suppression list, not already in another active sequence, and the deal must still be open.
- Draft the step. For each step, the model reads the deal’s recent notes and the previous messages and writes a short follow-up that adds something new (an answer to a question raised, a relevant example, a scheduling nudge). Steps use the rep’s approved tone and a fixed signature with unsubscribe text.
- Send at the right time. Messages go out from the rep’s mailbox as replies in the original thread, on business days, during the recipient’s working hours.
- Classify replies. Any inbound reply stops the sequence. The model sorts it into interested, not now (with a date if given), objection, wrong person, out of office, or unsubscribe, and the workflow acts on each.
- Hand back to the rep. Interested and objection replies go to the rep in Slack with a suggested response. The deal timeline records every step and reply.
When it is worth building
This is worth it when deals die quietly after the demo or quote because nobody followed up consistently. If your CRM shows lots of open deals with no activity in two weeks, that is the signal.
Do not build it to blast cold lists. That is a different job with different risks, covered in the sales email personalization guide. And if your CRM’s native sequences already stop correctly and your reps are happy with templated steps, the added value of AI drafting may not justify a custom build.
Architecture
| Layer | What we use | Why |
|---|---|---|
| Triggers | CRM deal events from HubSpot, Pipedrive or Salesforce, via n8n | Sequences start from what actually happened in the deal |
| State | A sequence table (contact, deal, step, next send time, status) in Postgres | Survives restarts and makes stop rules easy to audit |
| Drafting | Claude Sonnet 5.5 or OpenAI’s GPT models | Short, specific follow-ups grounded in notes and prior thread |
| Sending | Gmail or Outlook APIs from the rep’s mailbox, threaded | Looks and behaves like the rep’s own follow-up |
| Reply handling | Mailbox watch plus a classification step | Stops on any reply and routes by intent |
| Suppression | One shared list checked before every send | Unsubscribes from any tool are honored everywhere |
| Rep controls | Slack messages with Stop, Snooze and Edit next step | Reps stay in charge of their deals |
How to build it
- Pick two or three trigger events. Start with the ones that leak the most deals, often post-demo and post-quote. Define the steps, spacing and goal of each sequence.
- Write the guardrails. Maximum length, no new pricing or discounts, no claims not in the notes, a required call to action, and phrases to avoid (for example “just bumping this”).
- Build the state table. Store each enrollment with its next send time and status. A scheduled job picks up due steps every few minutes.
- Wire stop conditions before sending. Stop on reply from the contact or their domain, meeting booked, call logged, stage change, unsubscribe, or a rep’s Stop click. Test every one before the first live send.
- Draft, validate, send. Draft each step just before it is due so it reflects the latest notes. Validate length and banned content, then send threaded from the rep’s mailbox. For the first weeks, send drafts to the rep for approval.
- Classify and act on replies. Out of office pauses until the return date. Not now schedules a reminder task. Wrong person creates a task for the rep to follow up with the referral manually. Unsubscribe adds to suppression immediately.
- Report weekly. Reply rate per step, stops by reason, and any message sent after a reply (which should be zero).
The downloadable starter template below covers a HubSpot deal-stage trigger, the sequence state table, AI-drafted steps and reply-based stopping for Gmail.
Deliverability and compliance for follow-ups
Follow-ups go to people who already know you, which helps, but they are still judged by mailbox providers and still regulated. This is general information, not legal advice.
- Send from authenticated mailboxes. The rep’s domain should publish SPF, DKIM and DMARC records aligned with the From address. Google and Yahoo require this of bulk senders, along with spam complaint rates under 0.3%.
- Keep volume human. A rep sending forty threaded follow-ups a day looks normal. A rep’s mailbox sending hundreds of near-identical messages does not.
- Include the CAN-SPAM basics. An accurate From name, a subject line that is not misleading (a threaded “Re:” must belong to a real prior conversation), your physical postal address, and a clear way to opt out. Honor opt-outs within 10 business days, and keep the mechanism working for at least 30 days after each send.
- Make unsubscribes global. An opt-out from one rep’s sequence must suppress the contact across every rep, sequence and marketing tool. Liability under CAN-SPAM falls on the company whose product is promoted, not just the tool that sent the email.
- Respect stricter regimes. Contacts in Canada, the EU or the UK fall under consent-based rules. Filter or handle them separately.
What breaks in production
- Sending after a reply. The prospect replies from a mobile client that breaks threading, or forwards to a colleague who answers. Domain-level stop rules catch most of these.
- Double enrollment. A contact on two deals gets two sequences. Allow one active sequence per contact.
- Stale context. A step drafted days early references a question already answered on a call. Draft at send time, not at enrollment.
- Holidays and time zones. Messages land at 3 a.m. or on a public holiday. Store the contact’s time zone and keep a holiday calendar.
- Closed deals keep getting chased. A rep marks a deal won in a spreadsheet, not the CRM. Stop on stage change, and nudge reps to keep the CRM current.
- Out-of-office loops. An auto-reply triggers a reply classification that triggers another send. Detect auto-replies by header and content before classifying.
Timeline and cost
Two sequences on one CRM and one mail platform typically take one to three weeks, including a supervised period where reps approve drafts. Lumeor builds follow-up sequence automation in the range of $4,000 to $11,000, depending on the number of triggers, mail platforms and reply categories. Running costs are small: model usage per drafted step plus hosting. For the meeting-to-CRM piece that feeds these drafts, see AI meeting notes to CRM.
How to measure success
- Share of qualifying deal events followed up within one business day
- Reply rate and meeting rate per sequence step
- Messages sent after a reply or opt-out (target zero)
- Unsubscribe and complaint rate per sequence
- Win rate on deals that went through a sequence versus a prior-period baseline