01What can be automated in follow-up, and what cannot
Most sales follow-up is repeated work: remembering who to contact, checking where each deal stands, writing the same kind of message with different details, logging what was said. Much of that can be automated. What is not automated is judgement: what to offer, when to push and when to let go, how to answer an objection.
That is why the approach is not «the AI sells» but «the AI prepares and a person decides». There are no promised conversion rates and no automatic closing.
02Useful triggers
A trigger is the event that starts the next step. In sales follow-up, four are the most useful:
| Trigger | CRM entry | Draft | What a person approves |
|---|---|---|---|
| New lead | Record created with source and need | Personalised first reply | Sending the message |
| Silence after contact | Follow-up task with a date | Short follow-up message | Whether and when it is sent |
| Stalled proposal | Deal stage updated | Reminder with a summary of the proposal | The tone and sending |
| Customer who has not come back | Win-back flag | Win-back message | Who is contacted and what is offered |
The table is a generic example. Real triggers and their timing are agreed with the sales team in the diagnosis.
03What is written in the CRM and what is sent to the customer
It helps to separate two kinds of action. Internal ones —creating a task, updating a stage, logging a call summary— carry little risk and can run without review if agreed. External ones —any message that reaches a customer or a lead— speak for the business and need another level of care.
04Human approval for outgoing messages
In the work I propose, outgoing messages are prepared as drafts and a person approves, corrects or discards them. Only standard, low-risk notices agreed in advance can go out without review. The approval-queue pattern is explained in the guide to internal agents with human approval.
05How it connects to the CRM, email and calendar
The automation works on the existing CRM, email and calendar, if they have an API or a reliable connector. Sometimes a flow in n8n or Make is enough; sometimes a custom integration is better. The criteria are in the guide to connecting AI to n8n, Make and your CRM. For orders and bookings rather than sales deals, the approach is in the guide to automating orders, bookings and CRM with AI.
06Limits: brand tone, sensitive data and exceptions
Drafts follow a tone guide written with the team, with examples of good messages and of what is never said. Sensitive data —special terms, financial information, complaints— is not included in an automatic draft without review. And exceptions —a customer with an open issue, a delicate negotiation— stay out of the flow and go straight to a person.
A sensible first phase covers a single trigger and is observed for a while before the scope widens. The full method is described in the AI integration service.