← back
🔥 Production Case Study

Flirthaus

The founder walked away from a content agency. I walked in, automated every bottleneck with AI, and took it from bleeding money to $50K/month in sales — in 30 days.

Business AutomationAI AgentsTelegram BotsRemotionN8NPython TUIStreamlitAirtableDiscord Bot

// results

30 days

turnaround time

$50K/mo

in sales generated

0–2 hrs

weekly time spent

0 clients

lost during transition

// the-situation

Someone handed me a content creation agency — the kind that manages creators on platforms like OnlyFans and Fanvue — and said "good luck." The founder had walked away. No systems. No automation. Everything was manual. The business was bleeding.

The normal move? Hire a team. Build departments. Scale through headcount. I hired AI agents instead.

My first benchmark was survival: don't lose a single client in 30 days. Day 30 came. Zero churn. Green light.

// phase-1: fix operations

The first problem was chaos. Schedules lived in people's heads. Information was pinned in Discord channels nobody read. Managers spent their days looking things up instead of managing.

Login screen

// login screen

First thing: moved everyone from weekday/weekend schedules to an 8-day rotation. Two groups — A and B (I'm not good at naming). Four days on, four days off. No random swaps. This standardized performance tracking — same hours, comparable output.

Scheduling

// 8-day rotation scheduling

Chatter scheduling

// 4 days on, 4 days off per chatter

Discussions moved from Discord (where messages go to die) into the app with internal and creator-facing sharing. The right information shown to the right person.

Discussions

// discussions moved from Discord to the app

Managers whose time got freed up started innovating. Once people aren't drowning in grunt work, they actually start managing.

Manager suggestions

// managers started offering suggestions to improve the org

Managers

// suggestions and improvements

Privacy features

// ability to hide names during Zoom meetings

Peer reviews

// peer reviews linked to lessons and ideas from other team members

Chatter management

// managers create/edit/remove chatters with portal access

Sales overview

// quick overview to compare sales with breakdowns

Rankings

// chatters see their own rankings

Chatter schedule and incidents

// rovintest has been doing pretty good. no complaints so far.

// the-hype-bot

All that eliminated the biggest source of confusion and time waste — scrolling Discord for pinned messages. Now the right info goes to the right person, leaving Discord for the important things. Like...

An AI sales HYPE-BOT. The messages get louder the bigger the sale.

Hype bot

// AI sales hype bot

Hype bot loud

// gets louder with bigger amounts 🔥

Clock-in tracking

// tracks who DIDN'T clock in — more important than who did

// phase-2: make it data-centric

Operations were stable. Next problem: nobody knew why numbers went up or down. Without data, it was all vibes. I built a shareable Streamlit app so clients could see their own performance. Cohort reports showed WHY a week was lower or higher. Marketing changes? Bad sales on our part?

Analytics

// cohort analysis — why did this week look different?

Sometimes numbers go up because of marketing from weeks ago bringing in segments that kept spending. Sometimes vice versa. Segmenting by weeks helped clients understand their own marketing effects.

Cohort segmentation

// weekly cohort segmentation

Segments

// segment analysis

Spender segments

// big spenders vs ones who hadn't spent in a while

Spender details

// spending detail breakdown

Activity breakdown

// buying activity by day of week and hour of day

One client didn't realize that even though his traffic was decreasing, we were actually increasing the ROI on each segment. The data told the real story.

ROI analysis

// blue line (traffic down) vs bars (revenue up) — data told the real story

// phase-3: automate the money

A huge chunk of time was spent chasing money and paying people. The old process:

  • → Calculate sales per client, lookup invoice percentages
  • → Create PDF invoices, send to Telegram
  • → Ask if they saw it (even though double checkmarks say they did)
  • → Follow up. Follow up again. Until they pay.
  • → Calculate commissions, bonuses, hourly rates
  • → Send payroll through Wise, one at a time

2-day process. Every week. I had removed the person doing this but was now doing it myself. 2 days out of 7 just moving money around.

I had heard of this thing called N8N. Link Airtable, calculate commissions in JSON, save back to Airtable, generate HTML invoices, convert to PDF with Gotenberg (self-hosted of course), look up the client's Telegram room, send.

N8N workflow

// wait I wasn't done...

N8N invoice workflow

// it was a beautiful mess

I wanted something simpler because I had no idea what I had built a week later. Replaced it all with a Python TUI that does the same job in seconds and stays idempotent.

Python TUI

// from beautiful mess to simple CLI

TUI running

// runs in seconds, stays idempotent — agents can call any part of it

The bonus: it's a CLI, so AI agents can call any part of it. Clients ask for their invoice over Telegram? Agent runs the command, returns the PDF. Wise API handles payouts. A crypto wallet scanner checks for incoming payments.

Wallet scanner

// crypto wallet scanner — agents check if payments came through

// phase-4: content & growth

By now I had streamlined operations, made the business data-centric, created a sales hype bot, added a customer support bot on Telegram, let people tattle on each other anonymously, and automated invoicing and payroll. Time for growth.

The content pipeline: Remotion templates + OpenRouter to generate marketing content, pushed to Telegram channels. The plan is an agent that's helpful in community channels, builds trust, then drives traffic to our channel and schedules calls.

Remotion pipeline

// Remotion templates + OpenRouter content generation

Content examples

// generated content samples

More content

// work in progress

// phase-5: training the team

The other piece to prepare for growth — balance the seesaw of clients ↔ sales team. A standardized training system with lessons and an automated chat simulator.

"Think flight simulator, but for flirting. I wish I was joking."

They log in and start flirting with an AI. They're reviewed on how engaging they are. They learn to deal with a variety of "pushy" fans before they're released on a live account.

Training system

// training system login

Training chat

// practice flirting with AI before going live

Training review

// reviewed on engagement quality

// the-result

30 days to turn around a failing agency. Operations streamlined. Data-centric decision making. Automated invoicing and payroll. AI hype bot. Customer support bot. Anonymous peer feedback. Content pipeline. Training simulator.

From losing money → $50,000/month in sales. With invoice margins, that's $12K–$15K/month in revenue. Costs cut. Time spent: 0–2 hours per week.

Six-month target: $100K–$200K/month in sales. $24K–$60K in invoicing. The only remaining bottleneck is me closing calls. Maybe one day an AI agent does that too.

At that point the entire business runs without me. Which is either the dream or the start of a very confusing identity crisis. I'll let you know which one it turns out to be.