Prices hidden behind a call
Most agencies make a small business book a call just to hear a price. Owners want to know what it costs before they spend time talking.
Case study · Syndicate Holdings Group
The back officebehind the agency.
Small agencies usually run on a pile of rented tools that don’t talk to each other. I designed and built one system instead: a store, CRM, client portal, analytics and AI agents on one database. A small business can buy from us online without a sales call, every new client is set up automatically, and the AI does real work inside rules it can’t bend.
33klines of code, one codebase for every site and app
26products and plans sold online from one price catalog
8steps that run on their own after every paid order
4AI agent jobs running on schedules
152end-to-end checks on checkout, quotes, booking and sign-in
110help articles: 79 for clients, 31 for the team
Lines counted from the source when this page was built; the rest as of October 2026.Revenue and client figures stay private.
01 The problem
Lots of tools, one person.
Syndicate Marketing builds websites, runs ads and manages Discord communities for small businesses. I wanted it to run like a much bigger company without the cost of one. Three things stood in the way.
Busywork after every sale
Every new client meant the same manual setup: invoice, project, logins, kickoff, tracking code. That doesn’t scale, and it’s easy to miss a step.
Results clients couldn’t see
Rented analytics and reporting add monthly cost and cookie banners, and clients rarely get a clear view of what their money is doing.
02 The system
One system, start to finish.
Every stage of a client’s journey runs through the same database, so nothing is copied by hand from one tool to the next.
- Found
A free website checkup, a blog, local landing pages and rule-bound cold email bring new businesses in.
- Reached
AI agents research prospects and write personal emails; the system decides what can actually be sent.
- Bought
Published prices, a cart and a four-step Stripe checkout, with one-off and monthly plans in one order.
- Set up
Payment triggers the whole onboarding: contact, invoice, project, portal access and analytics.
- Built
The client fills in a kickoff form, an AI agent drafts the site, and I review it before the client sees it.
- Cared for
Uptime, security and speed checks, a monthly report, renewals through Stripe and online cancellation.
03 Automation
What happens when a client clicks Pay.
The checkout never trusts prices from the browser: the server prices every cart again from our own catalog. Once Stripe confirms the payment, the onboarding runs by itself, and each payment is handled exactly once even if Stripe sends it twice.
If a confirmation ever goes missing, a scheduled job checks with Stripe directly every 15 minutes, so no order falls through the cracks.
- Client and company created in the CRM, with where they came from
- Deal marked won in the sales pipeline
- Paid invoice recorded
- Project created with the right steps for what they bought
- Client portal access emailed as a one-click sign-in link
- Analytics set up for their new site, with a live dashboard
- Kickoff form and call link sent, plus my to-do tasks
- Confirmation emails to the client and to me
04 Where the AI fits
Agents inside the rules.
Scheduled AI agents do the judgment work: researching prospects, writing emails, answering replies, drafting websites and blog posts. Everything with a rule stays in plain, tested code.
Judgment work
Who’s worth contacting, what to say, how to answer a reply, how a client’s site should look. One narrow job per agent.
The rules
Daily send limits, opt-outs honored at once, only approved links, one job claimed by one run, and a stop if bounces climb. The agent can’t talk its way past any of it.
What clients see
No website draft or blog post goes out until I’ve reviewed it, and anything an agent isn’t sure about is handed to me.
05 The pieces
What clients actually see.
Every screenshot is a live page. The client portal is a demo account for a made-up business, so you can click around without seeing real client data.
A store instead of a sales call
- Every price comes from one catalog, so the site, cart, invoices and help center can’t disagree
- Cart rules stop orders that don’t make sense, like an add-on without its package
- Offers, creator codes and referrals, each checked again at payment
A portal that shows the work
- Sign in with an emailed link, no password to forget
- Progress, drafts to approve, files, invoices and their own analytics
- Cancel a plan online, with the end date shown first
A free tool that finds leads
- A plain-English report on any website in about a minute
- Each problem links to the product that fixes it
- One click turns a report into a draft quote I review before sending
06 Engineering decisions
The calls I made, and why.
The judgment calls that shaped it, with what each one cost.
One source of truth
Prices live in one catalog and every rule lives in one spec the AI agents read before they change anything, so the code, the site and the docs can’t drift apart.
Trade-offEvery change updates the spec too.
Keep the stack simple
No frameworks and zero npm dependencies. It runs on ordinary shared hosting, and every push to GitHub builds and deploys the sites.
Trade-offSome things a framework gives you had to be built.
Guardrails in code, not prompts
The agents are told the rules, but the system enforces them. A clever prompt can’t send to an opted-out address or post a draft to a client.
Trade-offSlower than letting automation run unchecked, on purpose.
Test where money moves
Checkout, quotes, booking and sign-in run against a local copy with a fake Stripe server and fake mail before anything ships: 152 checks.
Trade-offA test harness to keep up alongside the code.
Own the analytics
Cookie-free analytics that store no personal data, so our sites and every client’s get honest numbers without renting another tool.
Trade-offMore to maintain than a free tracker.
Security from day one
It handles payments and client data, so staff sign in with passkeys or two-step codes, leaked passwords are refused, sign-ins are logged and backups are checked nightly.
Trade-offA few extra steps to sign in.
07 How I built it
AI does the typing. I do the engineering.
The same loop for every change, from a typo to a payment flow.
- Research
How it should be built: OWASP, NIST, the RFCs, Stripe’s and Google’s own docs.
- Spec
The rules, written down where the agents read them before touching anything.
- Build
AI coding agents write most of the code, one scoped task at a time.
- Review
I read the code, catch the errors and shortcuts, and send them back.
- Test & ship
End-to-end tests, then a push to GitHub deploys it.
How I actually work
I read and understand the code, spot bugs and bad patterns before they ship, and research best practices instead of taking an agent’s word for it. That isn’t vibe coding: it’s specifying, reviewing, testing and owning what goes live.
08 What it shows
Why it matters for your team.
You probably aren’t hiring someone to build a CRM. This project is about how I take a messy business problem and ship an AI-powered system that solves it, safely.
Problem to production
I mapped every handoff in the business, automated the ones that didn’t need a person, and put AI where judgment was the bottleneck.
Agents you can trust
Narrow jobs, rules in code, human gates and stop switches. I run these agents every day and tighten the rules when they drift.
Built for the person paying
Published prices, a portal that shows the work, plain-English reports and compliance built in: CAN-SPAM, FTC rules, auto-renewal laws.
Open to AI roles
Want the walkthrough?
I’m happy to share my screen and show how any part of this works, down to the code. Email me and I’ll reply within one business day.