Try OnsaTL;DR: Build an automated meeting intelligence pipeline analytics with n8n that captures, transcribes, and analyzes sales calls in real time. Extract action items, deal risks, and coaching insights without manual note-taking — then route them to your CRM automatically.
I’ll be honest: I’m a bit of a disaster when it comes to manual admin. If a task requires me to spend 20 minutes “stalking” a prospect on LinkedIn or manually typing notes into a CRM, there is a 100% chance I will find a way to procrastinate. In this post, I’m going to show you how I reclaimed my sanity by building a meeting intelligence engine that does the heavy lifting for me.
We’ve all been there. You have a back-to-back schedule, and you’re jumping into a Discovery Call with a “VP of Something” at a “Company You’ve Barely Heard Of.” You have two choices: go in blind and look like an amateur, or spend your lunch break frantically Googling their recent Series B funding.
Most sales reps spend about 20% of their week just preparing for meetings or cleaning up the mess afterward in HubSpot or Salesforce. That’s a day a week lost to the “Admin Tax.” I decided to stop paying it. By using N8N (an open-source automation tool) and a few AI legos, you can build a system that delivers a “cheat sheet” to your inbox before the call and logs everything after it.
Here is exactly how I built my workflow. It’s not as intimidating as it looks, and it saves me about five hours of grunt work every week.
Most people try to do this with just a basic Google search, but the results are usually noisy. I like Firecrawl because it actually “crawls” the prospect’s site and recent news, pulling out the context that actually matters for a sales call—like a recent product launch or a shift in their market positioning.
When I jump on a call now, I’m not asking, “So, what does your company do?” I’m saying, “I saw you recently expanded into the EMEA market; how is that affecting your sales ops?” That’s the difference between a vendor and a partner.
Building this in N8N is a fantastic Saturday project if you’re a tinkerer like me. It gives you total control over the logic. But I also realize that most sales leaders don’t have the time to debug a JSON error at 10:00 PM on a Tuesday.
That’s exactly why we’re building Onsa.ai. We took the core philosophy of this workflow—eliminating manual research and CRM friction—and turned it into a seamless, “it just works” experience for sales teams who want the power of AI without the setup headache.
P.S. If you do build this yourself, make sure your OpenAI prompt specifically tells the AI to “keep it brief.” There’s nothing worse than an automated memo that’s longer than the actual meeting.
If you are using Google Gemini instead of OpenAI for transcription in your n8n workflow, you may run into a specific error that stops the pipeline.
The most common issue: your n8n Gemini node throws a "transcribe a recording" error or a "resource not found" error when processing audio/video files (MP4, MP3, WAV).
Root causes and fixes:
• File format mismatch: Gemini expects specific MIME types. If you are passing an MP4 file, make sure the MIME type is set to video/mp4, not audio/mp4. For audio-only files, use audio/mp3 or audio/wav.
• File size limits: Gemini has a ~20MB inline limit for media files. For larger recordings, upload the file to Google Cloud Storage first and pass the GCS URI instead of the raw binary. Use the Google Cloud Storage node in n8n to upload, then reference the gs://bucket/file URI in your Gemini prompt.
• Resource not found: This usually means the file URI or reference is invalid. If you are chaining nodes (e.g., HTTP Request → Gemini), verify the binary data is being passed correctly between nodes. Add a Set node between them to explicitly map the binary field name.
• API version mismatch: Make sure your n8n Google Gemini node is using the latest API version. Older n8n versions may reference deprecated Gemini endpoints. Update n8n to the latest version (npm update n8n -g) and reconnect your Google credentials.
• Credentials scope: Your Google service account needs the Generative Language API enabled. Go to the Google Cloud Console → APIs & Services → Enable APIs → search for "Generative Language API" and enable it.
If the Gemini transcription issues persist, consider using OpenAI Whisper for the transcription step and Gemini for the analysis/summarization. This split approach is more reliable: Whisper handles audio-to-text (it accepts MP3, MP4, WAV, M4A up to 25MB), and Gemini handles the meeting intelligence extraction from the transcript text. In n8n, this means: Recording file → OpenAI Whisper node (transcribe) → Gemini node (analyze and extract action items).
A: Not necessarily, but you do need to understand “if-this-then-that” logic. It’s a low-code tool, meaning you’ll be dragging and dropping blocks, but you might need to copy-paste a few snippets of JSON here and there.
A: N8N can be self-hosted for free or used via their cloud for a small fee. Your main costs will be OpenAI API credits and Firecrawl, which are usually just a few cents per meeting. It’s significantly cheaper than hiring a virtual assistant.
A: Absolutely. N8N has a native Microsoft Outlook integration that works almost exactly like the Google Calendar trigger.