Insights

My studio runs on its own CRM, and here is what that costs

My studio runs on its own CRM, and here is what that costs

On October 19 I sent my first batch of letters to businesses I wanted to work with, each one pointing to a demo I had built for them. I sent them by hand from Gmail, thirty in all, and kept track in a spreadsheet and my own memory. One bounced, one person asked me not to write again, and by the next morning I knew the spreadsheet would not survive a second batch. Over the following three days I built the studio its own CRM, on the same engine my clients use.

Renting was the obvious answer

The usual move is to rent one. The big products are good and their prices are public. When Salesforce reshaped its editions, its September 3, 2026 announcement put the core sales edition at $195 per user per month. HubSpot's Sales Hub Professional, as listed when I checked this month, is $90 a seat each month plus a $1,500 onboarding fee. For a sales team of twenty, either is a sensible bill.

Price was not what stopped me. Shape was. My pipeline does not start with a lead filling in a form. It starts with me picking a business I would like to help, building them a demo of a better site, writing a short audit of the one they have, and only then writing to them. A rented CRM would hold the contact. The demo, the audit and the pitch page would live somewhere else, and I would spend my mornings copying links between tabs. I already had an engine with contacts, mail and a dashboard. What it lacked were three studio parts: proposals, campaigns and an organizer.

Diagram titled Contact, demo, letter, reply, all one record. The same engine my clients use, with three parts only the studio needs: proposals, campaigns and an organizer. Four steps. One, contact: a business I would like to help, imported or typed in, with a shared inbox like info@ kept apart from a person's name. Two, proposal: a demo built for them, with the concept site, the audit and the pitch page linked on the contact's record with the date it was sent. Three, letter: sent by my hand or by a campaign, on weekdays from 8 to 5, a few minutes apart, under a daily cap, and nothing launches until I press Launch. Four, reply: the inbox is read on a schedule, and a real reply stops the sequence and makes me a task. Below, the organizer's five tabs: never emailed, emailed and waiting, replied, bounced or opted out, and done for won, lost or not now. A footnote says the tabs are worked out from each record and its mail history, there is no status column to drift, and it was built October 21 to 23, 2026.

A record for every demo

The proposals module is the record behind each lead I have built something for. It holds the address of the concept demo, the audit, the pitch page, the score I gave their current site and the single biggest problem I found. Those become fields a letter can use, written as {demo_url}, {audit_url} or {top_issue}, so the sentence that says "I rebuilt your home page, have a look" points to the right demo for the right person every time. On October 23 I matched all 38 demos I had built so far to their contacts, and imported my sent Gmail history so the letters from the 19th sit on the records they belong to.

The module's own header carries one rule I care about: nothing in it sends mail. It listens. When a letter goes out, the proposal moves to sent. When a reply comes in, it moves to replied. The proposal is a record, not a generator, and keeping those jobs apart means a mistake in one cannot send anything from the other.

Letters with a cap and a button

The outreach module writes campaigns meant to read like one person writing to another, because that is what they are. A campaign is a short sequence: a first letter and its follow-ups, each a few days after the last. The sending rules sit in settings where I can see them. Letters go out on weekdays between 8 and 5 Pacific, no more than two at a time, with a random gap of two to six minutes between them. A daily cap of thirty is shared across every campaign, so starting a second one never doubles the volume. Each campaign's calendar says so in plain words: "Assumes nobody replies; the shared daily cap across campaigns can push sends later."

Every letter ends with my postal address and a line telling the reader how to stop hearing from me, with a page that does it in one press. Click tracking is off unless I turn it on for a campaign, so the links in a letter go straight to the demo and not through a counter of mine.

The rule I wrote first is the one I trust most. A comment at the top of the module says it: "Nothing in this module launches a campaign: only the Launch button does." Before that button does anything, a preflight looks for the gaps that would strand a letter halfway. If the key that signs each unsubscribe link is missing, Launch is blocked. If the site's address is empty, plain http or a local one, Launch is blocked too, because every way out in every letter is built from it. A contact with no first name does not block anything; the letter simply opens with "Hi there," rather than a guess. Then the button asks twice. I built all of that on October 22, and at the time of writing the studio's letters have still gone out by my own hand. The machine is ready. I would rather press the button than find out it pressed itself.

What comes back

Letters are the easy half. The hard half is the answers, because most of what lands in a sales inbox is not a person. Since November 9 the studio's mail runs through Google Workspace, and the CRM reads that inbox on a schedule. Each new message in a prospect's thread is sorted before anything acts on it.

Diagram titled Four kinds of answer, and only one is a person. Every message in a prospect's thread is sorted before anything acts on it, by a plain function with no database and no network, so a test can prove it. Reply: a person wrote back; the sequence stops, the contact turns warm, and I get a task and one email about it. Auto-reply: out of office or a ticket robot; not a reply, so the next step simply waits a week. Opt-out: no thanks or remove me; suppressed for good, and no campaign can enroll them again. Bounce: a delivery failure; suppressed and stopped, matched to the letter that failed. A footnote says that since November 9, 2026 the studio's mail runs through Google Workspace and the CRM reads that inbox directly.

The sorter is a plain function. It takes a message and returns one word: reply, auto, optout, bounce, or ours for my own letters. It touches no database and no network, which means a test harness can feed it a pile of real-shaped messages and prove each one lands where it should. An out-of-office is not a reply, so it does not stop the sequence; the next step just waits a week. A bounce or an opt-out stops everything and adds the address to a list no campaign can write to again. A real reply stops the sequence, turns the contact warm, makes me a task, and sends me one email about it. One, so a long thread does not bury me in notices about itself.

This is the same lesson I learned running a directory: the robot may retire a lead nobody has heard from, but it never overwrites what a person actually said.

An organizer that cannot drift

The contacts list has five tabs: never emailed, emailed and waiting, replied, bounced or opted out, and done. None of them is stored. Each one is worked out every time the page loads, from the contact and the history of mail with them, in a fixed order where bounced beats done and done beats replied. There is no status column for me to forget to update, so the tab a person sits in is always the truth about my letters to them. I can add coloured labels on top for my own sorting, eight of them, but the tabs are never mine to set.

The import has guards of its own. A shared address like info@ or hello@ is kept apart from a person's name, and a name that is clearly a company or a role is not dressed up as a person in a greeting. "Hi Info" is the fastest way to tell someone a machine wrote to them.

What owning it costs

Owning has a bill, just not a monthly one. When something breaks, I am the support line. Backups, upgrades and the odd strange message from a mail server are mine. There is no app on my phone, no marketplace of add-ons, and nobody else has tested my edge cases for me. The test harnesses exist because nobody else will run them.

What I get back is the ability to change a rule the same day I learn it. When getting the first campaigns ready showed me those three preflight gaps, closing them took an afternoon and a new section of tests, not a feature request and a wait. The pieces I build for the studio also become pieces my clients can have. The guards in the contractor demo and most of the businesses in one CRM, five businesses run on the same engine, and the studio is the first customer of every rule.

The page I open most mornings is the organizer. It carries a small badge with one number: replies I have not handled yet. Most days it says zero. On the days it does not, there is a person on the other side of it who looked at a demo I built for them and wrote back, and that is the whole reason the rest of this exists.

Common questions

Why not use Salesforce or HubSpot?

They are good products priced per seat for teams. The studio's pipeline starts with a demo built for each prospect, so the demo, audit and pitch links had to live on the same record as the contact and the mail.

Can a campaign start sending on its own?

No. Only the Launch button starts a campaign, after a preflight check and a second confirmation. Sends then run on weekdays under a daily cap shared across all campaigns.

How does the CRM tell a real reply from an auto-reply?

Every incoming message is sorted by a plain function into reply, auto-reply, opt-out or bounce. Only a real reply stops the sequence and creates a task; an auto-reply just delays the next step by a week.

What happens when someone opts out?

Their address is suppressed for good, every sequence they are in stops, and no campaign can enroll them again. Every letter carries a postal address and a one-press way to stop.

Related

← All insights