TL;DR
- We run 8 active ventures with effectively 1 operations person because AI handles the repetitive work.
- Our stack is n8n + Supabase + OpenRouter + Make.com + custom dashboards — nothing proprietary or locked behind someone else's platform.
- SchoolRegistry.ng gets 30K sessions/month with zero manual blog posts. TaxLienSimple publishes 50+ articles/month. The Boring Receipt publishes daily — all faceless.
- This took 2 years to build, not 2 weeks. Anyone selling you an "overnight AI transformation" is lying.
- If a venture needs more than 20 hours/week of human attention, it gets its own ops — this model has a hard ceiling.
The Honest Truth: This Took 2 Years
I need to say this upfront because every AI influencer on LinkedIn is selling you a fantasy: we did not automate eight companies in a weekend.
It took two years of failed workflows, broken API connections, agents hallucinating customer responses, and one memorable incident where an AI phone agent tried to schedule a dental appointment for a tax lien investor. (The transcript is saved. It is funny now. It was not funny then.)
What you are about to read is the result of 24 months of iteration, not a $997 course.
What Each Venture Gets
We run eight ventures under our umbrella. Each one shares the same infrastructure but gets custom modules and dedicated AI agents:
| Venture | What It Does | AI Agent Role | Human Touchpoint |
|---|---|---|---|
| SchoolRegistry.ng | Nigerian school discovery platform (15K+ listings) | Content agent publishes exam-funnel articles; support agent handles parent inquiries | Final content approval; school verification |
| TaxLienSimple | Tax lien investing education | 50+ articles/month; lead qualification; email nurture sequences | Investment accuracy review; compliance check |
| The Boring Receipt | Faceless daily content property | Daily article + social distribution; fully autonomous | Monthly brand voice audit |
| SceneHost | 360° virtual tour platform for STR hosts | Guest guide generation; listing optimization | Customer onboarding calls |
| AutoWalk | AI marketing for auction dealers | Studio-quality image generation; 360° spin; listing copy | Dealer relationship management |
| VettyDrive | Rental fleet management software | Compliance documentation; AI-generated marketing assets | Fleet onboarding; compliance review |
| CDLSchoolsUSA.com | CDL school directory + DOT training | Content generation; lead routing; white-label fulfillment | School partnership calls |
| AIScripts.Studio | GPT workflow marketplace | Prompt block generation; copy asset creation | Creator support; quality control |
Every venture sits on shared infrastructure but operates through custom modules tuned to its business logic. The AI agents don't just share a brain — each one has a memory, a script, and a specific job description.
The Stack (No Vaporware, No Affiliate Links)
This is what we actually pay for and run:
n8n (Self-Hosted) — The Workflow Engine
Everything flows through n8n. Content pipelines, lead routing, reporting, data synchronization between ventures. Self-hosted means we own the data, control the execution, and never hit rate limits at the wrong moment.
Supabase — The Database
All venture data lives in Supabase: customer records, content calendars, agent memory, conversation logs, analytics. PostgreSQL under the hood. We can query across ventures for cross-selling insights.
OpenRouter — The AI Model Layer
We don't bet on one model. OpenRouter lets us route prompts to the best model for the job: GPT-4o for complex reasoning, Claude for long-form content, Llama for cost-sensitive bulk tasks. If one provider goes down, we switch in seconds.
Make.com — The Quick Integration Layer
When we need a fast, no-code connection to a third-party tool (new CRM, random SaaS, a client's legacy system), Make.com handles it. n8n does the heavy orchestration; Make.com handles the quick glue.
Custom Dashboards — The Command Center
One dashboard. Eight ventures. All monitored in real time. We see:
- Content pipeline status per venture
- Lead flow and qualification rates
- Agent error rates and escalations
- Revenue and cost per venture
- System health across all integrations
If something breaks, we know in minutes, not days.
The Weekly Time Breakdown
Here's what a typical week looks like for our operations lead:
| Activity | Time Spent | Automation Level |
|---|---|---|
| Content production | 0 hours | Fully automated: research → outline → draft → human approval queue |
| Reporting | 30 minutes | Automated dashboards; human reviews anomalies only |
| Phone coverage | 0 hours | AI agents answer, qualify, book, route. Humans get summaries. |
| Email triage | 1 hour | AI sorts by urgency; human handles edge cases and approvals |
| Approvals & edge cases | 4-5 hours | Content final sign-off, complex customer issues, system fixes |
| Strategic ops | 3-4 hours | New workflow design, vendor evaluation, venture scaling decisions |
| Total | ~8-10 hours/week |
The remaining time? Building the next automation, improving agent performance, and occasionally fixing whatever broke on Tuesday afternoon.
Real Numbers, No Rounding Up
SchoolRegistry.ng: 30K Sessions/Month, Zero Manual Blog Posts
Our content agent publishes exam-funnel articles targeting Nigerian standardized tests (WAEC, NECO, JAMB). These articles drive organic traffic to school listings. Every article is researched, drafted, optimized, and queued for human approval. Once approved, it publishes automatically.
Result: 15 exam-funnel articles delivered, 83,000 words, zero manual writing.
The Boring Receipt: Daily Content, Fully Faceless
A content property that publishes every single day. The entire pipeline — topic selection, research, writing, image generation, social distribution — runs without human intervention. A human audits monthly for brand drift.
TaxLienSimple: 50+ Articles Per Month
Tax lien investing is niche, regulated, and trust-dependent. Our agent produces educational content at scale, but every article is fact-checked by a human before publishing. The agent does the research and drafting; the human validates accuracy.
The Command Center Concept
Running eight ventures without a central nervous system is impossible. Our command center is a set of custom dashboards built on top of Supabase + n8n:
- Venture Health Panel — Traffic, revenue, lead flow, content queue depth per venture
- Agent Operations Panel — Conversation logs, error rates, escalation reasons, model performance
- Content Pipeline View — Every piece of content from idea to published, across all ventures
- Alert Center — Broken workflows, failed API calls, threshold breaches
One screen. All eight ventures. If something is red, we click into it. If everything is green, we go build something else.
When This Model Breaks
I promised radical honesty. Here is where our model fails:
If a venture needs more than 20 hours per week of human attention, it needs its own ops.
We learned this the hard way. One venture started requiring heavy partner management, custom sales calls, and manual fulfillment. We tried to force it into the shared model. It broke everything — delayed content for other ventures, missed agent alerts, unhappy customers.
We spun it out. It now has dedicated ops. Everyone is happier.
Other situations where this model does not work:
| Scenario | Why It Breaks | What We Do Instead |
|---|---|---|
| High-touch B2B sales | AI can't build trust in complex deals | Dedicated sales ops |
| Regulatory-heavy services | Compliance requires human judgment | Dedicated compliance review |
| Real-time crisis management | AI is too slow for true emergencies | Human escalation protocols |
| Creative direction | AI generates; humans direct | Hybrid creative lead |
The Philosophy: AI Operations Before Employees
We do not hire for repetition. We hire for judgment.
Every time we consider adding a person, we ask: "Can an AI agent do 80% of this?" If yes, we build the agent first. If the remaining 20% is high-value human work, we hire for that 20%.
This is not about replacing people. It is about not hiring people to do work that makes them miserable.
Our operations lead is not a content farmer. They are a systems architect who happens to run eight companies.
What's Next
This stack is not finished. We are currently testing voice agents for outbound follow-up, expanding our GEO (Generative Engine Optimization) capabilities, and building cross-venture recommendation engines.
If you are a business owner wondering whether AI operations can work for you, the answer is yes — but only if you are willing to build it honestly, iteratively, and without buying into overnight transformation fantasies.
Book a call with us to discuss your operations stack. We will tell you what is realistic and what is not.
Related reading: