CRM Automation Rules: Triggers, Actions and the Autonomous AI Step
Almost every small firm has the same gap in its week: a customer calls while two people are on site, the call is missed, and by the evening nobody remembers to call back. Nothing here is a discipline problem. It is a timing problem, and timing is exactly what software is good at. The Automation section in Firmspace CRM is built as a simple sentence: when this happens in the CRM, do that. The sentence can wait an hour or three days before it acts, and one of the things it can do is hand the situation to the AI assistant.
What starts a rule are events, not schedules. Firmspace offers eighteen of them, and they are grouped by the part of the business they belong to. Deals: a new deal appears, or a deal moves into a chosen stage. Contacts and inbound: a contact is created, a lead comes in from an incoming e-mail, a message arrives in a connected channel, or a call is missed. Tasks: a task is created, or a deadline is approaching. Finance: an order is created or changes status, a quote is sent, a contract is signed, an invoice is sent, paid, becomes overdue, gets a credit note, is created as a recurring invoice, or is reminded.
Between the event and the work stands a delay you choose: immediately, after one hour, after one day or after three days. The delay is what makes a rule feel human instead of robotic. A missed call does not need a reaction in the same second, it needs a reaction before the end of the day. A quote that has not been answered for three days deserves a friendly nudge, and it is much easier to configure that once than to remember it every evening. One detail about the deadline event is worth knowing: it fires in the browser of a person who has the CRM open, so it is a reminder for people who are working, not an alarm clock for an empty office.
A rule can then do one of seven things: send a notification to the team, create a task, add a note to the history of the deal or contact, move the deal to another stage, send an e-mail from the connected mailbox, call an external address by webhook, or let the AI decide. The first six are predictable by design, which is their value: a notification to the sales person with the deal name in the text, a task named after the customer with a responsible colleague already chosen, a note that says what happened and when. The seventh is the interesting one, and it deserves its own paragraph.
The AI action step works differently from everything else on the list. You write an instruction in ordinary words, for example: look at this missed call, find out who the person is in the CRM and create a follow-up task for the responsible colleague. When the rule fires, the assistant reads the event together with the data attached to it, may look things up first, and then performs one write action that fits the instruction. It cannot perform two — a second write call in the same run is skipped by design, which keeps an automatic step from turning into an uncontrolled chain of changes.
Because this step acts while nobody is watching, the safeguards are part of the feature rather than an afterthought. No confirmation is asked — the owner agreed to autonomous execution by enabling the step. The instruction field carries a visible warning that says exactly this. The assistant works with the rights of the firm owner, but never sees more than what it would see in the CRM chat. Text found inside CRM records, notes, e-mails and documents is treated as data, not as orders, so a clever sentence pasted into an incoming message cannot redirect the assistant. And if the instruction is unclear or the assistant is not confident, it performs no write action at all and explains in one sentence why.
Texts in rules support placeholders, so the message a customer receives can name the person and the document. The tokens are shown as small clickable chips under the text field; a click copies one to the clipboard. Common ones include the contact name, the customer e-mail, the deal name, the stage name, the task title, the text and the sender of an incoming message, the invoice or contract number, and your own constants and variables defined once in the section. This is the difference between a generic reminder and a message that reads as if a colleague wrote it.
The webhook action is the quiet workhorse for firms that already own a toolbox. It sends a POST request with JSON — the rule, the event, its data and your text — to an address you provide. Only https addresses of public hosts are accepted, and if the other side does not answer with a success code, the action is recorded as an error. That is enough to push a signed contract into an accounting tool, drop a paid invoice into a bookkeeping spreadsheet, or send a notification into a team messenger. The rule stays inside Firmspace; the outside world hears about it only when you decide it should.
Rules are not trusted blindly, they are observable. Every run is written to a log with the rule name, the date, a status and a message — including the cases you would otherwise never notice, such as an action that failed because no mailbox was connected or because the deal had already left the stage while the rule was waiting. There is also a test tab, so you can see what a rule would do before it starts doing it for real. A waiting rule survives an edit as long as you keep it enabled; switch it off and the queued action does not fire. If it has quietly stopped making sense, the log is where you find out, not in a customer's complaint.
Two limits are honest to mention before you build your first rule. The AI action needs the AI feature of the highest plan and consumes the same daily quota as the AI chat, the receipt scanning and the finance analysis — the quota counts conversations per firm per day, so a team that talks to the assistant a lot should keep an eye on it. The number of rules is capped by plan as well. And invoice events carry the customer name and the invoice number but no e-mail address, so an e-mail addressed to the client will not find a recipient there; use a notification to your team or an e-mail to the owner instead. Knowing this in advance saves an afternoon of confusion.
The practical way to start is small. Pick the one moment in your week where work most often slips — the missed call, the quote that goes quiet, the invoice that passes its due date — and write a single rule for it: an event, a delay, one action, a text with a placeholder or two. Let it run for a week and read the log. When the first rule behaves, add a second, and only then consider giving one of them an AI instruction for the cases where the right next step depends on what the CRM already knows. Automation in a small firm is not about replacing people; it is about making sure the customer does not fall into the gap between two busy afternoons.