Invoice processing · Production

TMNZ saves 400 hours per month and runs 150,000 monthly workflow executions with n8n agentic AI

The problem

TMNZ needed a self-hostable, secure workflow automation platform that non-technical staff could use without heavy training while giving engineers fine-grained control — requirements that Zapier, Make, MuleSoft, and the Microsoft suite could not meet simultaneously.

First attempt

Zapier and Make could only automate isolated workflows and were heavily limited in scope; MuleSoft and the Microsoft suite were either not self-hostable or cost-prohibitive for a 60-person Fintech.

Workflow diagram · grounded in source
1
Staff submits automation request
Trigger
More than a third of TMNZ staff initiate n8n workflows spanning market research, HR, finance, and operations.
▾ source quote
“more than a third of TMNZ staff are using n8n to automate tasks from market research and HR processes, to agentic workflows, running around 150,000 executions each month”
2
LLM models execute agentic tasks
Ai action
n8n drives agentic AI use cases including invoice processing and data enrichment for Marketing, HR, Finance, and Operations teams.
▾ source quote
“TMNZ runs an agnostic LLM stack with a number of foundational models including OpenAI, Azure, and Anthopic, atop a self-hosted infrastructure including PostgreSQL databases and vector stores. They now leverage n8n as part of this stack, for agentic AI use …”
3
MCP servers extend agent capabilities
Integration
MCP servers and interactive AI agents are called and combined with LLM models using n8n.
▾ source quote
“created a number of MCP servers and AI agents that are more interactive. Using n8n, these can be called and combined with LLM models”
4
Results delivered via Open WebUI
Output
Open WebUI serves as the user experience layer for AI agents and workflow results.
▾ source quote
“using Open WebUI as a user experience layer”
5
Results drive continuous improvement
Feedback loop
The team uses workflow results, including negative outcomes, to continuously iterate on workflows and improve resilience.
▾ source quote
“using the results to continuously iterate on workflows, to improve resilience and add new functionality as needed. Even negative results guide the team towards improved automations”
Reported outcome

n8n is now used by more than a third of TMNZ staff, running around 150,000 executions per month and saving 400 hours per month, with AI POCs saving up to an hour per execution; TMNZ is scaling toward one million executions per month.

Reported metrics
Monthly workflow executionsaround 150,000
Monthly hours saved400 hours
time saved per AI POC executionup to an hour
Staff adoptionmore than a third of TMNZ staff
Show all 5 reported metrics
monthly workflow executionsaround 150,000
monthly hours saved400 hours
time saved per AI POC executionup to an hour
staff adoptionmore than a third of TMNZ staff
target monthly executions (scaling goal)up to one million
Reported stack
n8nOpenAIAzurePostgreSQLvector storesOpen WebUI
◆ Does this fit your context?

Compare to your context

Tell us your scale, team, and constraints. We'll show what changes at your size, what fails at your scale, and whether this case is a fit, needs adaptation, or won't scale to you. Free demo, no signup.

Compare to your context →
~30 seconds · free
Source
https://n8n.io/case-studies/tmnz/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

n8n is now used by more than a third of TMNZ staff, running around 150,000 executions per month and saving 400 hours per month, with AI POCs saving up to an hour per execution; TMNZ is scaling toward one million execu…

What tools did this team use?

n8n, OpenAI, Azure, PostgreSQL, vector stores, Open WebUI.

What results were reported?

Monthly workflow executions: around 150,000; Monthly hours saved: 400 hours; time saved per AI POC execution: up to an hour; Staff adoption: more than a third of TMNZ staff (source-reported, not independently verified).

What failed first in this deployment?

Zapier and Make could only automate isolated workflows and were heavily limited in scope; MuleSoft and the Microsoft suite were either not self-hostable or cost-prohibitive for a 60-person Fintech.

How is this invoice processing AI workflow structured?

Staff submits automation request → LLM models execute agentic tasks → MCP servers extend agent capabilities → Results delivered via Open WebUI → Results drive continuous improvement.

Related invoice processing cases

WHAT TO DO WITH THIS

Now compare it to your context

This case is one data point. Whether its pattern fits you depends on your volumes, your stack, and your exception load — that comparison is the step no case study can do for you.

Browse the Invoice processing cases →