The most common reason AI automation projects stall isn’t technical. It’s the planning phase.
Someone decides they want to automate their workflow. They start designing the full system — AI that handles emails, schedules meetings, updates the CRM, preps briefings, drafts proposals, follows up with leads. All of it, connected, running automatically.
They spend weeks drawing it out. Then either never build it, or build something so complex that it breaks on day three.
I’ve watched this happen enough times that I’ve started giving everyone the same first piece of advice: start with a voice note.
What “start with a voice note” actually means
After every meeting, drop a two-minute voice note. What was discussed. What you committed to. Anything important about the person that you want to remember.
That voice note gets automatically transcribed and parsed, and the relevant details get pushed to update the CRM record for that contact.
On the surface, this sounds simple to the point of being obvious. But the impact of this one workflow is larger than it looks.
Most CRM data is stale, incomplete, or wrong. Not because people don’t care about accuracy — they do. It’s because nobody wants to sit down after a call and type notes. It’s tedious, it breaks the flow of the day, and it always gets deprioritized.
Voice is different. Two minutes while you’re still in the mental context of the meeting. The AI handles the transcription and the data entry. The result is accurate CRM records — maybe for the first time.
Accurate records mean better handoffs between teammates. Fewer things falling through the cracks. A real history of every relationship in one place.
That’s already a meaningful win.
How one workflow becomes two
Once the data is accurate and you trust it, something interesting happens: the next step practically builds itself.
The day before any meeting, the system can pull notes from your last interaction and send you a briefing. You walk into every call knowing exactly where you left off — what was discussed, what you promised, what the person is working on.
That’s not a separate second workflow. It’s the same workflow extended. The voice note created the data; the briefing is just reading it back to you at the right time.
This is the pattern I see with every good automation system: what starts as one simple step reveals the next step. The system designs itself once you’re in motion.
What this looks like in practice
I have an agent called Teddy that manages my CRM backlog overnight. It enriches contact information, drafts follow-up emails, updates records based on recent activity. By morning, it’s completed a batch of work that would have taken an hour of manual data entry.
My job is quality control. I review what Teddy handled, approve or adjust anything that needs a human touch, and move on.
That shift — from data entry to quality control — only became possible because I started with a simple foundation. Accurate data. Consistent logging. One trigger that ran reliably.
I’ve also built a meeting follow-up workflow where if I mention during a call that I’ll send something, the agent auto-drafts the email. Within one minute of hanging up, there’s a completed draft in my inbox. No to-do item added. No forgetting. Just a draft ready to review.
The follow-up workflow depends on the transcription. The transcription depends on the voice note habit. Simple foundation, then layers.
The start-small-iterate approach
There’s a principle I use in AI consulting called start-small-iterate. It sounds self-evident, but most people don’t apply it. They want to see the full architecture before committing to anything.
The problem is that the full architecture isn’t visible until you’ve started. You learn what the next step should be by running the first step and watching what happens.
One voice note workflow will tell you more about your actual workflow gaps than a week of planning sessions.
How to find your voice note equivalent
A voice note after meetings isn’t the only entry point. It’s just one example of the pattern.
The pattern is: find the manual, repetitive task that happens consistently but never gets done well. Something that should happen after every meeting, or every sales call, or every customer interaction — but doesn’t, because typing it out is too slow.
That’s your starting point.
Other examples that follow the same logic:
- After a sales call, the agent reads the transcript and populates the proposal template. Proposal goes out in minutes instead of sitting on a to-do list.
- After a client onboarding call, the agent triggers account setup, sends the welcome sequence, and logs the new client in the system.
- After any meeting where a commitment was made, the agent drafts the follow-up email so it’s ready to send immediately.
All of these start from the same place: one trigger, one action, run consistently.
A simple way to start this week
Pick the single most annoying manual thing that happens after meetings in your workflow right now. Not the most ambitious thing — the most annoying thing.
Build one agent around that. Let it run for two weeks.
Then ask yourself: what data does this workflow create, and what’s the natural next use of that data?
That’s your second workflow.
You don’t need to plan the whole system in advance. You just need to start somewhere specific and let the system design itself from there.
Recommended for you
Want the full system? 25X is the flagship productivity system we teach.
Explore 25X →