Where AI helps in a CRM project, and where it does not
AI tools do a large share of the typing and checking on every project, and none of the deciding. Which parts of a CRM and automation build they handle, and which stay with a person.
- · updated
- 4 min
Clients sometimes ask whether an AI builds their CRM bridge. AI tools draft the tests, the documentation and a first review on every project. A person makes every decision. Here is where the line sits in a CRM and automation project.
What the tools do
- Draft the automated tests while a feature is being built, so the tests arrive with the code instead of weeks after it.
- Write and update the technical documentation: what each workflow reads, what it writes, and what it retries when a call fails.
- Run a first review pass on every change: a missing error branch, a field mapped twice, an empty value nobody handled.
- Run accessibility and performance checks on the screens a client will actually use.
The numbers are in the case studies: 144 end-to-end tests and 24 passing axe-core checks on the Atharli platform, 101 tests on the Zahrat app.
What stays a human decision
- The data model: which system is the source of truth for a contact, a call, an invoice.
- The business rules: who gets the next untreated lead, when a quote becomes an invoice, what counts as a duplicate.
- What goes to production, and when.
- Anything a client or their customers will see: wording, the numbers on a dashboard, an e-mail sent in their name.
These are the decisions that break a CRM when they are wrong, and no tool can read a client's sales process from the outside. They are settled with the client, in writing, before anything is built, and a person reviews and tests every change before it ships. The reasoning behind one of them is in why the source of truth is a database, not the CRM.
What it changes for a client
Mostly coverage and speed. Tests and documentation come by default rather than as paid extras, and the written scope with a fixed price arrives within 48 hours of the review call. The price stays fixed once signed. The offers and the monthly care plan are on the services page.
What it does not mean
AI-generated images appear in a deliverable only if the scope says so, and no chatbot is added to a CRM because it is fashionable. When a project does call for an AI step, such as a short summary of a call written onto the CRM record, it appears in the scope like any other step, with what it costs and where the data goes.
How an automation project is scoped, built and kept running every month: n8n automation agency →