Josephine Umana
Sales/Marketing Automation • AI Data Collection • n8n, Make
New to Contra
Josephine is ready for their next project!
An AI-powered Email Classification Workflow
0
7
A RAG AI Agent using Pinecone Vector
0
5
Built a full Gym subscription and renewal workflow in n8n
1
18
"The workflows you built on n8n 2 months ago just stopped working."
One of the reasons workflows error after weeks, months, and maybe a year is your data structure.
Testing and working with APIs and data have taught me the importance of testing workflows for future errors.
It's easier for someone to build some fancy workflow, maybe even take random templates they don't know how to build and pose as theirs, but hey, a real project oftentimes work differently from a demo.
I'm going to break this down in the best way possible, so grab a pen and a note or just save this post (Coach Jojo reporting for duty 🫡😉).
There are several reasons why your workflow might error, but let's talk about data as one of them:
Let's say you want to log a Google calendar event in Airtable and Airtable expects the date to come in an ISO format like this "yyyy-mm-dd", but it comes in like this instead "dd-mm-yyyy". An error.
You build an admirable AI voice agent that connects to your n8n workflow and everything seems fine. The agent is meant to send the number of minutes the call lasted, a number "24", but somehow it sends "24 minutes" even when it had sent "24" several times when you were testing. Workflow down because a string is sent instead of a number 👎
You connect your AI agent to an MCP server, send an example payload, and somehow, the agent sends back a different data type.
We could go on and on, but let's look at the remedy: Write down all the possible future errors before you build any workflow (this happen when you're planning the workflow) and then test your workflows to fail.
If you're expecting a number, what happens if a string is sent? Test your workflow with a string (text) and ensure it fails so you can fix it.
In Make, use the tools module (set variable or multiple variables) and write functions to structure the data. In n8n, use the code node to catch those data errors. Test, rewrite your code in the code node, test, again until everything works well. You can also use regular expressions to fix minor data errors on both platforms.
.
.
You can call me Jojo. I build AI-powered workflows and AI Agents for sales & marketing automation for agencies & companies using Agentic AI and low-code AI tools (n8n, Make, Vapi, Claude Code, Codex, and many others).
Beyond building workflows, I handle the technical side of things; I read and interpret API documentation, set up webhooks, test APIs and payloads inside Postman, and structure conversational design & system prompts so your AI agents behave exactly as expected.
I’m also open to AI data collection and data annotation roles. You can send a DM if you need any of my services.
Just joined here, but I also share content around AI, tech, Automation, and AI/tech careers. You’re welcome to follow 😊
1
34