Here's a scenario I see a version of all the time. A small service business starts with a CRM. Then it adds a form builder for the website, a project board for jobs, an e-signature tool for contracts, and finally a connector like Zapier so the other four can talk. Each one made sense on the day it was bought. Together they cost a few hundred dollars a month and still leave someone retyping data between tabs. This post is a walkthrough of how I'd replace that kind of stack with one custom CRM for small business use, in a single afternoon, using the tools I build with today.
To be clear up front: this is a hypothetical example, not a client case study. The point is to show the steps, so you can judge whether something like it fits your business.
The example: five tools, one broken handoff
Picture a 6-person residential cleaning company. Let's say its stack looks like this:
- A CRM, priced per user
- A form builder for website quote requests
- A project board to schedule jobs once a deal closes
- An e-signature tool for service agreements
- A connector tool to move data between all of them
For this example, call the total about $300 a month. Your numbers will be different, so plug in your own later.
The bill isn't the real problem. The handoffs are. A quote request lands in the form tool, someone copies it into the CRM, someone else creates the job on the project board after the deal closes, and the connector quietly fails every so often. Leads go cold in the gaps.
The stack I'd use
- >n8n for the workflows. It can run on n8n's cloud or be self-hosted, and it replaces the connector tool entirely.
- Supabase for the data. It's a real Postgres database with a >free plan to start.
- Claude for lead scoring and drafting. ChatGPT or Gemini work too; I pick based on the job.
- Your existing email and calendar. No new inbox to check.
Everything gets created in the owner's accounts. If they ever stop working with me, it all keeps running.
The afternoon, step by step
Before starting: map the process (30 minutes)
I don't touch a tool until I can draw the path on one page. Where do leads come from? What makes a good one? What happens the moment a deal closes? Who needs to know? This half hour saves the rest of the afternoon.
Hour 1: one database instead of five
In Supabase I'd create four linked tables: leads, clients, jobs and documents. A lead becomes a client by changing its status, not by being copied into another app. Then I'd turn on >row level security so each role only sees what it should. That last part is the step DIY builds usually skip.
Hour 2: intake and AI lead scoring
The website form either posts to an n8n webhook or gets replaced with an >n8n form. Either way, the workflow:
- Checks for a duplicate by email and phone.
- Saves the lead in Supabase.
- Sends the answers to Claude with a short ideal-customer description ("recurring service, within our service area, home size we handle well").
- Gets back a 1 to 10 score and a one-line reason in a fixed format.
- Writes both to the lead record, and texts or emails the owner when the score clears a threshold.
Now the best leads get a call the same hour, and nobody reads every submission by hand.
Hour 3: from "won" to scheduled, automatically
When a lead is marked won, n8n creates the client record, opens a job, fills a service agreement template from data already in the database, and emails it.
One honest caveat: emailing a PDF is not the same as a legally binding e-signature with an audit trail. If your agreements need that, keep a signing tool in the flow. You might still drop to a cheaper plan once it's only doing one job.
Hour 4: give the team what they need without a new screen
Instead of building a dashboard on day one, I'd have n8n send a morning digest: new leads, scores, jobs this week, anything stuck. The team keeps working in email. A proper screen can come later. I'd sketch it in Lovable to agree on the layout, then build it properly on the same Supabase data.
What doesn't fit in an afternoon
Being straight about this matters. The afternoon gets you a working core. It doesn't cover cleaning up years of messy CRM exports, testing weird edge cases, tightening permissions for every role, or teaching the team the new routine. Those take more hours, and they're what make a build dependable instead of a demo. If you're curious why that gap matters, I wrote about it in The Hammer Doesn't Make You a Carpenter.
Run your own numbers
Grab your last bank statement and fill this in:
- Current monthly cost: add up every tool this would replace.
- New monthly cost: hosting, database plan and AI usage.
- Hours of retyping per week: multiply by 4.3, then by what that person's hour costs you.
- Monthly savings: (1 minus 2) plus 3.
- Payback: divide the one-time build cost by line 4. That's how many months until it's paid for itself.
If line 4 is small, don't build. Connect what you have and move on. If you want a broader look at what disconnected tools cost, read Your SaaS Stack Is a Subscription Graveyard.
How I'd approach this
I wouldn't start with the CRM. I'd start with the bottleneck. Most small businesses have good people and a product that works; the process in between is what leaks. If the leak is slow lead follow-up, the fix might be one n8n workflow on top of your existing CRM. Only when several tools overlap and still don't fit does a build like this make sense. You can see the kinds of work I automate, and my rates are on the pricing page.
Frequently asked questions
Can you really build a custom CRM for a small business in an afternoon?
The core of one, yes: the database, intake, AI scoring, alerts and document generation for a business with a clear process. What doesn't fit in an afternoon is cleaning up years of old data, testing edge cases, setting permissions for every role, and training the team. Plan on the afternoon for the build and more hours after it to make it dependable.
What do I need to run a custom CRM built with n8n and Supabase?
An n8n account or a self-hosted n8n server for the workflows, a Supabase project for the database, an API key for an AI model like Claude, and your existing email. Both n8n and Supabase have free options to start. Everything should be created in your accounts, not the builder's, so you own it from day one.
How does AI lead scoring work?
When a lead submits your form, the workflow sends their answers to an AI model along with a short description of your ideal customer. The model returns a score and a one-sentence reason in a fixed format. That score is saved on the lead record, and anything above your threshold triggers an alert so the best-fit leads get a call first.
Should I replace my current CRM with a custom one?
Only if it keeps fighting your process. If your CRM fits and the problem is that it doesn't talk to your other tools, connect them instead. That's faster and cheaper. A custom CRM makes sense when you're paying for several overlapping tools, paying a connector to glue them together, and still retyping data by hand.
Find the leak before you rebuild anything
You probably don't need a new CRM. You need to know where leads and hours are slipping through. Book a 30-minute conversation and I'll look at your biggest bottleneck with you and tell you whether it calls for a connection, a build, or nothing at all. No pitch.
