Notion + n8n · a company operating system
How an education company went from Excel and 40 hours of planning to a system that runs three programs at once
An education company spent 40 hours planning each off-site program in Excel and 15–30 hours a month moving payments by hand. Over 15+ months I designed and shipped an operating system on Notion and n8n: a project with 400–700 tasks is created in one click, payments are received and reconciled automatically, and the production workflows test themselves every day before a program.
At a glance
- Client
- An education company (name withheld)
- Scale
- 16–20 off-site programs a year, 15–30 participants each
- Team
- 8+ staff, up to 100+ people in the system as guests
- Timeline
- Since April 2025, 15+ months, ongoing
- Role
- Systems architect and automation developer, solo
Results
40 h → 15 min
program planning: 400–700 tasks in one click
15–30 h → 0
manual payment entry per month
85% → 99%+
payment records accuracy
30 → 2 min
per participant questionnaire
2–3 h → 5 min
rebuilding the schedule after a change
1 → 3+
programs in parallel instead of one per quarter
₽678,910
documented savings per year (as of March 2026)
≈$28
per month for the infrastructure
Before
- Each program meant 400–700 tasks across three levels of the team, created by hand from memory and old lists: about 40 hours per event.
- Payments arrived in two GetCourse accounts, domestic and international, and were entered by hand: 15–30 hours a month at roughly 85% accuracy, with lost payments and mixed-up currencies.
- Data lived in Google Sheets, chats and paper lists. Finding out who had paid and who had arrived meant calling colleagues.
- Manual management hit a ceiling of one program per quarter; the team could not run several in parallel.
- Moving one session in the schedule meant 2–3 hours of recalculation and sign-offs.
What was built
Three-level Notion architecture
15+ linked databases; leadership, coordinators and staff each get their own tasks and their own dashboard. Moved to a plan with 100+ guest seats instead of 10.
A project in one click
Templates and cascading buttons create a project with every task and role assignment. Deadlines are computed from the event date, so deadline errors went from 5–10% to zero.
Payments without a human in the loop
Webhooks from both GetCourse accounts → n8n → Notion, with multi-currency, de-duplication, repeated webhooks and dates in any format. Plus direct payments: Russian SBP via T-Bank and Stripe for international participants, with a role-based finance dashboard. A money report takes 10 seconds instead of 5–10 minutes.
AI questionnaire analysis
Each participant questionnaire runs through an AI pipeline and becomes a Notion profile with insights and recommendations: 2 minutes instead of 30. The profile builds up from program to program.
QR attendance
Personal QR codes scanned at the door, with a live Telegram message showing who has arrived and who is still expected. First live run: 90 participants.
Zoom and access control
Call recordings land in Notion and Google Drive, linked to events. Assigning a role grants access to the right pages automatically and revokes it after the program.
A system that tests itself
Seven, three and one day before each program, the production workflows run an end-to-end test on test data every day and report to Telegram. The team learns about a breakage before participants arrive, not during the program. Any production error arrives as a notification.
Stack
- Notion
- n8n (self-hosted, Docker, VPS)
- GetCourse
- SBP (T-Bank)
- Stripe
- Google Forms, Sheets, Drive
- Zoom
- Telegram
- AI models via API
Questions
How many workflows are in the system?
A system of 66 workflows in the registry, 46 of them active in production: about 1,100 nodes, 46 of them AI nodes. I designed it and have been developing it for 15+ months.
Why self-hosted n8n rather than Zapier or Make?
Cost and control. The infrastructure costs about $28 a month; a comparable load on Zapier or Make would start at $100 a month. Data and logic stay on our own server.
What happens if an automation breaks?
Before each program the production workflows run a daily end-to-end test on test data, and any production error is reported immediately. Breakages are found before participants arrive, not a week later.
How were the savings calculated?
₽678,910 a year is documented savings as of March 2026, calculated for the first 14 workflows. There are now 66, so the current figure is higher, but I will not quote it until it has been recalculated. Payback period: 12 months.
Where does a project start?
With a process audit: a map of bottlenecks and an estimate of savings, before any development. Then design, and only then automation.