· Saul Mateos · Claude and AI Tools · 5 min read
How to Use AI in Finance Team Work: An Honest Scorecard
How to use AI in finance team work, after a year of doing it: six areas I tested, what stuck, what I quietly stopped doing, and where to start.
89% of finance executives plan to upskill their teams for AI. That’s from EY’s latest survey. The harder question is how to use AI in finance team work once the plan is made.
I’ve been doing it for a year. Not planning. Doing. And the gap between “plan to upskill” and “actually changed how we work” is enormous.
Here’s an honest scorecard. Six areas I tested, what actually stuck, and what I quietly stopped doing.
01 Parallel Working
The single biggest unlock was parallel working. I keep three Claude Code windows side by side across departments: finance reporting, marketing site, internal tools. One finishes a task, I review it. While I’m reviewing, the other two are building. When I send feedback on the first, I rotate to the third. Nobody waits. Nothing sits idle. A complex build takes 5-10 minutes. Single-threading that same work means 2-3 hours a day watching loading spinners. The bottleneck was never the AI. It was me sitting there like a project manager with one direct report.
Fair warning: context switching has a cost. You need enough familiarity with each project to review output quickly, or you’ll introduce errors trying to move too fast. Start with two sessions, not eight. And keep a log of what each window is doing, because when you come back to a finished task, you need to remember what you asked for.
02 Financial Modeling
Financial modeling landed in a good spot. Claude Code for driver-based models, then Claude within Excel (the plugin) for scaffolding, then manual refinement for the nuanced parts. AI gets you to 80% fast. The last 20% is where your judgment matters. One gotcha: version control. New Claude-generated versions will override manual changes you made outside the agent. You have to explicitly tell it to preserve your edits.
03 Reporting & Dashboards
Reporting and dashboards is where AI shines most. I still don’t understand why people are using Power BI or Tableau when you can build flexible dashboards with local data and see the analysis in real time. Claude Code makes this accessible even if you’re not a developer. I’ve built dashboards in a few minutes that replaced manual processes taking someone half a day every week. Two things I learned the hard way, though. First, your data and calculations need to live in a database (backend), not in the web page itself (frontend). Claude usually handles this correctly, but you need to understand the concept or you’ll end up with something fragile. Second, security. When you’re building dashboards with real financial data, add security layers. Don’t skip this.
04 Writing & Communication
Writing is a different problem. I built a humanizer skill for Claude that catches AI writing patterns and fixes them automatically. But the bigger issue is AI slop. It’s everywhere now. LinkedIn, Reddit, work emails. You can spot it instantly. And when I see a careless AI-written email, I don’t just notice the bad writing. I wonder if the person also let AI handle the actual analysis behind it. Careless writing signals careless thinking. If you’re using AI to write, you need a system that makes the output sound like you actually wrote it.
05 Operations & Adoption
Then there’s adoption. Using AI yourself is one thing. Putting it at scale is something else entirely. I trained 10 people, accepted that 3 actually adopted. Those 3 now produce 3x output. The math works even at 30% adoption. You don’t need everyone on board. You need your best people multiplied. But the tools aren’t there yet for company-wide deployment. Not everyone should be learning Claude Code. I showed the whole team ChatGPT. Gave them prompts. Did a live demo. Nobody used it the next week. What actually worked was finding one person’s specific pain point, building an automation for that one thing, and watching them become an evangelist. Start with one person’s real problem, not a company-wide initiative.
06 Shadow IT & Data Security
Nobody talks about the shadow IT part. AI adoption means people will use tools you didn’t approve. They’ll paste financial data into random AI platforms because it’s faster than waiting for your approved workflow. Finance data in unvetted AI tools is a real risk. Your technology team needs to oversee this. Data management policies aren’t optional.
How to Use AI in Finance Team Work: The Pattern Behind Every Win
None of these wins came from buying a tool or attending a webinar. Every single one followed the same path: I did the task manually first, documented the repeatable steps, and then automated the boring part.
I should be honest about how I got here. I’ve spent countless hours on this. I started with Make.com automations that constantly broke and drove me crazy. ChatGPT lied to me nonstop and wasted more time than it saved. Then I found Claude. Then Claude Artifacts. I switched to Lovable.dev for building things until I watched a YouTuber explain that Claude Code is way cheaper. Haven’t looked back since.
The honest summary is simple. AI saves real time on specific, repeatable tasks. It’s terrible at anything requiring judgment, relationships, or context your team carries in their heads.
Stop collecting AI tool lists. Start with one workflow that’s eating your team’s time and automate that. One. Not five. Not a roadmap. One workflow, this week.