You can automate most of the admin around a CRM: capturing leads, enriching and deduplicating records, creating follow-up tasks, logging mails and calls, updating stages and building reports. Use plain rules wherever the logic is clear and AI only for reading free text. Leave decisions on deal value and customer status with a person.
What can you automate, and with what?
Most of this list doesn't need AI at all. The right-hand column is the one that saves you money.
| Task | What happens | Needs |
|---|---|---|
| Capture leads | A form or mail becomes a contact and a deal | Rules, plus AI if it's free text |
| Route leads | The right owner gets it by region or type | Rules |
| Create follow-up tasks | A stage change creates a task with a due date | Rules |
| Flag stale deals | No activity for a set time triggers a reminder | Rules |
| Spot duplicates | Same domain or mail address gets merged or flagged | Rules, AI for fuzzy cases |
| Enrich records | Company details are added from a source | A connection |
| Log conversations | Mails and call notes are summarised on the record | AI |
| Pull out next steps | Commitments in a mail become tasks | AI |
| Report | A weekly pipeline overview, with a short summary | Rules, AI for the summary |
Start with what your CRM can do by itself. Built-in workflows differ per product and plan, so check there before buying anything extra.
Where does AI really help in a CRM?
The most useful thing AI does is close the gap between conversations and records. CRMs go stale because people don't enter what they've agreed by mail or on a call. AI can read those conversations and propose the update, so a salesperson only has to confirm it. It also qualifies free-text enquiries, as described in how AI lead qualification works, and can sort the inbox that feeds the CRM.
What shouldn't run without a person?
- Changes to deal value or expected close date
- Marking a deal lost or a customer as cancelled
- Merging records when the match is uncertain
- Mass updates over many records
- Mails to customers that are generated from CRM data
Use the rule from human in the loop: if it's hard to undo or reaches a customer, a person approves it first.
What goes wrong most often?
Automating on top of messy data. Automation doesn't clean up a CRM, it spreads what's there faster. A wrong segment now gets wrong mails sent to it automatically. Before you build, run this short check:
- Who owns each field? A person or the system. Not both without a rule.
- Do you agree on the stages? If each salesperson uses them differently, fix that first.
- Can you see what changed? Every automated change should be traceable to its source.
- Is AI output marked? Keep AI-written values in their own field until someone has confirmed them.
- Can you roll back? Export before you run anything across many records.
That fourth point is the one most teams skip, and it's the one that keeps your data trustworthy. Once people stop trusting the CRM, they go back to spreadsheets and the automation has no value left.
Where do you start?
Pick the task your sales team complains about most and measure it for a week. Often it's logging or follow-up. One defined flow fits an Automation Sprint, from €2,950. A Business Agent, from €7,950, is made for complete processes, including CRM automation with AI and approval moments. Prices exclude VAT. For general advice, see which process to automate first.
