From n8n to a Custom CRM: When Automation Needs a Source of Truth

n8n will happily become your CRM if you let it. Every quote writes a row. Every status change is a node. Six months later the only person who can answer “where is this customer?” is the person who knows which execution to open. That is not a system. That is a very expensive memory palace.

We hit that wall after the Zapier-to-n8n move. The workflows were healthier. The records were not. Monday.com held jobs. Sheets held quote history. The site held a form. n8n held the story of how those things had once been in sync. Nobody held the customer.

A CRM is not software you buy because the category exists. It is a source of truth for the objects the business sells. For an EV charging operator those objects were not “leads” and “deals”. They were households, quotes, jobs, and the awkward states between a survey and an install. Off-the-shelf CRMs wanted us to pretend otherwise.

The test: where does the team look first?

Ask the people who run the week. When a customer calls, what do they open? If the honest answer is “Monday, then the sheet, then I search my inbox”, the CRM is a collage. n8n will not fix a collage. It will decorate it.

• A board is a view of work. It is not a customer record.
• A sheet is a dump of history. It is not a permissioned object.
• A workflow run is an event. It is not the thing the event happened to.

We wanted one screen that could answer: who is this, what did we quote, what job is live, and what is blocked. Everything else — email, Slack, firmware, ads — could stay around it. n8n would keep moving the edges. The centre had to sit still.

Buy vs build is a shape question

HubSpot, Salesforce, and Pipedrive are excellent if you sell like the companies they were designed for. If your pipeline is really a job board with money attached, you will spend a year mapping “opportunity stages” onto site visits. That mapping is the cost. The licence is the receipt.

We built because the process was the product. An EV install is not a deal that closes. It is a sequence of physical and admin states that a generic funnel always lies about. A thin CRM that told the truth was cheaper than a famous one we would spend the year fighting.

What “custom” actually meant

It did not mean a platform play. It meant four objects and the states between them.

1. Person — the household or site contact, with a stable ID.
2. Quote — what we offered, what they accepted, what expired.
3. Job — the live piece of work, with the states an installer would recognise.
4. Event — the trail n8n already knew: emails, board changes, payments.

n8n stayed. It created quotes from the calculator, flipped job states, and notified the board. It stopped being the filing cabinet. The CRM became the thing you opened when a human asked a human question.

How the cutover ran

Same discipline as the Zapier move. Dual-run. The board stayed the daily view until the CRM view was less wrong. We imported live jobs first, archive second, vanity never. n8n wrote to both until the CRM was trusted, then the board became a projection of the CRM instead of a source.

The failure mode to watch is split-brain: a state that exists in n8n, the board, and the CRM at once. Pick a writer. Everything else reads. If two systems can update the same job, they will, and you will spend Friday deciding which Tuesday was real.

What this is not

This is not an argument for building software as a personality. Most businesses should buy. Most of those who should build still over-build. A custom CRM is justified when the records are the operation — and when you are willing to let n8n, the site, and the inbox stay as edges.

If you are earlier than that, you do not have a CRM problem. You have a process you have not mapped. That is a different piece of work, and it is cheaper. I wrote that up as a guide to automating the business without buying another tool.

A workflow without a record is a rumour. A record without a workflow is a filing cabinet. You need both, and they are not the same product.