I’ve watched ops people spend hours copying numbers from one screen into another. Nobody signs up for that kind of work, and it never seems to go away.
So I stopped scrolling when I saw this post from a LinkedIn creator about how Deel runs its back office. The claim is bold: 100,000 back-office workflows automated every month, without touching a single API. The author explains how it happened. It’s a useful case study in cleaning up messy operations without waiting on engineers.
😩 The Problem: A Back Office Built on Chaos
Deel runs payroll and compliance for 1.5 million workers across 150+ countries. That sounds impressive until you picture what it takes behind the scenes.
According to the post’s author, the ops team was dealing with:
- Government portals that can’t connect to anything
- Bank logins that need a human every single time
- Legacy systems that were never built to talk to each other
For years, the team lived inside that mess. They screenshotted invoices. They logged into three separate systems just to process one payment. They copied numbers by hand.
The usual fix is to ask developers to build integrations. But you can’t build an API connection to a government portal that doesn’t offer one. And dev teams always have a backlog.
🛠️ The Solution: Akai Learns by Watching
Deel stopped waiting on developers and built a tool called Akai.
This is the part that got me. Akai watches someone do a task once, recording both their screen and their voice. From that single recording, it learns the entire workflow.
Picture how that plays out. An ops lead who knows the process walks through it one time, talking out loud as they click: “Open the bank portal, grab the invoice total, paste it into the payroll system.” Akai picks up the steps and the reasoning behind them.
There’s no API and no engineer writing custom code. The tool works on the screen, just like a person would.
📊 The Results Inside Deel
The creator shared some big numbers from inside Deel right now:
- 100,000+ workflows automated every month
- 91,000+ hours saved every single month
- Built and owned by ops leads, not engineers
- Gets smarter with every run, across any system
- Works on portals that were never built to connect to anything
That 91,000 hours figure stuck with me. It’s roughly the monthly workload of more than 500 full-time employees. And it’s happening every month, not once.
Deel is valued at 17.3 billion dollars. According to the post, Akai is how the company runs its entire back office now.
💡 Why This Approach Matters
Screen-based automation isn’t a brand-new idea. Older robotic process automation (RPA) tools also clicked through screens like a human. The catch was that they were fragile. Move one button, and the whole bot broke. Someone technical then had to go in and fix it.
What stands out here is the mix of three things:
- Learning from one recording instead of weeks of scripting
- Voice plus screen, so the tool gets the “why” and not just the clicks
- Improving with every run instead of breaking when something changes
The biggest shift is who owns the automation. The people who actually know the process build it themselves. That cuts out the slow back-and-forth between ops and engineering, where half the details get lost in translation.
✅ How to Apply This Thinking
You don’t have to be a company the size of Deel to borrow the idea. Here’s a simple way to start spotting automation opportunities in your own ops:
- List the repetitive tasks that make someone jump between two or more systems.
- Flag the ones where those systems have no API or integration.
- Find the person who does each task best, and have them record it once while explaining their steps out loud.
- Start with one high-volume workflow and track the hours saved before you scale up.
Even if you never automate a thing, step three is worth doing. A narrated recording is the best process doc your team will ever have.
I think this is a great reminder that the messiest workflows often aren’t a tech problem. They’re a “who gets to fix it” problem. Check out the full LinkedIn post for the complete story from the one who posted it.