· Saul Mateos · AI-Native CFO · 5 min read
Automate FP&A Manager Tasks? Start With the Worst Process
I sent my best FP&A manager to deposit checks. A month later, 75% of the manual work was automated. A playbook to automate FP&A manager tasks.
I sent my best FP&A manager to deposit checks. Not as punishment. Not because we were short-staffed. If you want to automate FP&A manager tasks, or anyone’s, this is the playbook I’d steal.
Three months ago, I ran a Claude Code training session with my team. Out of everyone, two people fully embraced it. He was one of them. So when I needed someone to see a broken process up close and figure out how to fix it with AI, I knew exactly who to send.
Here’s the problem he walked into.
Our accounting team processes checks from a bank portal. Every day, someone logs in, downloads each check individually (sometimes hundreds in a day), opens it, matches it to a plaintiff in Salesforce, creates a cash receipt, posts it, and moves on to the next one. One. At. A. Time.
The workflow looks like it was designed in 2009. Because it was.
I’ve seen this pattern at every company I’ve worked at. The process is painful, everyone knows it’s painful, and nobody fixes it. Why? Because the people who understand the finance don’t know automation. And the people who know automation don’t understand the finance.
So I decided to close that gap with one person.
The assignment
I told him exactly what he was there to do. Sit with accounting. Learn every click. Document every step. Then figure out how to make it disappear.
He’s been using Claude Code for months. One of maybe three or four people at our company outside of Technology who actually adopted it beyond the basics. So he had the tools. What he didn’t have was the context.
That first couple of weeks, he did the work and got trained. Downloaded checks. Matched plaintiffs. Created cash receipts. Got to know the system the way you only can when you’re the one clicking through it at 8 AM on a Monday.
Then he started building.
What he actually built
He used Claude Cowork to create a skill that reads check images from our bank portal, extracts the key data, and matches it to the right opportunity in Salesforce. It cross-references the check amount against the payment amount on file, matches the plaintiff name, and if everything lines up, creates the cash receipt automatically. Right GL codes, right reference numbers, right accounts.
He even fed two weeks of his own call recordings into the tool as training context. Not the documentation (which barely existed). The actual conversations where people explained the process to him. That’s what gave the tool enough context to handle edge cases.
After one month, 75% of the manual work was automated.
Why this works and consultants don’t
I could have hired someone to map the process. I could have asked IT to build a tool. I’ve done both before. Here’s what happens:
The consultant maps the process in a nice diagram, delivers a PDF, and leaves. Nobody implements it because the diagram doesn’t capture the 40 exceptions that the accounting team handles from muscle memory.
The IT team builds a tool based on requirements they gathered in a meeting. It handles the happy path perfectly. It breaks on day one because nobody mentioned that sometimes the check amount doesn’t match and you have to look up a secondary field. By the way, it also takes forever. Good luck trying to get approval and prioritization in whatever process your Technology team has for bringing projects above the line.
The FP&A manager who sat in the chair for a month? He saw all of it. The exceptions, the workarounds, the “oh we just know to check this other screen” moments. And he built the automation around the real process, not the documented one.
What happens next
He built a prototype. It works. But it’s running on his machine in Cowork sessions, not in production. Let’s be honest. I’m not going to unleash my entire accounting team with a solution that was vibe-coded by one finance person. There are all sorts of potential issues with that.
So now I’m pulling in our technology team. Their job is to take what he built and turn it into something the accounting team can run themselves. A script, an app, whatever makes sense. Real security, real error handling, real handoff.
And I’m already thinking about where to send him next.
Most likely: our financial close. Same playbook. Same person. Sit in the process, learn every manual step, then automate it from the inside out.
The playbook to automate FP&A manager tasks (if you want to steal it)
- Find your person. Someone who understands finance AND is willing to learn AI tools. This is rare. Most finance people won’t touch it. Most tech people don’t understand the accounting.
- Send them into the worst process. Not the one that’s “kind of manual.” The one that makes people quit. That’s where the biggest ROI is hiding.
- Make the assignment clear. You’re not there to do the work forever. You’re there to learn it well enough to kill it.
- Give them time. A month is about right. Two weeks to learn, two weeks to build.
- Bring in engineering after, not before. Let the finance person define what needs to be built. Then let the tech team make it production-ready.
Most CFOs protect their best people from grunt work. I think that’s backwards. Your best people are the only ones who can see both the problem and the solution. Everyone else only sees one side.
Send them into the mess. That’s where the automation actually starts.