Ask a sales team where their week goes and the answer is rarely selling. Between transferring enquiries into the CRM, writing follow-ups, chasing quotes and preparing reports, the administrative share is commonly 25 to 35 percent. Most of that is automatable with tools the company already pays for.
Start by measuring, not by buying
For two weeks, have the team log how long each recurring task takes and how often it occurs. You will end up with five to eight candidates. Automate the one with the highest product of frequency and duration first, not the one that sounds most impressive. This exercise costs almost nothing and prevents the classic mistake of automating an exception.
Five automations that pay off almost always
- Enquiry to CRM: web form, e-mail and phone lead land as a deal with source, service interest and consent status, with no retyping.
- Enrichment: company data from the commercial register or a business directory is added automatically, so nobody googles the VAT number.
- Follow-up sequences: three reminders at day 2, 7 and 21, stopped automatically once the prospect replies.
- Quote generation: a template is filled from CRM fields and returned as a PDF, cutting a 40-minute task to five.
- Pipeline hygiene: deals untouched for 30 days move to a review list with a named owner instead of silently rotting.
For a team of five, these five workflows typically save 25 to 40 hours a month. At a loaded cost of 60 EUR per hour that is 1,500 to 2,400 EUR of capacity returned monthly, against a build cost of 4,000 to 9,000 EUR. The payback period is therefore three to six months, and unlike a new hire the capacity does not take holidays.
Which tools you use matters less than most vendors suggest. HubSpot, Pipedrive, Zoho and the open-source options all expose the APIs these workflows need, and orchestration can run in n8n, Make or a few hundred lines of code in your own environment. Choose by two criteria: whether your team will actually open it every day, and whether the data can be exported in full at any time. Everything else is negotiable.
Where AI adds something and where it does not
A language model is useful for summarising a long e-mail thread into a next step, drafting a first reply for a human to edit, and classifying enquiries by topic and urgency. It is the wrong tool for calculating discounts, checking stock or deciding credit terms. Those are rules, and rules should be code: deterministic, testable and auditable.
Automate the typing, not the judgement. The moment a customer notices the difference, you have automated the wrong step.
Data quality decides everything
Automation multiplies whatever is already in the system, including the mess. Before you build, define mandatory fields, a single deduplication rule, and one owner per record. A pipeline built on duplicate contacts produces duplicate follow-ups, and the fastest way to lose a prospect is to send the same reminder three times from three colleagues.
Adoption is a management task
- Involve two sales people in the design, not only the head of department.
- Roll out one workflow at a time, with a two-week observation period each.
- Keep a manual override for every automation and log when it is used.
- Review the logs monthly: frequent overrides mean the rule is wrong.
Finally, write down what happens when an automation fails. A workflow that silently stops is worse than no workflow, because everyone assumes it is still running. Error notifications to a named person, a retry policy and a monthly health check belong in the scope from day one.
